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         return handler ? xattr_prefix(handler) : NULL;
 548 }
 549
 550 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 551                                  const char *name,
 552                                  int name_len)
 553 {
 554         /* Get hash value of uuid from super block */
 555         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 556         int i;
 557
 558         /* hash extended attribute name */
 559         for (i = 0; i < name_len; i++) {
 560                 hash = (hash << OCFS2_HASH_SHIFT) ^
 561                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 562                        *name++;
 563         }
 564
 565         return hash;
 566 }
 567
 568 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 569 {
 570         return namevalue_size(name_len, value_len) +
 571                 sizeof(struct ocfs2_xattr_entry);
 572 }
 573
 574 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 575 {
 576         return namevalue_size_xi(xi) +
 577                 sizeof(struct ocfs2_xattr_entry);
 578 }
 579
 580 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 581 {
 582         return namevalue_size_xe(xe) +
 583                 sizeof(struct ocfs2_xattr_entry);
 584 }
 585
 586 int ocfs2_calc_security_init(struct inode *dir,
 587                              struct ocfs2_security_xattr_info *si,
 588                              int *want_clusters,
 589                              int *xattr_credits,
 590                              struct ocfs2_alloc_context **xattr_ac)
 591 {
 592         int ret = 0;
 593         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 594         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 595                                                  si->value_len);
 596
 597         /*
 598          * The max space of security xattr taken inline is
 599          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 600          * So reserve one metadata block for it is ok.
 601          */
 602         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 603             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 604                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 605                 if (ret) {
 606                         mlog_errno(ret);
 607                         return ret;
 608                 }
 609                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 610         }
 611
 612         /* reserve clusters for xattr value which will be set in B tree*/
 613         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 614                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 615                                                             si->value_len);
 616
 617                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 618                                                            new_clusters);
 619                 *want_clusters += new_clusters;
 620         }
 621         return ret;
 622 }
 623
 624 int ocfs2_calc_xattr_init(struct inode *dir,
 625                           struct buffer_head *dir_bh,
 626                           umode_t mode,
 627                           struct ocfs2_security_xattr_info *si,
 628                           int *want_clusters,
 629                           int *xattr_credits,
 630                           int *want_meta)
 631 {
 632         int ret = 0;
 633         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 634         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 635
 636         if (si->enable)
 637                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 638                                                      si->value_len);
 639
 640         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 641                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 642                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 643                                         "", NULL, 0);
 644                 if (acl_len > 0) {
 645                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 646                         if (S_ISDIR(mode))
 647                                 a_size <<= 1;
 648                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 649                         mlog_errno(ret);
 650                         return ret;
 651                 }
 652         }
 653
 654         if (!(s_size + a_size))
 655                 return ret;
 656
 657         /*
 658          * The max space of security xattr taken inline is
 659          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 660          * The max space of acl xattr taken inline is
 661          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 662          * when blocksize = 512, may reserve one more cluser for
 663          * xattr bucket, otherwise reserve one metadata block
 664          * for them is ok.
 665          * If this is a new directory with inline data,
 666          * we choose to reserve the entire inline area for
 667          * directory contents and force an external xattr block.
 668          */
 669         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 670             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 671             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 672                 *want_meta = *want_meta + 1;
 673                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 674         }
 675
 676         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 677             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 678                 *want_clusters += 1;
 679                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 680         }
 681
 682         /*
 683          * reserve credits and clusters for xattrs which has large value
 684          * and have to be set outside
 685          */
 686         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 687                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 688                                                         si->value_len);
 689                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 690                                                            new_clusters);
 691                 *want_clusters += new_clusters;
 692         }
 693         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 694             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 695                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 696                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 697                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 698                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 699                                                            new_clusters);
 700                 *want_clusters += new_clusters;
 701         }
 702
 703         return ret;
 704 }
 705
 706 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 707                                          u32 clusters_to_add,
 708                                          struct ocfs2_xattr_value_buf *vb,
 709                                          struct ocfs2_xattr_set_ctxt *ctxt)
 710 {
 711         int status = 0, credits;
 712         handle_t *handle = ctxt->handle;
 713         enum ocfs2_alloc_restarted why;
 714         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 715         struct ocfs2_extent_tree et;
 716
 717         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 718
 719         while (clusters_to_add) {
 720                 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 721
 722                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 723                                        OCFS2_JOURNAL_ACCESS_WRITE);
 724                 if (status < 0) {
 725                         mlog_errno(status);
 726                         break;
 727                 }
 728
 729                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 730                 status = ocfs2_add_clusters_in_btree(handle,
 731                                                      &et,
 732                                                      &logical_start,
 733                                                      clusters_to_add,
 734                                                      0,
 735                                                      ctxt->data_ac,
 736                                                      ctxt->meta_ac,
 737                                                      &why);
 738                 if ((status < 0) && (status != -EAGAIN)) {
 739                         if (status != -ENOSPC)
 740                                 mlog_errno(status);
 741                         break;
 742                 }
 743
 744                 ocfs2_journal_dirty(handle, vb->vb_bh);
 745
 746                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 747                                          prev_clusters;
 748
 749                 if (why != RESTART_NONE && clusters_to_add) {
 750                         /*
 751                          * We can only fail in case the alloc file doesn't give
 752                          * up enough clusters.
 753                          */
 754                         BUG_ON(why == RESTART_META);
 755
 756                         credits = ocfs2_calc_extend_credits(inode->i_sb,
 757                                                             &vb->vb_xv->xr_list);
 758                         status = ocfs2_extend_trans(handle, credits);
 759                         if (status < 0) {
 760                                 status = -ENOMEM;
 761                                 mlog_errno(status);
 762                                 break;
 763                         }
 764                 }
 765         }
 766
 767         return status;
 768 }
 769
 770 static int __ocfs2_remove_xattr_range(struct inode *inode,
 771                                       struct ocfs2_xattr_value_buf *vb,
 772                                       u32 cpos, u32 phys_cpos, u32 len,
 773                                       unsigned int ext_flags,
 774                                       struct ocfs2_xattr_set_ctxt *ctxt)
 775 {
 776         int ret;
 777         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 778         handle_t *handle = ctxt->handle;
 779         struct ocfs2_extent_tree et;
 780
 781         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 782
 783         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 784                             OCFS2_JOURNAL_ACCESS_WRITE);
 785         if (ret) {
 786                 mlog_errno(ret);
 787                 goto out;
 788         }
 789
 790         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 791                                   &ctxt->dealloc);
 792         if (ret) {
 793                 mlog_errno(ret);
 794                 goto out;
 795         }
 796
 797         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 798         ocfs2_journal_dirty(handle, vb->vb_bh);
 799
 800         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 801                 ret = ocfs2_decrease_refcount(inode, handle,
 802                                         ocfs2_blocks_to_clusters(inode->i_sb,
 803                                                                  phys_blkno),
 804                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 805         else
 806                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 807                                                   phys_blkno, len);
 808         if (ret)
 809                 mlog_errno(ret);
 810
 811 out:
 812         return ret;
 813 }
 814
 815 static int ocfs2_xattr_shrink_size(struct inode *inode,
 816                                    u32 old_clusters,
 817                                    u32 new_clusters,
 818                                    struct ocfs2_xattr_value_buf *vb,
 819                                    struct ocfs2_xattr_set_ctxt *ctxt)
 820 {
 821         int ret = 0;
 822         unsigned int ext_flags;
 823         u32 trunc_len, cpos, phys_cpos, alloc_size;
 824         u64 block;
 825
 826         if (old_clusters <= new_clusters)
 827                 return 0;
 828
 829         cpos = new_clusters;
 830         trunc_len = old_clusters - new_clusters;
 831         while (trunc_len) {
 832                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 833                                                &alloc_size,
 834                                                &vb->vb_xv->xr_list, &ext_flags);
 835                 if (ret) {
 836                         mlog_errno(ret);
 837                         goto out;
 838                 }
 839
 840                 if (alloc_size > trunc_len)
 841                         alloc_size = trunc_len;
 842
 843                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 844                                                  phys_cpos, alloc_size,
 845                                                  ext_flags, ctxt);
 846                 if (ret) {
 847                         mlog_errno(ret);
 848                         goto out;
 849                 }
 850
 851                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 852                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 853                                                        block, alloc_size);
 854                 cpos += alloc_size;
 855                 trunc_len -= alloc_size;
 856         }
 857
 858 out:
 859         return ret;
 860 }
 861
 862 static int ocfs2_xattr_value_truncate(struct inode *inode,
 863                                       struct ocfs2_xattr_value_buf *vb,
 864                                       int len,
 865                                       struct ocfs2_xattr_set_ctxt *ctxt)
 866 {
 867         int ret;
 868         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 869         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 870
 871         if (new_clusters == old_clusters)
 872                 return 0;
 873
 874         if (new_clusters > old_clusters)
 875                 ret = ocfs2_xattr_extend_allocation(inode,
 876                                                     new_clusters - old_clusters,
 877                                                     vb, ctxt);
 878         else
 879                 ret = ocfs2_xattr_shrink_size(inode,
 880                                               old_clusters, new_clusters,
 881                                               vb, ctxt);
 882
 883         return ret;
 884 }
 885
 886 static int ocfs2_xattr_list_entry(struct super_block *sb,
 887                                   char *buffer, size_t size,
 888                                   size_t *result, int type,
 889                                   const char *name, int name_len)
 890 {
 891         char *p = buffer + *result;
 892         const char *prefix;
 893         int prefix_len;
 894         int total_len;
 895
 896         switch(type) {
 897         case OCFS2_XATTR_INDEX_USER:
 898                 if (OCFS2_SB(sb)->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
 899                         return 0;
 900                 break;
 901
 902         case OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS:
 903         case OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT:
 904                 if (!(sb->s_flags & MS_POSIXACL))
 905                         return 0;
 906                 break;
 907
 908         case OCFS2_XATTR_INDEX_TRUSTED:
 909                 if (!capable(CAP_SYS_ADMIN))
 910                         return 0;
 911                 break;
 912         }
 913
 914         prefix = ocfs2_xattr_prefix(type);
 915         if (!prefix)
 916                 return 0;
 917         prefix_len = strlen(prefix);
 918         total_len = prefix_len + name_len + 1;
 919         *result += total_len;
 920
 921         /* we are just looking for how big our buffer needs to be */
 922         if (!size)
 923                 return 0;
 924
 925         if (*result > size)
 926                 return -ERANGE;
 927
 928         memcpy(p, prefix, prefix_len);
 929         memcpy(p + prefix_len, name, name_len);
 930         p[prefix_len + name_len] = '\0';
 931
 932         return 0;
 933 }
 934
 935 static int ocfs2_xattr_list_entries(struct inode *inode,
 936                                     struct ocfs2_xattr_header *header,
 937                                     char *buffer, size_t buffer_size)
 938 {
 939         size_t result = 0;
 940         int i, type, ret;
 941         const char *name;
 942
 943         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 944                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 945                 type = ocfs2_xattr_get_type(entry);
 946                 name = (const char *)header +
 947                         le16_to_cpu(entry->xe_name_offset);
 948
 949                 ret = ocfs2_xattr_list_entry(inode->i_sb,
 950                                              buffer, buffer_size,
 951                                              &result, type, name,
 952                                              entry->xe_name_len);
 953                 if (ret)
 954                         return ret;
 955         }
 956
 957         return result;
 958 }
 959
 960 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 961                                          struct ocfs2_dinode *di)
 962 {
 963         struct ocfs2_xattr_header *xh;
 964         int i;
 965
 966         xh = (struct ocfs2_xattr_header *)
 967                  ((void *)di + inode->i_sb->s_blocksize -
 968                  le16_to_cpu(di->i_xattr_inline_size));
 969
 970         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 971                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 972                         return 1;
 973
 974         return 0;
 975 }
 976
 977 static int ocfs2_xattr_ibody_list(struct inode *inode,
 978                                   struct ocfs2_dinode *di,
 979                                   char *buffer,
 980                                   size_t buffer_size)
 981 {
 982         struct ocfs2_xattr_header *header = NULL;
 983         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 984         int ret = 0;
 985
 986         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 987                 return ret;
 988
 989         header = (struct ocfs2_xattr_header *)
 990                  ((void *)di + inode->i_sb->s_blocksize -
 991                  le16_to_cpu(di->i_xattr_inline_size));
 992
 993         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 994
 995         return ret;
 996 }
 997
 998 static int ocfs2_xattr_block_list(struct inode *inode,
 999                                   struct ocfs2_dinode *di,
1000                                   char *buffer,
1001                                   size_t buffer_size)
1002 {
1003         struct buffer_head *blk_bh = NULL;
1004         struct ocfs2_xattr_block *xb;
1005         int ret = 0;
1006
1007         if (!di->i_xattr_loc)
1008                 return ret;
1009
1010         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
1011                                      &blk_bh);
1012         if (ret < 0) {
1013                 mlog_errno(ret);
1014                 return ret;
1015         }
1016
1017         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1018         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1019                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1020                 ret = ocfs2_xattr_list_entries(inode, header,
1021                                                buffer, buffer_size);
1022         } else
1023                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1024                                                    buffer, buffer_size);
1025
1026         brelse(blk_bh);
1027
1028         return ret;
1029 }
1030
1031 ssize_t ocfs2_listxattr(struct dentry *dentry,
1032                         char *buffer,
1033                         size_t size)
1034 {
1035         int ret = 0, i_ret = 0, b_ret = 0;
1036         struct buffer_head *di_bh = NULL;
1037         struct ocfs2_dinode *di = NULL;
1038         struct ocfs2_inode_info *oi = OCFS2_I(d_inode(dentry));
1039
1040         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1041                 return -EOPNOTSUPP;
1042
1043         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1044                 return ret;
1045
1046         ret = ocfs2_inode_lock(d_inode(dentry), &di_bh, 0);
1047         if (ret < 0) {
1048                 mlog_errno(ret);
1049                 return ret;
1050         }
1051
1052         di = (struct ocfs2_dinode *)di_bh->b_data;
1053
1054         down_read(&oi->ip_xattr_sem);
1055         i_ret = ocfs2_xattr_ibody_list(d_inode(dentry), di, buffer, size);
1056         if (i_ret < 0)
1057                 b_ret = 0;
1058         else {
1059                 if (buffer) {
1060                         buffer += i_ret;
1061                         size -= i_ret;
1062                 }
1063                 b_ret = ocfs2_xattr_block_list(d_inode(dentry), di,
1064                                                buffer, size);
1065                 if (b_ret < 0)
1066                         i_ret = 0;
1067         }
1068         up_read(&oi->ip_xattr_sem);
1069         ocfs2_inode_unlock(d_inode(dentry), 0);
1070
1071         brelse(di_bh);
1072
1073         return i_ret + b_ret;
1074 }
1075
1076 static int ocfs2_xattr_find_entry(int name_index,
1077                                   const char *name,
1078                                   struct ocfs2_xattr_search *xs)
1079 {
1080         struct ocfs2_xattr_entry *entry;
1081         size_t name_len;
1082         int i, cmp = 1;
1083
1084         if (name == NULL)
1085                 return -EINVAL;
1086
1087         name_len = strlen(name);
1088         entry = xs->here;
1089         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1090                 cmp = name_index - ocfs2_xattr_get_type(entry);
1091                 if (!cmp)
1092                         cmp = name_len - entry->xe_name_len;
1093                 if (!cmp)
1094                         cmp = memcmp(name, (xs->base +
1095                                      le16_to_cpu(entry->xe_name_offset)),
1096                                      name_len);
1097                 if (cmp == 0)
1098                         break;
1099                 entry += 1;
1100         }
1101         xs->here = entry;
1102
1103         return cmp ? -ENODATA : 0;
1104 }
1105
1106 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1107                                          struct ocfs2_xattr_value_root *xv,
1108                                          void *buffer,
1109                                          size_t len)
1110 {
1111         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1112         u64 blkno;
1113         int i, ret = 0;
1114         size_t cplen, blocksize;
1115         struct buffer_head *bh = NULL;
1116         struct ocfs2_extent_list *el;
1117
1118         el = &xv->xr_list;
1119         clusters = le32_to_cpu(xv->xr_clusters);
1120         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1121         blocksize = inode->i_sb->s_blocksize;
1122
1123         cpos = 0;
1124         while (cpos < clusters) {
1125                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1126                                                &num_clusters, el, NULL);
1127                 if (ret) {
1128                         mlog_errno(ret);
1129                         goto out;
1130                 }
1131
1132                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1133                 /* Copy ocfs2_xattr_value */
1134                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1135                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1136                                                &bh, NULL);
1137                         if (ret) {
1138                                 mlog_errno(ret);
1139                                 goto out;
1140                         }
1141
1142                         cplen = len >= blocksize ? blocksize : len;
1143                         memcpy(buffer, bh->b_data, cplen);
1144                         len -= cplen;
1145                         buffer += cplen;
1146
1147                         brelse(bh);
1148                         bh = NULL;
1149                         if (len == 0)
1150                                 break;
1151                 }
1152                 cpos += num_clusters;
1153         }
1154 out:
1155         return ret;
1156 }
1157
1158 static int ocfs2_xattr_ibody_get(struct inode *inode,
1159                                  int name_index,
1160                                  const char *name,
1161                                  void *buffer,
1162                                  size_t buffer_size,
1163                                  struct ocfs2_xattr_search *xs)
1164 {
1165         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1166         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1167         struct ocfs2_xattr_value_root *xv;
1168         size_t size;
1169         int ret = 0;
1170
1171         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1172                 return -ENODATA;
1173
1174         xs->end = (void *)di + inode->i_sb->s_blocksize;
1175         xs->header = (struct ocfs2_xattr_header *)
1176                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1177         xs->base = (void *)xs->header;
1178         xs->here = xs->header->xh_entries;
1179
1180         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1181         if (ret)
1182                 return ret;
1183         size = le64_to_cpu(xs->here->xe_value_size);
1184         if (buffer) {
1185                 if (size > buffer_size)
1186                         return -ERANGE;
1187                 if (ocfs2_xattr_is_local(xs->here)) {
1188                         memcpy(buffer, (void *)xs->base +
1189                                le16_to_cpu(xs->here->xe_name_offset) +
1190                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1191                 } else {
1192                         xv = (struct ocfs2_xattr_value_root *)
1193                                 (xs->base + le16_to_cpu(
1194                                  xs->here->xe_name_offset) +
1195                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1196                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1197                                                             buffer, size);
1198                         if (ret < 0) {
1199                                 mlog_errno(ret);
1200                                 return ret;
1201                         }
1202                 }
1203         }
1204
1205         return size;
1206 }
1207
1208 static int ocfs2_xattr_block_get(struct inode *inode,
1209                                  int name_index,
1210                                  const char *name,
1211                                  void *buffer,
1212                                  size_t buffer_size,
1213                                  struct ocfs2_xattr_search *xs)
1214 {
1215         struct ocfs2_xattr_block *xb;
1216         struct ocfs2_xattr_value_root *xv;
1217         size_t size;
1218         int ret = -ENODATA, name_offset, name_len, i;
1219         int uninitialized_var(block_off);
1220
1221         xs->bucket = ocfs2_xattr_bucket_new(inode);
1222         if (!xs->bucket) {
1223                 ret = -ENOMEM;
1224                 mlog_errno(ret);
1225                 goto cleanup;
1226         }
1227
1228         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1229         if (ret) {
1230                 mlog_errno(ret);
1231                 goto cleanup;
1232         }
1233
1234         if (xs->not_found) {
1235                 ret = -ENODATA;
1236                 goto cleanup;
1237         }
1238
1239         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1240         size = le64_to_cpu(xs->here->xe_value_size);
1241         if (buffer) {
1242                 ret = -ERANGE;
1243                 if (size > buffer_size)
1244                         goto cleanup;
1245
1246                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1247                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1248                 i = xs->here - xs->header->xh_entries;
1249
1250                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1251                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1252                                                                 bucket_xh(xs->bucket),
1253                                                                 i,
1254                                                                 &block_off,
1255                                                                 &name_offset);
1256                         if (ret) {
1257                                 mlog_errno(ret);
1258                                 goto cleanup;
1259                         }
1260                         xs->base = bucket_block(xs->bucket, block_off);
1261                 }
1262                 if (ocfs2_xattr_is_local(xs->here)) {
1263                         memcpy(buffer, (void *)xs->base +
1264                                name_offset + name_len, size);
1265                 } else {
1266                         xv = (struct ocfs2_xattr_value_root *)
1267                                 (xs->base + name_offset + name_len);
1268                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1269                                                             buffer, size);
1270                         if (ret < 0) {
1271                                 mlog_errno(ret);
1272                                 goto cleanup;
1273                         }
1274                 }
1275         }
1276         ret = size;
1277 cleanup:
1278         ocfs2_xattr_bucket_free(xs->bucket);
1279
1280         brelse(xs->xattr_bh);
1281         xs->xattr_bh = NULL;
1282         return ret;
1283 }
1284
1285 int ocfs2_xattr_get_nolock(struct inode *inode,
1286                            struct buffer_head *di_bh,
1287                            int name_index,
1288                            const char *name,
1289                            void *buffer,
1290                            size_t buffer_size)
1291 {
1292         int ret;
1293         struct ocfs2_dinode *di = NULL;
1294         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1295         struct ocfs2_xattr_search xis = {
1296                 .not_found = -ENODATA,
1297         };
1298         struct ocfs2_xattr_search xbs = {
1299                 .not_found = -ENODATA,
1300         };
1301
1302         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1303                 return -EOPNOTSUPP;
1304
1305         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1306                 return -ENODATA;
1307
1308         xis.inode_bh = xbs.inode_bh = di_bh;
1309         di = (struct ocfs2_dinode *)di_bh->b_data;
1310
1311         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1312                                     buffer_size, &xis);
1313         if (ret == -ENODATA && di->i_xattr_loc)
1314                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1315                                             buffer_size, &xbs);
1316
1317         return ret;
1318 }
1319
1320 /* ocfs2_xattr_get()
1321  *
1322  * Copy an extended attribute into the buffer provided.
1323  * Buffer is NULL to compute the size of buffer required.
1324  */
1325 static int ocfs2_xattr_get(struct inode *inode,
1326                            int name_index,
1327                            const char *name,
1328                            void *buffer,
1329                            size_t buffer_size)
1330 {
1331         int ret, had_lock;
1332         struct buffer_head *di_bh = NULL;
1333         struct ocfs2_lock_holder oh;
1334
1335         had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
1336         if (had_lock < 0) {
1337                 mlog_errno(had_lock);
1338                 return had_lock;
1339         }
1340         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1341         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1342                                      name, buffer, buffer_size);
1343         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1344
1345         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1346
1347         brelse(di_bh);
1348
1349         return ret;
1350 }
1351
1352 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1353                                            handle_t *handle,
1354                                            struct ocfs2_xattr_value_buf *vb,
1355                                            const void *value,
1356                                            int value_len)
1357 {
1358         int ret = 0, i, cp_len;
1359         u16 blocksize = inode->i_sb->s_blocksize;
1360         u32 p_cluster, num_clusters;
1361         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1362         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1363         u64 blkno;
1364         struct buffer_head *bh = NULL;
1365         unsigned int ext_flags;
1366         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1367
1368         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1369
1370         while (cpos < clusters) {
1371                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1372                                                &num_clusters, &xv->xr_list,
1373                                                &ext_flags);
1374                 if (ret) {
1375                         mlog_errno(ret);
1376                         goto out;
1377                 }
1378
1379                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1380
1381                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1382
1383                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1384                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1385                                                &bh, NULL);
1386                         if (ret) {
1387                                 mlog_errno(ret);
1388                                 goto out;
1389                         }
1390
1391                         ret = ocfs2_journal_access(handle,
1392                                                    INODE_CACHE(inode),
1393                                                    bh,
1394                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1395                         if (ret < 0) {
1396                                 mlog_errno(ret);
1397                                 goto out;
1398                         }
1399
1400                         cp_len = value_len > blocksize ? blocksize : value_len;
1401                         memcpy(bh->b_data, value, cp_len);
1402                         value_len -= cp_len;
1403                         value += cp_len;
1404                         if (cp_len < blocksize)
1405                                 memset(bh->b_data + cp_len, 0,
1406                                        blocksize - cp_len);
1407
1408                         ocfs2_journal_dirty(handle, bh);
1409                         brelse(bh);
1410                         bh = NULL;
1411
1412                         /*
1413                          * XXX: do we need to empty all the following
1414                          * blocks in this cluster?
1415                          */
1416                         if (!value_len)
1417                                 break;
1418                 }
1419                 cpos += num_clusters;
1420         }
1421 out:
1422         brelse(bh);
1423
1424         return ret;
1425 }
1426
1427 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1428                                        int num_entries)
1429 {
1430         int free_space;
1431
1432         if (!needed_space)
1433                 return 0;
1434
1435         free_space = free_start -
1436                 sizeof(struct ocfs2_xattr_header) -
1437                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1438                 OCFS2_XATTR_HEADER_GAP;
1439         if (free_space < 0)
1440                 return -EIO;
1441         if (free_space < needed_space)
1442                 return -ENOSPC;
1443
1444         return 0;
1445 }
1446
1447 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1448                                    int type)
1449 {
1450         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1451 }
1452
1453 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1454 {
1455         loc->xl_ops->xlo_journal_dirty(handle, loc);
1456 }
1457
1458 /* Give a pointer into the storage for the given offset */
1459 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1460 {
1461         BUG_ON(offset >= loc->xl_size);
1462         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1463 }
1464
1465 /*
1466  * Wipe the name+value pair and allow the storage to reclaim it.  This
1467  * must be followed by either removal of the entry or a call to
1468  * ocfs2_xa_add_namevalue().
1469  */
1470 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1471 {
1472         loc->xl_ops->xlo_wipe_namevalue(loc);
1473 }
1474
1475 /*
1476  * Find lowest offset to a name+value pair.  This is the start of our
1477  * downward-growing free space.
1478  */
1479 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1480 {
1481         return loc->xl_ops->xlo_get_free_start(loc);
1482 }
1483
1484 /* Can we reuse loc->xl_entry for xi? */
1485 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1486                                     struct ocfs2_xattr_info *xi)
1487 {
1488         return loc->xl_ops->xlo_can_reuse(loc, xi);
1489 }
1490
1491 /* How much free space is needed to set the new value */
1492 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1493                                 struct ocfs2_xattr_info *xi)
1494 {
1495         return loc->xl_ops->xlo_check_space(loc, xi);
1496 }
1497
1498 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1499 {
1500         loc->xl_ops->xlo_add_entry(loc, name_hash);
1501         loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1502         /*
1503          * We can't leave the new entry's xe_name_offset at zero or
1504          * add_namevalue() will go nuts.  We set it to the size of our
1505          * storage so that it can never be less than any other entry.
1506          */
1507         loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1508 }
1509
1510 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1511                                    struct ocfs2_xattr_info *xi)
1512 {
1513         int size = namevalue_size_xi(xi);
1514         int nameval_offset;
1515         char *nameval_buf;
1516
1517         loc->xl_ops->xlo_add_namevalue(loc, size);
1518         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1519         loc->xl_entry->xe_name_len = xi->xi_name_len;
1520         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1521         ocfs2_xattr_set_local(loc->xl_entry,
1522                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1523
1524         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1525         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1526         memset(nameval_buf, 0, size);
1527         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1528 }
1529
1530 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1531                                     struct ocfs2_xattr_value_buf *vb)
1532 {
1533         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1534         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1535
1536         /* Value bufs are for value trees */
1537         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1538         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1539                (name_size + OCFS2_XATTR_ROOT_SIZE));
1540
1541         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1542         vb->vb_xv =
1543                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1544                                                         nameval_offset +
1545                                                         name_size);
1546 }
1547
1548 static int ocfs2_xa_block_journal_access(handle_t *handle,
1549                                          struct ocfs2_xa_loc *loc, int type)
1550 {
1551         struct buffer_head *bh = loc->xl_storage;
1552         ocfs2_journal_access_func access;
1553
1554         if (loc->xl_size == (bh->b_size -
1555                              offsetof(struct ocfs2_xattr_block,
1556                                       xb_attrs.xb_header)))
1557                 access = ocfs2_journal_access_xb;
1558         else
1559                 access = ocfs2_journal_access_di;
1560         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1561 }
1562
1563 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1564                                          struct ocfs2_xa_loc *loc)
1565 {
1566         struct buffer_head *bh = loc->xl_storage;
1567
1568         ocfs2_journal_dirty(handle, bh);
1569 }
1570
1571 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1572                                            int offset)
1573 {
1574         return (char *)loc->xl_header + offset;
1575 }
1576
1577 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1578                                     struct ocfs2_xattr_info *xi)
1579 {
1580         /*
1581          * Block storage is strict.  If the sizes aren't exact, we will
1582          * remove the old one and reinsert the new.
1583          */
1584         return namevalue_size_xe(loc->xl_entry) ==
1585                 namevalue_size_xi(xi);
1586 }
1587
1588 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1589 {
1590         struct ocfs2_xattr_header *xh = loc->xl_header;
1591         int i, count = le16_to_cpu(xh->xh_count);
1592         int offset, free_start = loc->xl_size;
1593
1594         for (i = 0; i < count; i++) {
1595                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1596                 if (offset < free_start)
1597                         free_start = offset;
1598         }
1599
1600         return free_start;
1601 }
1602
1603 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1604                                       struct ocfs2_xattr_info *xi)
1605 {
1606         int count = le16_to_cpu(loc->xl_header->xh_count);
1607         int free_start = ocfs2_xa_get_free_start(loc);
1608         int needed_space = ocfs2_xi_entry_usage(xi);
1609
1610         /*
1611          * Block storage will reclaim the original entry before inserting
1612          * the new value, so we only need the difference.  If the new
1613          * entry is smaller than the old one, we don't need anything.
1614          */
1615         if (loc->xl_entry) {
1616                 /* Don't need space if we're reusing! */
1617                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1618                         needed_space = 0;
1619                 else
1620                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1621         }
1622         if (needed_space < 0)
1623                 needed_space = 0;
1624         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1625 }
1626
1627 /*
1628  * Block storage for xattrs keeps the name+value pairs compacted.  When
1629  * we remove one, we have to shift any that preceded it towards the end.
1630  */
1631 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1632 {
1633         int i, offset;
1634         int namevalue_offset, first_namevalue_offset, namevalue_size;
1635         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1636         struct ocfs2_xattr_header *xh = loc->xl_header;
1637         int count = le16_to_cpu(xh->xh_count);
1638
1639         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1640         namevalue_size = namevalue_size_xe(entry);
1641         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1642
1643         /* Shift the name+value pairs */
1644         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1645                 (char *)xh + first_namevalue_offset,
1646                 namevalue_offset - first_namevalue_offset);
1647         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1648
1649         /* Now tell xh->xh_entries about it */
1650         for (i = 0; i < count; i++) {
1651                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1652                 if (offset <= namevalue_offset)
1653                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1654                                      namevalue_size);
1655         }
1656
1657         /*
1658          * Note that we don't update xh_free_start or xh_name_value_len
1659          * because they're not used in block-stored xattrs.
1660          */
1661 }
1662
1663 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1664 {
1665         int count = le16_to_cpu(loc->xl_header->xh_count);
1666         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1667         le16_add_cpu(&loc->xl_header->xh_count, 1);
1668         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1669 }
1670
1671 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1672 {
1673         int free_start = ocfs2_xa_get_free_start(loc);
1674
1675         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1676 }
1677
1678 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1679                                           struct ocfs2_xattr_value_buf *vb)
1680 {
1681         struct buffer_head *bh = loc->xl_storage;
1682
1683         if (loc->xl_size == (bh->b_size -
1684                              offsetof(struct ocfs2_xattr_block,
1685                                       xb_attrs.xb_header)))
1686                 vb->vb_access = ocfs2_journal_access_xb;
1687         else
1688                 vb->vb_access = ocfs2_journal_access_di;
1689         vb->vb_bh = bh;
1690 }
1691
1692 /*
1693  * Operations for xattrs stored in blocks.  This includes inline inode
1694  * storage and unindexed ocfs2_xattr_blocks.
1695  */
1696 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1697         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1698         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1699         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1700         .xlo_check_space        = ocfs2_xa_block_check_space,
1701         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1702         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1703         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1704         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1705         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1706         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1707 };
1708
1709 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1710                                           struct ocfs2_xa_loc *loc, int type)
1711 {
1712         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1713
1714         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1715 }
1716
1717 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1718                                           struct ocfs2_xa_loc *loc)
1719 {
1720         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1721
1722         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1723 }
1724
1725 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1726                                             int offset)
1727 {
1728         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1729         int block, block_offset;
1730
1731         /* The header is at the front of the bucket */
1732         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1733         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1734
1735         return bucket_block(bucket, block) + block_offset;
1736 }
1737
1738 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1739                                      struct ocfs2_xattr_info *xi)
1740 {
1741         return namevalue_size_xe(loc->xl_entry) >=
1742                 namevalue_size_xi(xi);
1743 }
1744
1745 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1746 {
1747         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1748         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1749 }
1750
1751 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1752                                          int free_start, int size)
1753 {
1754         /*
1755          * We need to make sure that the name+value pair fits within
1756          * one block.
1757          */
1758         if (((free_start - size) >> sb->s_blocksize_bits) !=
1759             ((free_start - 1) >> sb->s_blocksize_bits))
1760                 free_start -= free_start % sb->s_blocksize;
1761
1762         return free_start;
1763 }
1764
1765 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1766                                        struct ocfs2_xattr_info *xi)
1767 {
1768         int rc;
1769         int count = le16_to_cpu(loc->xl_header->xh_count);
1770         int free_start = ocfs2_xa_get_free_start(loc);
1771         int needed_space = ocfs2_xi_entry_usage(xi);
1772         int size = namevalue_size_xi(xi);
1773         struct super_block *sb = loc->xl_inode->i_sb;
1774
1775         /*
1776          * Bucket storage does not reclaim name+value pairs it cannot
1777          * reuse.  They live as holes until the bucket fills, and then
1778          * the bucket is defragmented.  However, the bucket can reclaim
1779          * the ocfs2_xattr_entry.
1780          */
1781         if (loc->xl_entry) {
1782                 /* Don't need space if we're reusing! */
1783                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1784                         needed_space = 0;
1785                 else
1786                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1787         }
1788         BUG_ON(needed_space < 0);
1789
1790         if (free_start < size) {
1791                 if (needed_space)
1792                         return -ENOSPC;
1793         } else {
1794                 /*
1795                  * First we check if it would fit in the first place.
1796                  * Below, we align the free start to a block.  This may
1797                  * slide us below the minimum gap.  By checking unaligned
1798                  * first, we avoid that error.
1799                  */
1800                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1801                                                  count);
1802                 if (rc)
1803                         return rc;
1804                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1805                                                            size);
1806         }
1807         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1808 }
1809
1810 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1811 {
1812         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1813                      -namevalue_size_xe(loc->xl_entry));
1814 }
1815
1816 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1817 {
1818         struct ocfs2_xattr_header *xh = loc->xl_header;
1819         int count = le16_to_cpu(xh->xh_count);
1820         int low = 0, high = count - 1, tmp;
1821         struct ocfs2_xattr_entry *tmp_xe;
1822
1823         /*
1824          * We keep buckets sorted by name_hash, so we need to find
1825          * our insert place.
1826          */
1827         while (low <= high && count) {
1828                 tmp = (low + high) / 2;
1829                 tmp_xe = &xh->xh_entries[tmp];
1830
1831                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1832                         low = tmp + 1;
1833                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1834                         high = tmp - 1;
1835                 else {
1836                         low = tmp;
1837                         break;
1838                 }
1839         }
1840
1841         if (low != count)
1842                 memmove(&xh->xh_entries[low + 1],
1843                         &xh->xh_entries[low],
1844                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1845
1846         le16_add_cpu(&xh->xh_count, 1);
1847         loc->xl_entry = &xh->xh_entries[low];
1848         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1849 }
1850
1851 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1852 {
1853         int free_start = ocfs2_xa_get_free_start(loc);
1854         struct ocfs2_xattr_header *xh = loc->xl_header;
1855         struct super_block *sb = loc->xl_inode->i_sb;
1856         int nameval_offset;
1857
1858         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1859         nameval_offset = free_start - size;
1860         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1861         xh->xh_free_start = cpu_to_le16(nameval_offset);
1862         le16_add_cpu(&xh->xh_name_value_len, size);
1863
1864 }
1865
1866 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1867                                            struct ocfs2_xattr_value_buf *vb)
1868 {
1869         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1870         struct super_block *sb = loc->xl_inode->i_sb;
1871         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1872         int size = namevalue_size_xe(loc->xl_entry);
1873         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1874
1875         /* Values are not allowed to straddle block boundaries */
1876         BUG_ON(block_offset !=
1877                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1878         /* We expect the bucket to be filled in */
1879         BUG_ON(!bucket->bu_bhs[block_offset]);
1880
1881         vb->vb_access = ocfs2_journal_access;
1882         vb->vb_bh = bucket->bu_bhs[block_offset];
1883 }
1884
1885 /* Operations for xattrs stored in buckets. */
1886 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1887         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1888         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1889         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1890         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1891         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1892         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1893         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1894         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1895         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1896         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1897 };
1898
1899 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1900 {
1901         struct ocfs2_xattr_value_buf vb;
1902
1903         if (ocfs2_xattr_is_local(loc->xl_entry))
1904                 return 0;
1905
1906         ocfs2_xa_fill_value_buf(loc, &vb);
1907         return le32_to_cpu(vb.vb_xv->xr_clusters);
1908 }
1909
1910 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1911                                    struct ocfs2_xattr_set_ctxt *ctxt)
1912 {
1913         int trunc_rc, access_rc;
1914         struct ocfs2_xattr_value_buf vb;
1915
1916         ocfs2_xa_fill_value_buf(loc, &vb);
1917         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1918                                               ctxt);
1919
1920         /*
1921          * The caller of ocfs2_xa_value_truncate() has already called
1922          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1923          * calls ocfs2_extend_trans().  This may commit the previous
1924          * transaction and open a new one.  If this is a bucket, truncate
1925          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1926          * the caller is expecting to dirty the entire bucket.  So we must
1927          * reset the journal work.  We do this even if truncate has failed,
1928          * as it could have failed after committing the extend.
1929          */
1930         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1931                                             OCFS2_JOURNAL_ACCESS_WRITE);
1932
1933         /* Errors in truncate take precedence */
1934         return trunc_rc ? trunc_rc : access_rc;
1935 }
1936
1937 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1938 {
1939         int index, count;
1940         struct ocfs2_xattr_header *xh = loc->xl_header;
1941         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1942
1943         ocfs2_xa_wipe_namevalue(loc);
1944         loc->xl_entry = NULL;
1945
1946         le16_add_cpu(&xh->xh_count, -1);
1947         count = le16_to_cpu(xh->xh_count);
1948
1949         /*
1950          * Only zero out the entry if there are more remaining.  This is
1951          * important for an empty bucket, as it keeps track of the
1952          * bucket's hash value.  It doesn't hurt empty block storage.
1953          */
1954         if (count) {
1955                 index = ((char *)entry - (char *)&xh->xh_entries) /
1956                         sizeof(struct ocfs2_xattr_entry);
1957                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1958                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1959                 memset(&xh->xh_entries[count], 0,
1960                        sizeof(struct ocfs2_xattr_entry));
1961         }
1962 }
1963
1964 /*
1965  * If we have a problem adjusting the size of an external value during
1966  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1967  * in an intermediate state.  For example, the value may be partially
1968  * truncated.
1969  *
1970  * If the value tree hasn't changed, the extend/truncate went nowhere.
1971  * We have nothing to do.  The caller can treat it as a straight error.
1972  *
1973  * If the value tree got partially truncated, we now have a corrupted
1974  * extended attribute.  We're going to wipe its entry and leak the
1975  * clusters.  Better to leak some storage than leave a corrupt entry.
1976  *
1977  * If the value tree grew, it obviously didn't grow enough for the
1978  * new entry.  We're not going to try and reclaim those clusters either.
1979  * If there was already an external value there (orig_clusters != 0),
1980  * the new clusters are attached safely and we can just leave the old
1981  * value in place.  If there was no external value there, we remove
1982  * the entry.
1983  *
1984  * This way, the xattr block we store in the journal will be consistent.
1985  * If the size change broke because of the journal, no changes will hit
1986  * disk anyway.
1987  */
1988 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1989                                             const char *what,
1990                                             unsigned int orig_clusters)
1991 {
1992         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1993         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1994                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1995
1996         if (new_clusters < orig_clusters) {
1997                 mlog(ML_ERROR,
1998                      "Partial truncate while %s xattr %.*s.  Leaking "
1999                      "%u clusters and removing the entry\n",
2000                      what, loc->xl_entry->xe_name_len, nameval_buf,
2001                      orig_clusters - new_clusters);
2002                 ocfs2_xa_remove_entry(loc);
2003         } else if (!orig_clusters) {
2004                 mlog(ML_ERROR,
2005                      "Unable to allocate an external value for xattr "
2006                      "%.*s safely.  Leaking %u clusters and removing the "
2007                      "entry\n",
2008                      loc->xl_entry->xe_name_len, nameval_buf,
2009                      new_clusters - orig_clusters);
2010                 ocfs2_xa_remove_entry(loc);
2011         } else if (new_clusters > orig_clusters)
2012                 mlog(ML_ERROR,
2013                      "Unable to grow xattr %.*s safely.  %u new clusters "
2014                      "have been added, but the value will not be "
2015                      "modified\n",
2016                      loc->xl_entry->xe_name_len, nameval_buf,
2017                      new_clusters - orig_clusters);
2018 }
2019
2020 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2021                            struct ocfs2_xattr_set_ctxt *ctxt)
2022 {
2023         int rc = 0;
2024         unsigned int orig_clusters;
2025
2026         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2027                 orig_clusters = ocfs2_xa_value_clusters(loc);
2028                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2029                 if (rc) {
2030                         mlog_errno(rc);
2031                         /*
2032                          * Since this is remove, we can return 0 if
2033                          * ocfs2_xa_cleanup_value_truncate() is going to
2034                          * wipe the entry anyway.  So we check the
2035                          * cluster count as well.
2036                          */
2037                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2038                                 rc = 0;
2039                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2040                                                         orig_clusters);
2041                         if (rc)
2042                                 goto out;
2043                 }
2044         }
2045
2046         ocfs2_xa_remove_entry(loc);
2047
2048 out:
2049         return rc;
2050 }
2051
2052 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2053 {
2054         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2055         char *nameval_buf;
2056
2057         nameval_buf = ocfs2_xa_offset_pointer(loc,
2058                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2059         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2060 }
2061
2062 /*
2063  * Take an existing entry and make it ready for the new value.  This
2064  * won't allocate space, but it may free space.  It should be ready for
2065  * ocfs2_xa_prepare_entry() to finish the work.
2066  */
2067 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2068                                 struct ocfs2_xattr_info *xi,
2069                                 struct ocfs2_xattr_set_ctxt *ctxt)
2070 {
2071         int rc = 0;
2072         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2073         unsigned int orig_clusters;
2074         char *nameval_buf;
2075         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2076         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2077
2078         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2079                name_size);
2080
2081         nameval_buf = ocfs2_xa_offset_pointer(loc,
2082                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2083         if (xe_local) {
2084                 memset(nameval_buf + name_size, 0,
2085                        namevalue_size_xe(loc->xl_entry) - name_size);
2086                 if (!xi_local)
2087                         ocfs2_xa_install_value_root(loc);
2088         } else {
2089                 orig_clusters = ocfs2_xa_value_clusters(loc);
2090                 if (xi_local) {
2091                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2092                         if (rc < 0)
2093                                 mlog_errno(rc);
2094                         else
2095                                 memset(nameval_buf + name_size, 0,
2096                                        namevalue_size_xe(loc->xl_entry) -
2097                                        name_size);
2098                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2099                            xi->xi_value_len) {
2100                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2101                                                      ctxt);
2102                         if (rc < 0)
2103                                 mlog_errno(rc);
2104                 }
2105
2106                 if (rc) {
2107                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2108                                                         orig_clusters);
2109                         goto out;
2110                 }
2111         }
2112
2113         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2114         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2115
2116 out:
2117         return rc;
2118 }
2119
2120 /*
2121  * Prepares loc->xl_entry to receive the new xattr.  This includes
2122  * properly setting up the name+value pair region.  If loc->xl_entry
2123  * already exists, it will take care of modifying it appropriately.
2124  *
2125  * Note that this modifies the data.  You did journal_access already,
2126  * right?
2127  */
2128 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2129                                   struct ocfs2_xattr_info *xi,
2130                                   u32 name_hash,
2131                                   struct ocfs2_xattr_set_ctxt *ctxt)
2132 {
2133         int rc = 0;
2134         unsigned int orig_clusters;
2135         __le64 orig_value_size = 0;
2136
2137         rc = ocfs2_xa_check_space(loc, xi);
2138         if (rc)
2139                 goto out;
2140
2141         if (loc->xl_entry) {
2142                 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2143                         orig_value_size = loc->xl_entry->xe_value_size;
2144                         rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2145                         if (rc)
2146                                 goto out;
2147                         goto alloc_value;
2148                 }
2149
2150                 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2151                         orig_clusters = ocfs2_xa_value_clusters(loc);
2152                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2153                         if (rc) {
2154                                 mlog_errno(rc);
2155                                 ocfs2_xa_cleanup_value_truncate(loc,
2156                                                                 "overwriting",
2157                                                                 orig_clusters);
2158                                 goto out;
2159                         }
2160                 }
2161                 ocfs2_xa_wipe_namevalue(loc);
2162         } else
2163                 ocfs2_xa_add_entry(loc, name_hash);
2164
2165         /*
2166          * If we get here, we have a blank entry.  Fill it.  We grow our
2167          * name+value pair back from the end.
2168          */
2169         ocfs2_xa_add_namevalue(loc, xi);
2170         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2171                 ocfs2_xa_install_value_root(loc);
2172
2173 alloc_value:
2174         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2175                 orig_clusters = ocfs2_xa_value_clusters(loc);
2176                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2177                 if (rc < 0) {
2178                         ctxt->set_abort = 1;
2179                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2180                                                         orig_clusters);
2181                         /*
2182                          * If we were growing an existing value,
2183                          * ocfs2_xa_cleanup_value_truncate() won't remove
2184                          * the entry. We need to restore the original value
2185                          * size.
2186                          */
2187                         if (loc->xl_entry) {
2188                                 BUG_ON(!orig_value_size);
2189                                 loc->xl_entry->xe_value_size = orig_value_size;
2190                         }
2191                         mlog_errno(rc);
2192                 }
2193         }
2194
2195 out:
2196         return rc;
2197 }
2198
2199 /*
2200  * Store the value portion of the name+value pair.  This will skip
2201  * values that are stored externally.  Their tree roots were set up
2202  * by ocfs2_xa_prepare_entry().
2203  */
2204 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2205                                 struct ocfs2_xattr_info *xi,
2206                                 struct ocfs2_xattr_set_ctxt *ctxt)
2207 {
2208         int rc = 0;
2209         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2210         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2211         char *nameval_buf;
2212         struct ocfs2_xattr_value_buf vb;
2213
2214         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2215         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2216                 ocfs2_xa_fill_value_buf(loc, &vb);
2217                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2218                                                      ctxt->handle, &vb,
2219                                                      xi->xi_value,
2220                                                      xi->xi_value_len);
2221         } else
2222                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2223
2224         return rc;
2225 }
2226
2227 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2228                         struct ocfs2_xattr_info *xi,
2229                         struct ocfs2_xattr_set_ctxt *ctxt)
2230 {
2231         int ret;
2232         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2233                                               xi->xi_name_len);
2234
2235         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2236                                       OCFS2_JOURNAL_ACCESS_WRITE);
2237         if (ret) {
2238                 mlog_errno(ret);
2239                 goto out;
2240         }
2241
2242         /*
2243          * From here on out, everything is going to modify the buffer a
2244          * little.  Errors are going to leave the xattr header in a
2245          * sane state.  Thus, even with errors we dirty the sucker.
2246          */
2247
2248         /* Don't worry, we are never called with !xi_value and !xl_entry */
2249         if (!xi->xi_value) {
2250                 ret = ocfs2_xa_remove(loc, ctxt);
2251                 goto out_dirty;
2252         }
2253
2254         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2255         if (ret) {
2256                 if (ret != -ENOSPC)
2257                         mlog_errno(ret);
2258                 goto out_dirty;
2259         }
2260
2261         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2262         if (ret)
2263                 mlog_errno(ret);
2264
2265 out_dirty:
2266         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2267
2268 out:
2269         return ret;
2270 }
2271
2272 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2273                                      struct inode *inode,
2274                                      struct buffer_head *bh,
2275                                      struct ocfs2_xattr_entry *entry)
2276 {
2277         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2278
2279         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2280
2281         loc->xl_inode = inode;
2282         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2283         loc->xl_storage = bh;
2284         loc->xl_entry = entry;
2285         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2286         loc->xl_header =
2287                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2288                                               loc->xl_size);
2289 }
2290
2291 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2292                                           struct inode *inode,
2293                                           struct buffer_head *bh,
2294                                           struct ocfs2_xattr_entry *entry)
2295 {
2296         struct ocfs2_xattr_block *xb =
2297                 (struct ocfs2_xattr_block *)bh->b_data;
2298
2299         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2300
2301         loc->xl_inode = inode;
2302         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2303         loc->xl_storage = bh;
2304         loc->xl_header = &(xb->xb_attrs.xb_header);
2305         loc->xl_entry = entry;
2306         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2307                                              xb_attrs.xb_header);
2308 }
2309
2310 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2311                                            struct ocfs2_xattr_bucket *bucket,
2312                                            struct ocfs2_xattr_entry *entry)
2313 {
2314         loc->xl_inode = bucket->bu_inode;
2315         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2316         loc->xl_storage = bucket;
2317         loc->xl_header = bucket_xh(bucket);
2318         loc->xl_entry = entry;
2319         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2320 }
2321
2322 /*
2323  * In xattr remove, if it is stored outside and refcounted, we may have
2324  * the chance to split the refcount tree. So need the allocators.
2325  */
2326 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2327                                         struct ocfs2_xattr_value_root *xv,
2328                                         struct ocfs2_caching_info *ref_ci,
2329                                         struct buffer_head *ref_root_bh,
2330                                         struct ocfs2_alloc_context **meta_ac,
2331                                         int *ref_credits)
2332 {
2333         int ret, meta_add = 0;
2334         u32 p_cluster, num_clusters;
2335         unsigned int ext_flags;
2336
2337         *ref_credits = 0;
2338         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2339                                        &num_clusters,
2340                                        &xv->xr_list,
2341                                        &ext_flags);
2342         if (ret) {
2343                 mlog_errno(ret);
2344                 goto out;
2345         }
2346
2347         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2348                 goto out;
2349
2350         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2351                                                  ref_root_bh, xv,
2352                                                  &meta_add, ref_credits);
2353         if (ret) {
2354                 mlog_errno(ret);
2355                 goto out;
2356         }
2357
2358         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2359                                                 meta_add, meta_ac);
2360         if (ret)
2361                 mlog_errno(ret);
2362
2363 out:
2364         return ret;
2365 }
2366
2367 static int ocfs2_remove_value_outside(struct inode*inode,
2368                                       struct ocfs2_xattr_value_buf *vb,
2369                                       struct ocfs2_xattr_header *header,
2370                                       struct ocfs2_caching_info *ref_ci,
2371                                       struct buffer_head *ref_root_bh)
2372 {
2373         int ret = 0, i, ref_credits;
2374         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2375         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2376         void *val;
2377
2378         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2379
2380         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2381                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2382
2383                 if (ocfs2_xattr_is_local(entry))
2384                         continue;
2385
2386                 val = (void *)header +
2387                         le16_to_cpu(entry->xe_name_offset);
2388                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2389                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2390
2391                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2392                                                          ref_ci, ref_root_bh,
2393                                                          &ctxt.meta_ac,
2394                                                          &ref_credits);
2395
2396                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2397                                         ocfs2_remove_extent_credits(osb->sb));
2398                 if (IS_ERR(ctxt.handle)) {
2399                         ret = PTR_ERR(ctxt.handle);
2400                         mlog_errno(ret);
2401                         break;
2402                 }
2403
2404                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2405
2406                 ocfs2_commit_trans(osb, ctxt.handle);
2407                 if (ctxt.meta_ac) {
2408                         ocfs2_free_alloc_context(ctxt.meta_ac);
2409                         ctxt.meta_ac = NULL;
2410                 }
2411
2412                 if (ret < 0) {
2413                         mlog_errno(ret);
2414                         break;
2415                 }
2416
2417         }
2418
2419         if (ctxt.meta_ac)
2420                 ocfs2_free_alloc_context(ctxt.meta_ac);
2421         ocfs2_schedule_truncate_log_flush(osb, 1);
2422         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2423         return ret;
2424 }
2425
2426 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2427                                     struct buffer_head *di_bh,
2428                                     struct ocfs2_caching_info *ref_ci,
2429                                     struct buffer_head *ref_root_bh)
2430 {
2431
2432         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2433         struct ocfs2_xattr_header *header;
2434         int ret;
2435         struct ocfs2_xattr_value_buf vb = {
2436                 .vb_bh = di_bh,
2437                 .vb_access = ocfs2_journal_access_di,
2438         };
2439
2440         header = (struct ocfs2_xattr_header *)
2441                  ((void *)di + inode->i_sb->s_blocksize -
2442                  le16_to_cpu(di->i_xattr_inline_size));
2443
2444         ret = ocfs2_remove_value_outside(inode, &vb, header,
2445                                          ref_ci, ref_root_bh);
2446
2447         return ret;
2448 }
2449
2450 struct ocfs2_rm_xattr_bucket_para {
2451         struct ocfs2_caching_info *ref_ci;
2452         struct buffer_head *ref_root_bh;
2453 };
2454
2455 static int ocfs2_xattr_block_remove(struct inode *inode,
2456                                     struct buffer_head *blk_bh,
2457                                     struct ocfs2_caching_info *ref_ci,
2458                                     struct buffer_head *ref_root_bh)
2459 {
2460         struct ocfs2_xattr_block *xb;
2461         int ret = 0;
2462         struct ocfs2_xattr_value_buf vb = {
2463                 .vb_bh = blk_bh,
2464                 .vb_access = ocfs2_journal_access_xb,
2465         };
2466         struct ocfs2_rm_xattr_bucket_para args = {
2467                 .ref_ci = ref_ci,
2468                 .ref_root_bh = ref_root_bh,
2469         };
2470
2471         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2472         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2473                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2474                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2475                                                  ref_ci, ref_root_bh);
2476         } else
2477                 ret = ocfs2_iterate_xattr_index_block(inode,
2478                                                 blk_bh,
2479                                                 ocfs2_rm_xattr_cluster,
2480                                                 &args);
2481
2482         return ret;
2483 }
2484
2485 static int ocfs2_xattr_free_block(struct inode *inode,
2486                                   u64 block,
2487                                   struct ocfs2_caching_info *ref_ci,
2488                                   struct buffer_head *ref_root_bh)
2489 {
2490         struct inode *xb_alloc_inode;
2491         struct buffer_head *xb_alloc_bh = NULL;
2492         struct buffer_head *blk_bh = NULL;
2493         struct ocfs2_xattr_block *xb;
2494         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2495         handle_t *handle;
2496         int ret = 0;
2497         u64 blk, bg_blkno;
2498         u16 bit;
2499
2500         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2501         if (ret < 0) {
2502                 mlog_errno(ret);
2503                 goto out;
2504         }
2505
2506         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2507         if (ret < 0) {
2508                 mlog_errno(ret);
2509                 goto out;
2510         }
2511
2512         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2513         blk = le64_to_cpu(xb->xb_blkno);
2514         bit = le16_to_cpu(xb->xb_suballoc_bit);
2515         if (xb->xb_suballoc_loc)
2516                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2517         else
2518                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2519
2520         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2521                                 EXTENT_ALLOC_SYSTEM_INODE,
2522                                 le16_to_cpu(xb->xb_suballoc_slot));
2523         if (!xb_alloc_inode) {
2524                 ret = -ENOMEM;
2525                 mlog_errno(ret);
2526                 goto out;
2527         }
2528         inode_lock(xb_alloc_inode);
2529
2530         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2531         if (ret < 0) {
2532                 mlog_errno(ret);
2533                 goto out_mutex;
2534         }
2535
2536         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2537         if (IS_ERR(handle)) {
2538                 ret = PTR_ERR(handle);
2539                 mlog_errno(ret);
2540                 goto out_unlock;
2541         }
2542
2543         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2544                                        bit, bg_blkno, 1);
2545         if (ret < 0)
2546                 mlog_errno(ret);
2547
2548         ocfs2_commit_trans(osb, handle);
2549 out_unlock:
2550         ocfs2_inode_unlock(xb_alloc_inode, 1);
2551         brelse(xb_alloc_bh);
2552 out_mutex:
2553         inode_unlock(xb_alloc_inode);
2554         iput(xb_alloc_inode);
2555 out:
2556         brelse(blk_bh);
2557         return ret;
2558 }
2559
2560 /*
2561  * ocfs2_xattr_remove()
2562  *
2563  * Free extended attribute resources associated with this inode.
2564  */
2565 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2566 {
2567         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2568         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2569         struct ocfs2_refcount_tree *ref_tree = NULL;
2570         struct buffer_head *ref_root_bh = NULL;
2571         struct ocfs2_caching_info *ref_ci = NULL;
2572         handle_t *handle;
2573         int ret;
2574
2575         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2576                 return 0;
2577
2578         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2579                 return 0;
2580
2581         if (ocfs2_is_refcount_inode(inode)) {
2582                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2583                                                le64_to_cpu(di->i_refcount_loc),
2584                                                1, &ref_tree, &ref_root_bh);
2585                 if (ret) {
2586                         mlog_errno(ret);
2587                         goto out;
2588                 }
2589                 ref_ci = &ref_tree->rf_ci;
2590
2591         }
2592
2593         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2594                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2595                                                ref_ci, ref_root_bh);
2596                 if (ret < 0) {
2597                         mlog_errno(ret);
2598                         goto out;
2599                 }
2600         }
2601
2602         if (di->i_xattr_loc) {
2603                 ret = ocfs2_xattr_free_block(inode,
2604                                              le64_to_cpu(di->i_xattr_loc),
2605                                              ref_ci, ref_root_bh);
2606                 if (ret < 0) {
2607                         mlog_errno(ret);
2608                         goto out;
2609                 }
2610         }
2611
2612         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2613                                    OCFS2_INODE_UPDATE_CREDITS);
2614         if (IS_ERR(handle)) {
2615                 ret = PTR_ERR(handle);
2616                 mlog_errno(ret);
2617                 goto out;
2618         }
2619         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2620                                       OCFS2_JOURNAL_ACCESS_WRITE);
2621         if (ret) {
2622                 mlog_errno(ret);
2623                 goto out_commit;
2624         }
2625
2626         di->i_xattr_loc = 0;
2627
2628         spin_lock(&oi->ip_lock);
2629         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2630         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2631         spin_unlock(&oi->ip_lock);
2632         ocfs2_update_inode_fsync_trans(handle, inode, 0);
2633
2634         ocfs2_journal_dirty(handle, di_bh);
2635 out_commit:
2636         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2637 out:
2638         if (ref_tree)
2639                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2640         brelse(ref_root_bh);
2641         return ret;
2642 }
2643
2644 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2645                                         struct ocfs2_dinode *di)
2646 {
2647         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2648         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2649         int free;
2650
2651         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2652                 return 0;
2653
2654         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2655                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2656                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2657         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2658                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2659                         le64_to_cpu(di->i_size);
2660         } else {
2661                 struct ocfs2_extent_list *el = &di->id2.i_list;
2662                 free = (le16_to_cpu(el->l_count) -
2663                         le16_to_cpu(el->l_next_free_rec)) *
2664                         sizeof(struct ocfs2_extent_rec);
2665         }
2666         if (free >= xattrsize)
2667                 return 1;
2668
2669         return 0;
2670 }
2671
2672 /*
2673  * ocfs2_xattr_ibody_find()
2674  *
2675  * Find extended attribute in inode block and
2676  * fill search info into struct ocfs2_xattr_search.
2677  */
2678 static int ocfs2_xattr_ibody_find(struct inode *inode,
2679                                   int name_index,
2680                                   const char *name,
2681                                   struct ocfs2_xattr_search *xs)
2682 {
2683         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2684         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2685         int ret;
2686         int has_space = 0;
2687
2688         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2689                 return 0;
2690
2691         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2692                 down_read(&oi->ip_alloc_sem);
2693                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2694                 up_read(&oi->ip_alloc_sem);
2695                 if (!has_space)
2696                         return 0;
2697         }
2698
2699         xs->xattr_bh = xs->inode_bh;
2700         xs->end = (void *)di + inode->i_sb->s_blocksize;
2701         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2702                 xs->header = (struct ocfs2_xattr_header *)
2703                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2704         else
2705                 xs->header = (struct ocfs2_xattr_header *)
2706                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2707         xs->base = (void *)xs->header;
2708         xs->here = xs->header->xh_entries;
2709
2710         /* Find the named attribute. */
2711         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2712                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2713                 if (ret && ret != -ENODATA)
2714                         return ret;
2715                 xs->not_found = ret;
2716         }
2717
2718         return 0;
2719 }
2720
2721 static int ocfs2_xattr_ibody_init(struct inode *inode,
2722                                   struct buffer_head *di_bh,
2723                                   struct ocfs2_xattr_set_ctxt *ctxt)
2724 {
2725         int ret;
2726         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2727         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2728         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2729         unsigned int xattrsize = osb->s_xattr_inline_size;
2730
2731         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2732                 ret = -ENOSPC;
2733                 goto out;
2734         }
2735
2736         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2737                                       OCFS2_JOURNAL_ACCESS_WRITE);
2738         if (ret) {
2739                 mlog_errno(ret);
2740                 goto out;
2741         }
2742
2743         /*
2744          * Adjust extent record count or inline data size
2745          * to reserve space for extended attribute.
2746          */
2747         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2748                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2749                 le16_add_cpu(&idata->id_count, -xattrsize);
2750         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2751                 struct ocfs2_extent_list *el = &di->id2.i_list;
2752                 le16_add_cpu(&el->l_count, -(xattrsize /
2753                                              sizeof(struct ocfs2_extent_rec)));
2754         }
2755         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2756
2757         spin_lock(&oi->ip_lock);
2758         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2759         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2760         spin_unlock(&oi->ip_lock);
2761
2762         ocfs2_journal_dirty(ctxt->handle, di_bh);
2763
2764 out:
2765         return ret;
2766 }
2767
2768 /*
2769  * ocfs2_xattr_ibody_set()
2770  *
2771  * Set, replace or remove an extended attribute into inode block.
2772  *
2773  */
2774 static int ocfs2_xattr_ibody_set(struct inode *inode,
2775                                  struct ocfs2_xattr_info *xi,
2776                                  struct ocfs2_xattr_search *xs,
2777                                  struct ocfs2_xattr_set_ctxt *ctxt)
2778 {
2779         int ret;
2780         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2781         struct ocfs2_xa_loc loc;
2782
2783         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2784                 return -ENOSPC;
2785
2786         down_write(&oi->ip_alloc_sem);
2787         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2788                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2789                 if (ret) {
2790                         if (ret != -ENOSPC)
2791                                 mlog_errno(ret);
2792                         goto out;
2793                 }
2794         }
2795
2796         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2797                                  xs->not_found ? NULL : xs->here);
2798         ret = ocfs2_xa_set(&loc, xi, ctxt);
2799         if (ret) {
2800                 if (ret != -ENOSPC)
2801                         mlog_errno(ret);
2802                 goto out;
2803         }
2804         xs->here = loc.xl_entry;
2805
2806 out:
2807         up_write(&oi->ip_alloc_sem);
2808
2809         return ret;
2810 }
2811
2812 /*
2813  * ocfs2_xattr_block_find()
2814  *
2815  * Find extended attribute in external block and
2816  * fill search info into struct ocfs2_xattr_search.
2817  */
2818 static int ocfs2_xattr_block_find(struct inode *inode,
2819                                   int name_index,
2820                                   const char *name,
2821                                   struct ocfs2_xattr_search *xs)
2822 {
2823         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2824         struct buffer_head *blk_bh = NULL;
2825         struct ocfs2_xattr_block *xb;
2826         int ret = 0;
2827
2828         if (!di->i_xattr_loc)
2829                 return ret;
2830
2831         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2832                                      &blk_bh);
2833         if (ret < 0) {
2834                 mlog_errno(ret);
2835                 return ret;
2836         }
2837
2838         xs->xattr_bh = blk_bh;
2839         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2840
2841         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2842                 xs->header = &xb->xb_attrs.xb_header;
2843                 xs->base = (void *)xs->header;
2844                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2845                 xs->here = xs->header->xh_entries;
2846
2847                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2848         } else
2849                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2850                                                    name_index,
2851                                                    name, xs);
2852
2853         if (ret && ret != -ENODATA) {
2854                 xs->xattr_bh = NULL;
2855                 goto cleanup;
2856         }
2857         xs->not_found = ret;
2858         return 0;
2859 cleanup:
2860         brelse(blk_bh);
2861
2862         return ret;
2863 }
2864
2865 static int ocfs2_create_xattr_block(struct inode *inode,
2866                                     struct buffer_head *inode_bh,
2867                                     struct ocfs2_xattr_set_ctxt *ctxt,
2868                                     int indexed,
2869                                     struct buffer_head **ret_bh)
2870 {
2871         int ret;
2872         u16 suballoc_bit_start;
2873         u32 num_got;
2874         u64 suballoc_loc, first_blkno;
2875         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2876         struct buffer_head *new_bh = NULL;
2877         struct ocfs2_xattr_block *xblk;
2878
2879         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2880                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2881         if (ret < 0) {
2882                 mlog_errno(ret);
2883                 goto end;
2884         }
2885
2886         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2887                                    &suballoc_loc, &suballoc_bit_start,
2888                                    &num_got, &first_blkno);
2889         if (ret < 0) {
2890                 mlog_errno(ret);
2891                 goto end;
2892         }
2893
2894         new_bh = sb_getblk(inode->i_sb, first_blkno);
2895         if (!new_bh) {
2896                 ret = -ENOMEM;
2897                 mlog_errno(ret);
2898                 goto end;
2899         }
2900
2901         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2902
2903         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2904                                       new_bh,
2905                                       OCFS2_JOURNAL_ACCESS_CREATE);
2906         if (ret < 0) {
2907                 mlog_errno(ret);
2908                 goto end;
2909         }
2910
2911         /* Initialize ocfs2_xattr_block */
2912         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2913         memset(xblk, 0, inode->i_sb->s_blocksize);
2914         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2915         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2916         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2917         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2918         xblk->xb_fs_generation =
2919                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2920         xblk->xb_blkno = cpu_to_le64(first_blkno);
2921         if (indexed) {
2922                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2923                 xr->xt_clusters = cpu_to_le32(1);
2924                 xr->xt_last_eb_blk = 0;
2925                 xr->xt_list.l_tree_depth = 0;
2926                 xr->xt_list.l_count = cpu_to_le16(
2927                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2928                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2929                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2930         }
2931         ocfs2_journal_dirty(ctxt->handle, new_bh);
2932
2933         /* Add it to the inode */
2934         di->i_xattr_loc = cpu_to_le64(first_blkno);
2935
2936         spin_lock(&OCFS2_I(inode)->ip_lock);
2937         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2938         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2939         spin_unlock(&OCFS2_I(inode)->ip_lock);
2940
2941         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2942
2943         *ret_bh = new_bh;
2944         new_bh = NULL;
2945
2946 end:
2947         brelse(new_bh);
2948         return ret;
2949 }
2950
2951 /*
2952  * ocfs2_xattr_block_set()
2953  *
2954  * Set, replace or remove an extended attribute into external block.
2955  *
2956  */
2957 static int ocfs2_xattr_block_set(struct inode *inode,
2958                                  struct ocfs2_xattr_info *xi,
2959                                  struct ocfs2_xattr_search *xs,
2960                                  struct ocfs2_xattr_set_ctxt *ctxt)
2961 {
2962         struct buffer_head *new_bh = NULL;
2963         struct ocfs2_xattr_block *xblk = NULL;
2964         int ret;
2965         struct ocfs2_xa_loc loc;
2966
2967         if (!xs->xattr_bh) {
2968                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2969                                                0, &new_bh);
2970                 if (ret) {
2971                         mlog_errno(ret);
2972                         goto end;
2973                 }
2974
2975                 xs->xattr_bh = new_bh;
2976                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2977                 xs->header = &xblk->xb_attrs.xb_header;
2978                 xs->base = (void *)xs->header;
2979                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2980                 xs->here = xs->header->xh_entries;
2981         } else
2982                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2983
2984         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2985                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2986                                               xs->not_found ? NULL : xs->here);
2987
2988                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2989                 if (!ret)
2990                         xs->here = loc.xl_entry;
2991                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2992                         goto end;
2993                 else {
2994                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2995                         if (ret)
2996                                 goto end;
2997                 }
2998         }
2999
3000         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
3001                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
3002
3003 end:
3004         return ret;
3005 }
3006
3007 /* Check whether the new xattr can be inserted into the inode. */
3008 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
3009                                        struct ocfs2_xattr_info *xi,
3010                                        struct ocfs2_xattr_search *xs)
3011 {
3012         struct ocfs2_xattr_entry *last;
3013         int free, i;
3014         size_t min_offs = xs->end - xs->base;
3015
3016         if (!xs->header)
3017                 return 0;
3018
3019         last = xs->header->xh_entries;
3020
3021         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3022                 size_t offs = le16_to_cpu(last->xe_name_offset);
3023                 if (offs < min_offs)
3024                         min_offs = offs;
3025                 last += 1;
3026         }
3027
3028         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3029         if (free < 0)
3030                 return 0;
3031
3032         BUG_ON(!xs->not_found);
3033
3034         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3035                 return 1;
3036
3037         return 0;
3038 }
3039
3040 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3041                                      struct ocfs2_dinode *di,
3042                                      struct ocfs2_xattr_info *xi,
3043                                      struct ocfs2_xattr_search *xis,
3044                                      struct ocfs2_xattr_search *xbs,
3045                                      int *clusters_need,
3046                                      int *meta_need,
3047                                      int *credits_need)
3048 {
3049         int ret = 0, old_in_xb = 0;
3050         int clusters_add = 0, meta_add = 0, credits = 0;
3051         struct buffer_head *bh = NULL;
3052         struct ocfs2_xattr_block *xb = NULL;
3053         struct ocfs2_xattr_entry *xe = NULL;
3054         struct ocfs2_xattr_value_root *xv = NULL;
3055         char *base = NULL;
3056         int name_offset, name_len = 0;
3057         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3058                                                     xi->xi_value_len);
3059         u64 value_size;
3060
3061         /*
3062          * Calculate the clusters we need to write.
3063          * No matter whether we replace an old one or add a new one,
3064          * we need this for writing.
3065          */
3066         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3067                 credits += new_clusters *
3068                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3069
3070         if (xis->not_found && xbs->not_found) {
3071                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3072
3073                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3074                         clusters_add += new_clusters;
3075                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3076                                                         &def_xv.xv.xr_list);
3077                 }
3078
3079                 goto meta_guess;
3080         }
3081
3082         if (!xis->not_found) {
3083                 xe = xis->here;
3084                 name_offset = le16_to_cpu(xe->xe_name_offset);
3085                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3086                 base = xis->base;
3087                 credits += OCFS2_INODE_UPDATE_CREDITS;
3088         } else {
3089                 int i, block_off = 0;
3090                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3091                 xe = xbs->here;
3092                 name_offset = le16_to_cpu(xe->xe_name_offset);
3093                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3094                 i = xbs->here - xbs->header->xh_entries;
3095                 old_in_xb = 1;
3096
3097                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3098                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3099                                                         bucket_xh(xbs->bucket),
3100                                                         i, &block_off,
3101                                                         &name_offset);
3102                         base = bucket_block(xbs->bucket, block_off);
3103                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3104                 } else {
3105                         base = xbs->base;
3106                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3107                 }
3108         }
3109
3110         /*
3111          * delete a xattr doesn't need metadata and cluster allocation.
3112          * so just calculate the credits and return.
3113          *
3114          * The credits for removing the value tree will be extended
3115          * by ocfs2_remove_extent itself.
3116          */
3117         if (!xi->xi_value) {
3118                 if (!ocfs2_xattr_is_local(xe))
3119                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3120
3121                 goto out;
3122         }
3123
3124         /* do cluster allocation guess first. */
3125         value_size = le64_to_cpu(xe->xe_value_size);
3126
3127         if (old_in_xb) {
3128                 /*
3129                  * In xattr set, we always try to set the xe in inode first,
3130                  * so if it can be inserted into inode successfully, the old
3131                  * one will be removed from the xattr block, and this xattr
3132                  * will be inserted into inode as a new xattr in inode.
3133                  */
3134                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3135                         clusters_add += new_clusters;
3136                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3137                                     OCFS2_INODE_UPDATE_CREDITS;
3138                         if (!ocfs2_xattr_is_local(xe))
3139                                 credits += ocfs2_calc_extend_credits(
3140                                                         inode->i_sb,
3141                                                         &def_xv.xv.xr_list);
3142                         goto out;
3143                 }
3144         }
3145
3146         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3147                 /* the new values will be stored outside. */
3148                 u32 old_clusters = 0;
3149
3150                 if (!ocfs2_xattr_is_local(xe)) {
3151                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3152                                                                  value_size);
3153                         xv = (struct ocfs2_xattr_value_root *)
3154                              (base + name_offset + name_len);
3155                         value_size = OCFS2_XATTR_ROOT_SIZE;
3156                 } else
3157                         xv = &def_xv.xv;
3158
3159                 if (old_clusters >= new_clusters) {
3160                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3161                         goto out;
3162                 } else {
3163                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3164                         clusters_add += new_clusters - old_clusters;
3165                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3166                                                              &xv->xr_list);
3167                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3168                                 goto out;
3169                 }
3170         } else {
3171                 /*
3172                  * Now the new value will be stored inside. So if the new
3173                  * value is smaller than the size of value root or the old
3174                  * value, we don't need any allocation, otherwise we have
3175                  * to guess metadata allocation.
3176                  */
3177                 if ((ocfs2_xattr_is_local(xe) &&
3178                      (value_size >= xi->xi_value_len)) ||
3179                     (!ocfs2_xattr_is_local(xe) &&
3180                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3181                         goto out;
3182         }
3183
3184 meta_guess:
3185         /* calculate metadata allocation. */
3186         if (di->i_xattr_loc) {
3187                 if (!xbs->xattr_bh) {
3188                         ret = ocfs2_read_xattr_block(inode,
3189                                                      le64_to_cpu(di->i_xattr_loc),
3190                                                      &bh);
3191                         if (ret) {
3192                                 mlog_errno(ret);
3193                                 goto out;
3194                         }
3195
3196                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3197                 } else
3198                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3199
3200                 /*
3201                  * If there is already an xattr tree, good, we can calculate
3202                  * like other b-trees. Otherwise we may have the chance of
3203                  * create a tree, the credit calculation is borrowed from
3204                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3205                  * new tree will be cluster based, so no meta is needed.
3206                  */
3207                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3208                         struct ocfs2_extent_list *el =
3209                                  &xb->xb_attrs.xb_root.xt_list;
3210                         meta_add += ocfs2_extend_meta_needed(el);
3211                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3212                                                              el);
3213                 } else
3214                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3215
3216                 /*
3217                  * This cluster will be used either for new bucket or for
3218                  * new xattr block.
3219                  * If the cluster size is the same as the bucket size, one
3220                  * more is needed since we may need to extend the bucket
3221                  * also.
3222                  */
3223                 clusters_add += 1;
3224                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3225                 if (OCFS2_XATTR_BUCKET_SIZE ==
3226                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3227                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3228                         clusters_add += 1;
3229                 }
3230         } else {
3231                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3232                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3233                         struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3234                         meta_add += ocfs2_extend_meta_needed(el);
3235                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3236                                                              el);
3237                 } else {
3238                         meta_add += 1;
3239                 }
3240         }
3241 out:
3242         if (clusters_need)
3243                 *clusters_need = clusters_add;
3244         if (meta_need)
3245                 *meta_need = meta_add;
3246         if (credits_need)
3247                 *credits_need = credits;
3248         brelse(bh);
3249         return ret;
3250 }
3251
3252 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3253                                      struct ocfs2_dinode *di,
3254                                      struct ocfs2_xattr_info *xi,
3255                                      struct ocfs2_xattr_search *xis,
3256                                      struct ocfs2_xattr_search *xbs,
3257                                      struct ocfs2_xattr_set_ctxt *ctxt,
3258                                      int extra_meta,
3259                                      int *credits)
3260 {
3261         int clusters_add, meta_add, ret;
3262         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3263
3264         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3265
3266         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3267
3268         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3269                                         &clusters_add, &meta_add, credits);
3270         if (ret) {
3271                 mlog_errno(ret);
3272                 return ret;
3273         }
3274
3275         meta_add += extra_meta;
3276         trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3277                                         clusters_add, *credits);
3278
3279         if (meta_add) {
3280                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3281                                                         &ctxt->meta_ac);
3282                 if (ret) {
3283                         mlog_errno(ret);
3284                         goto out;
3285                 }
3286         }
3287
3288         if (clusters_add) {
3289                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3290                 if (ret)
3291                         mlog_errno(ret);
3292         }
3293 out:
3294         if (ret) {
3295                 if (ctxt->meta_ac) {
3296                         ocfs2_free_alloc_context(ctxt->meta_ac);
3297                         ctxt->meta_ac = NULL;
3298                 }
3299
3300                 /*
3301                  * We cannot have an error and a non null ctxt->data_ac.
3302                  */
3303         }
3304
3305         return ret;
3306 }
3307
3308 static int __ocfs2_xattr_set_handle(struct inode *inode,
3309                                     struct ocfs2_dinode *di,
3310                                     struct ocfs2_xattr_info *xi,
3311                                     struct ocfs2_xattr_search *xis,
3312                                     struct ocfs2_xattr_search *xbs,
3313                                     struct ocfs2_xattr_set_ctxt *ctxt)
3314 {
3315         int ret = 0, credits, old_found;
3316
3317         if (!xi->xi_value) {
3318                 /* Remove existing extended attribute */
3319                 if (!xis->not_found)
3320                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3321                 else if (!xbs->not_found)
3322                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3323         } else {
3324                 /* We always try to set extended attribute into inode first*/
3325                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3326                 if (!ret && !xbs->not_found) {
3327                         /*
3328                          * If succeed and that extended attribute existing in
3329                          * external block, then we will remove it.
3330                          */
3331                         xi->xi_value = NULL;
3332                         xi->xi_value_len = 0;
3333
3334                         old_found = xis->not_found;
3335                         xis->not_found = -ENODATA;
3336                         ret = ocfs2_calc_xattr_set_need(inode,
3337                                                         di,
3338                                                         xi,
3339                                                         xis,
3340                                                         xbs,
3341                                                         NULL,
3342                                                         NULL,
3343                                                         &credits);
3344                         xis->not_found = old_found;
3345                         if (ret) {
3346                                 mlog_errno(ret);
3347                                 goto out;
3348                         }
3349
3350                         ret = ocfs2_extend_trans(ctxt->handle, credits);
3351                         if (ret) {
3352                                 mlog_errno(ret);
3353                                 goto out;
3354                         }
3355                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3356                 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3357                         if (di->i_xattr_loc && !xbs->xattr_bh) {
3358                                 ret = ocfs2_xattr_block_find(inode,
3359                                                              xi->xi_name_index,
3360                                                              xi->xi_name, xbs);
3361                                 if (ret)
3362                                         goto out;
3363
3364                                 old_found = xis->not_found;
3365                                 xis->not_found = -ENODATA;
3366                                 ret = ocfs2_calc_xattr_set_need(inode,
3367                                                                 di,
3368                                                                 xi,
3369                                                                 xis,
3370                                                                 xbs,
3371                                                                 NULL,
3372                                                                 NULL,
3373                                                                 &credits);
3374                                 xis->not_found = old_found;
3375                                 if (ret) {
3376                                         mlog_errno(ret);
3377                                         goto out;
3378                                 }
3379
3380                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3381                                 if (ret) {
3382                                         mlog_errno(ret);
3383                                         goto out;
3384                                 }
3385                         }
3386                         /*
3387                          * If no space in inode, we will set extended attribute
3388                          * into external block.
3389                          */
3390                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3391                         if (ret)
3392                                 goto out;
3393                         if (!xis->not_found) {
3394                                 /*
3395                                  * If succeed and that extended attribute
3396                                  * existing in inode, we will remove it.
3397                                  */
3398                                 xi->xi_value = NULL;
3399                                 xi->xi_value_len = 0;
3400                                 xbs->not_found = -ENODATA;
3401                                 ret = ocfs2_calc_xattr_set_need(inode,
3402                                                                 di,
3403                                                                 xi,
3404                                                                 xis,
3405                                                                 xbs,
3406                                                                 NULL,
3407                                                                 NULL,
3408                                                                 &credits);
3409                                 if (ret) {
3410                                         mlog_errno(ret);
3411                                         goto out;
3412                                 }
3413
3414                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3415                                 if (ret) {
3416                                         mlog_errno(ret);
3417                                         goto out;
3418                                 }
3419                                 ret = ocfs2_xattr_ibody_set(inode, xi,
3420                                                             xis, ctxt);
3421                         }
3422                 }
3423         }
3424
3425         if (!ret) {
3426                 /* Update inode ctime. */
3427                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3428                                               xis->inode_bh,
3429                                               OCFS2_JOURNAL_ACCESS_WRITE);
3430                 if (ret) {
3431                         mlog_errno(ret);
3432                         goto out;
3433                 }
3434
3435                 inode->i_ctime = current_time(inode);
3436                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3437                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3438                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3439         }
3440 out:
3441         return ret;
3442 }
3443
3444 /*
3445  * This function only called duing creating inode
3446  * for init security/acl xattrs of the new inode.
3447  * All transanction credits have been reserved in mknod.
3448  */
3449 int ocfs2_xattr_set_handle(handle_t *handle,
3450                            struct inode *inode,
3451                            struct buffer_head *di_bh,
3452                            int name_index,
3453                            const char *name,
3454                            const void *value,
3455                            size_t value_len,
3456                            int flags,
3457                            struct ocfs2_alloc_context *meta_ac,
3458                            struct ocfs2_alloc_context *data_ac)
3459 {
3460         struct ocfs2_dinode *di;
3461         int ret;
3462
3463         struct ocfs2_xattr_info xi = {
3464                 .xi_name_index = name_index,
3465                 .xi_name = name,
3466                 .xi_name_len = strlen(name),
3467                 .xi_value = value,
3468                 .xi_value_len = value_len,
3469         };
3470
3471         struct ocfs2_xattr_search xis = {
3472                 .not_found = -ENODATA,
3473         };
3474
3475         struct ocfs2_xattr_search xbs = {
3476                 .not_found = -ENODATA,
3477         };
3478
3479         struct ocfs2_xattr_set_ctxt ctxt = {
3480                 .handle = handle,
3481                 .meta_ac = meta_ac,
3482                 .data_ac = data_ac,
3483         };
3484
3485         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3486                 return -EOPNOTSUPP;
3487
3488         /*
3489          * In extreme situation, may need xattr bucket when
3490          * block size is too small. And we have already reserved
3491          * the credits for bucket in mknod.
3492          */
3493         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3494                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3495                 if (!xbs.bucket) {
3496                         mlog_errno(-ENOMEM);
3497                         return -ENOMEM;
3498                 }
3499         }
3500
3501         xis.inode_bh = xbs.inode_bh = di_bh;
3502         di = (struct ocfs2_dinode *)di_bh->b_data;
3503
3504         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3505
3506         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3507         if (ret)
3508                 goto cleanup;
3509         if (xis.not_found) {
3510                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3511                 if (ret)
3512                         goto cleanup;
3513         }
3514
3515         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3516
3517 cleanup:
3518         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3519         brelse(xbs.xattr_bh);
3520         ocfs2_xattr_bucket_free(xbs.bucket);
3521
3522         return ret;
3523 }
3524
3525 /*
3526  * ocfs2_xattr_set()
3527  *
3528  * Set, replace or remove an extended attribute for this inode.
3529  * value is NULL to remove an existing extended attribute, else either
3530  * create or replace an extended attribute.
3531  */
3532 int ocfs2_xattr_set(struct inode *inode,
3533                     int name_index,
3534                     const char *name,
3535                     const void *value,
3536                     size_t value_len,
3537                     int flags)
3538 {
3539         struct buffer_head *di_bh = NULL;
3540         struct ocfs2_dinode *di;
3541         int ret, credits, had_lock, ref_meta = 0, ref_credits = 0;
3542         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3543         struct inode *tl_inode = osb->osb_tl_inode;
3544         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3545         struct ocfs2_refcount_tree *ref_tree = NULL;
3546         struct ocfs2_lock_holder oh;
3547
3548         struct ocfs2_xattr_info xi = {
3549                 .xi_name_index = name_index,
3550                 .xi_name = name,
3551                 .xi_name_len = strlen(name),
3552                 .xi_value = value,
3553                 .xi_value_len = value_len,
3554         };
3555
3556         struct ocfs2_xattr_search xis = {
3557                 .not_found = -ENODATA,
3558         };
3559
3560         struct ocfs2_xattr_search xbs = {
3561                 .not_found = -ENODATA,
3562         };
3563
3564         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3565                 return -EOPNOTSUPP;
3566
3567         /*
3568          * Only xbs will be used on indexed trees.  xis doesn't need a
3569          * bucket.
3570          */
3571         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3572         if (!xbs.bucket) {
3573                 mlog_errno(-ENOMEM);
3574                 return -ENOMEM;
3575         }
3576
3577         had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 1, &oh);
3578         if (had_lock < 0) {
3579                 ret = had_lock;
3580                 mlog_errno(ret);
3581                 goto cleanup_nolock;
3582         }
3583         xis.inode_bh = xbs.inode_bh = di_bh;
3584         di = (struct ocfs2_dinode *)di_bh->b_data;
3585
3586         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3587         /*
3588          * Scan inode and external block to find the same name
3589          * extended attribute and collect search information.
3590          */
3591         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3592         if (ret)
3593                 goto cleanup;
3594         if (xis.not_found) {
3595                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3596                 if (ret)
3597                         goto cleanup;
3598         }
3599
3600         if (xis.not_found && xbs.not_found) {
3601                 ret = -ENODATA;
3602                 if (flags & XATTR_REPLACE)
3603                         goto cleanup;
3604                 ret = 0;
3605                 if (!value)
3606                         goto cleanup;
3607         } else {
3608                 ret = -EEXIST;
3609                 if (flags & XATTR_CREATE)
3610                         goto cleanup;
3611         }
3612
3613         /* Check whether the value is refcounted and do some preparation. */
3614         if (ocfs2_is_refcount_inode(inode) &&
3615             (!xis.not_found || !xbs.not_found)) {
3616                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3617                                                    &xis, &xbs, &ref_tree,
3618                                                    &ref_meta, &ref_credits);
3619                 if (ret) {
3620                         mlog_errno(ret);
3621                         goto cleanup;
3622                 }
3623         }
3624
3625         inode_lock(tl_inode);
3626
3627         if (ocfs2_truncate_log_needs_flush(osb)) {
3628                 ret = __ocfs2_flush_truncate_log(osb);
3629                 if (ret < 0) {
3630                         inode_unlock(tl_inode);
3631                         mlog_errno(ret);
3632                         goto cleanup;
3633                 }
3634         }
3635         inode_unlock(tl_inode);
3636
3637         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3638                                         &xbs, &ctxt, ref_meta, &credits);
3639         if (ret) {
3640                 mlog_errno(ret);
3641                 goto cleanup;
3642         }
3643
3644         /* we need to update inode's ctime field, so add credit for it. */
3645         credits += OCFS2_INODE_UPDATE_CREDITS;
3646         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3647         if (IS_ERR(ctxt.handle)) {
3648                 ret = PTR_ERR(ctxt.handle);
3649                 mlog_errno(ret);
3650                 goto out_free_ac;
3651         }
3652
3653         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3654         ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3655
3656         ocfs2_commit_trans(osb, ctxt.handle);
3657
3658 out_free_ac:
3659         if (ctxt.data_ac)
3660                 ocfs2_free_alloc_context(ctxt.data_ac);
3661         if (ctxt.meta_ac)
3662                 ocfs2_free_alloc_context(ctxt.meta_ac);
3663         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3664                 ocfs2_schedule_truncate_log_flush(osb, 1);
3665         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3666
3667 cleanup:
3668         if (ref_tree)
3669                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3670         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3671         if (!value && !ret) {
3672                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3673                 if (ret)
3674                         mlog_errno(ret);
3675         }
3676         ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
3677 cleanup_nolock:
3678         brelse(di_bh);
3679         brelse(xbs.xattr_bh);
3680         ocfs2_xattr_bucket_free(xbs.bucket);
3681
3682         return ret;
3683 }
3684
3685 /*
3686  * Find the xattr extent rec which may contains name_hash.
3687  * e_cpos will be the first name hash of the xattr rec.
3688  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3689  */
3690 static int ocfs2_xattr_get_rec(struct inode *inode,
3691                                u32 name_hash,
3692                                u64 *p_blkno,
3693                                u32 *e_cpos,
3694                                u32 *num_clusters,
3695                                struct ocfs2_extent_list *el)
3696 {
3697         int ret = 0, i;
3698         struct buffer_head *eb_bh = NULL;
3699         struct ocfs2_extent_block *eb;
3700         struct ocfs2_extent_rec *rec = NULL;
3701         u64 e_blkno = 0;
3702
3703         if (el->l_tree_depth) {
3704                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3705                                       &eb_bh);
3706                 if (ret) {
3707                         mlog_errno(ret);
3708                         goto out;
3709                 }
3710
3711                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3712                 el = &eb->h_list;
3713
3714                 if (el->l_tree_depth) {
3715                         ret = ocfs2_error(inode->i_sb,
3716                                           "Inode %lu has non zero tree depth in xattr tree block %llu\n",
3717                                           inode->i_ino,
3718                                           (unsigned long long)eb_bh->b_blocknr);
3719                         goto out;
3720                 }
3721         }
3722
3723         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3724                 rec = &el->l_recs[i];
3725
3726                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3727                         e_blkno = le64_to_cpu(rec->e_blkno);
3728                         break;
3729                 }
3730         }
3731
3732         if (!e_blkno) {
3733                 ret = ocfs2_error(inode->i_sb, "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
3734                                   inode->i_ino,
3735                                   le32_to_cpu(rec->e_cpos),
3736                                   ocfs2_rec_clusters(el, rec));
3737                 goto out;
3738         }
3739
3740         *p_blkno = le64_to_cpu(rec->e_blkno);
3741         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3742         if (e_cpos)
3743                 *e_cpos = le32_to_cpu(rec->e_cpos);
3744 out:
3745         brelse(eb_bh);
3746         return ret;
3747 }
3748
3749 typedef int (xattr_bucket_func)(struct inode *inode,
3750                                 struct ocfs2_xattr_bucket *bucket,
3751                                 void *para);
3752
3753 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3754                                    struct ocfs2_xattr_bucket *bucket,
3755                                    int name_index,
3756                                    const char *name,
3757                                    u32 name_hash,
3758                                    u16 *xe_index,
3759                                    int *found)
3760 {
3761         int i, ret = 0, cmp = 1, block_off, new_offset;
3762         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3763         size_t name_len = strlen(name);
3764         struct ocfs2_xattr_entry *xe = NULL;
3765         char *xe_name;
3766
3767         /*
3768          * We don't use binary search in the bucket because there
3769          * may be multiple entries with the same name hash.
3770          */
3771         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3772                 xe = &xh->xh_entries[i];
3773
3774                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3775                         continue;
3776                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3777                         break;
3778
3779                 cmp = name_index - ocfs2_xattr_get_type(xe);
3780                 if (!cmp)
3781                         cmp = name_len - xe->xe_name_len;
3782                 if (cmp)
3783                         continue;
3784
3785                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3786                                                         xh,
3787                                                         i,
3788                                                         &block_off,
3789                                                         &new_offset);
3790                 if (ret) {
3791                         mlog_errno(ret);
3792                         break;
3793                 }
3794
3795
3796                 xe_name = bucket_block(bucket, block_off) + new_offset;
3797                 if (!memcmp(name, xe_name, name_len)) {
3798                         *xe_index = i;
3799                         *found = 1;
3800                         ret = 0;
3801                         break;
3802                 }
3803         }
3804
3805         return ret;
3806 }
3807
3808 /*
3809  * Find the specified xattr entry in a series of buckets.
3810  * This series start from p_blkno and last for num_clusters.
3811  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3812  * the num of the valid buckets.
3813  *
3814  * Return the buffer_head this xattr should reside in. And if the xattr's
3815  * hash is in the gap of 2 buckets, return the lower bucket.
3816  */
3817 static int ocfs2_xattr_bucket_find(struct inode *inode,
3818                                    int name_index,
3819                                    const char *name,
3820                                    u32 name_hash,
3821                                    u64 p_blkno,
3822                                    u32 first_hash,
3823                                    u32 num_clusters,
3824                                    struct ocfs2_xattr_search *xs)
3825 {
3826         int ret, found = 0;
3827         struct ocfs2_xattr_header *xh = NULL;
3828         struct ocfs2_xattr_entry *xe = NULL;
3829         u16 index = 0;
3830         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3831         int low_bucket = 0, bucket, high_bucket;
3832         struct ocfs2_xattr_bucket *search;
3833         u32 last_hash;
3834         u64 blkno, lower_blkno = 0;
3835
3836         search = ocfs2_xattr_bucket_new(inode);
3837         if (!search) {
3838                 ret = -ENOMEM;
3839                 mlog_errno(ret);
3840                 goto out;
3841         }
3842
3843         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3844         if (ret) {
3845                 mlog_errno(ret);
3846                 goto out;
3847         }
3848
3849         xh = bucket_xh(search);
3850         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3851         while (low_bucket <= high_bucket) {
3852                 ocfs2_xattr_bucket_relse(search);
3853
3854                 bucket = (low_bucket + high_bucket) / 2;
3855                 blkno = p_blkno + bucket * blk_per_bucket;
3856                 ret = ocfs2_read_xattr_bucket(search, blkno);
3857                 if (ret) {
3858                         mlog_errno(ret);
3859                         goto out;
3860                 }
3861
3862                 xh = bucket_xh(search);
3863                 xe = &xh->xh_entries[0];
3864                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3865                         high_bucket = bucket - 1;
3866                         continue;
3867                 }
3868
3869                 /*
3870                  * Check whether the hash of the last entry in our
3871                  * bucket is larger than the search one. for an empty
3872                  * bucket, the last one is also the first one.
3873                  */
3874                 if (xh->xh_count)
3875                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3876
3877                 last_hash = le32_to_cpu(xe->xe_name_hash);
3878
3879                 /* record lower_blkno which may be the insert place. */
3880                 lower_blkno = blkno;
3881
3882                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3883                         low_bucket = bucket + 1;
3884                         continue;
3885                 }
3886
3887                 /* the searched xattr should reside in this bucket if exists. */
3888                 ret = ocfs2_find_xe_in_bucket(inode, search,
3889                                               name_index, name, name_hash,
3890                                               &index, &found);
3891                 if (ret) {
3892                         mlog_errno(ret);
3893                         goto out;
3894                 }
3895                 break;
3896         }
3897
3898         /*
3899          * Record the bucket we have found.
3900          * When the xattr's hash value is in the gap of 2 buckets, we will
3901          * always set it to the previous bucket.
3902          */
3903         if (!lower_blkno)
3904                 lower_blkno = p_blkno;
3905
3906         /* This should be in cache - we just read it during the search */
3907         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3908         if (ret) {
3909                 mlog_errno(ret);
3910                 goto out;
3911         }
3912
3913         xs->header = bucket_xh(xs->bucket);
3914         xs->base = bucket_block(xs->bucket, 0);
3915         xs->end = xs->base + inode->i_sb->s_blocksize;
3916
3917         if (found) {
3918                 xs->here = &xs->header->xh_entries[index];
3919                 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3920                         name, name_index, name_hash,
3921                         (unsigned long long)bucket_blkno(xs->bucket),
3922                         index);
3923         } else
3924                 ret = -ENODATA;
3925
3926 out:
3927         ocfs2_xattr_bucket_free(search);
3928         return ret;
3929 }
3930
3931 static int ocfs2_xattr_index_block_find(struct inode *inode,
3932                                         struct buffer_head *root_bh,
3933                                         int name_index,
3934                                         const char *name,
3935                                         struct ocfs2_xattr_search *xs)
3936 {
3937         int ret;
3938         struct ocfs2_xattr_block *xb =
3939                         (struct ocfs2_xattr_block *)root_bh->b_data;
3940         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3941         struct ocfs2_extent_list *el = &xb_root->xt_list;
3942         u64 p_blkno = 0;
3943         u32 first_hash, num_clusters = 0;
3944         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3945
3946         if (le16_to_cpu(el->l_next_free_rec) == 0)
3947                 return -ENODATA;
3948
3949         trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3950                                         name, name_index, name_hash,
3951                                         (unsigned long long)root_bh->b_blocknr,
3952                                         -1);
3953
3954         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3955                                   &num_clusters, el);
3956         if (ret) {
3957                 mlog_errno(ret);
3958                 goto out;
3959         }
3960
3961         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3962
3963         trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3964                                         name, name_index, first_hash,
3965                                         (unsigned long long)p_blkno,
3966                                         num_clusters);
3967
3968         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3969                                       p_blkno, first_hash, num_clusters, xs);
3970
3971 out:
3972         return ret;
3973 }
3974
3975 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3976                                        u64 blkno,
3977                                        u32 clusters,
3978                                        xattr_bucket_func *func,
3979                                        void *para)
3980 {
3981         int i, ret = 0;
3982         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3983         u32 num_buckets = clusters * bpc;
3984         struct ocfs2_xattr_bucket *bucket;
3985
3986         bucket = ocfs2_xattr_bucket_new(inode);
3987         if (!bucket) {
3988                 mlog_errno(-ENOMEM);
3989                 return -ENOMEM;
3990         }
3991
3992         trace_ocfs2_iterate_xattr_buckets(
3993                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3994                 (unsigned long long)blkno, clusters);
3995
3996         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3997                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3998                 if (ret) {
3999                         mlog_errno(ret);
4000                         break;
4001                 }
4002
4003                 /*
4004                  * The real bucket num in this series of blocks is stored
4005                  * in the 1st bucket.
4006                  */
4007                 if (i == 0)
4008                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
4009
4010                 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
4011                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
4012                 if (func) {
4013                         ret = func(inode, bucket, para);
4014                         if (ret && ret != -ERANGE)
4015                                 mlog_errno(ret);
4016                         /* Fall through to bucket_relse() */
4017                 }
4018
4019                 ocfs2_xattr_bucket_relse(bucket);
4020                 if (ret)
4021                         break;
4022         }
4023
4024         ocfs2_xattr_bucket_free(bucket);
4025         return ret;
4026 }
4027
4028 struct ocfs2_xattr_tree_list {
4029         char *buffer;
4030         size_t buffer_size;
4031         size_t result;
4032 };
4033
4034 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4035                                              struct ocfs2_xattr_header *xh,
4036                                              int index,
4037                                              int *block_off,
4038                                              int *new_offset)
4039 {
4040         u16 name_offset;
4041
4042         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4043                 return -EINVAL;
4044
4045         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4046
4047         *block_off = name_offset >> sb->s_blocksize_bits;
4048         *new_offset = name_offset % sb->s_blocksize;
4049
4050         return 0;
4051 }
4052
4053 static int ocfs2_list_xattr_bucket(struct inode *inode,
4054                                    struct ocfs2_xattr_bucket *bucket,
4055                                    void *para)
4056 {
4057         int ret = 0, type;
4058         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4059         int i, block_off, new_offset;
4060         const char *name;
4061
4062         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4063                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4064                 type = ocfs2_xattr_get_type(entry);
4065
4066                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4067                                                         bucket_xh(bucket),
4068                                                         i,
4069                                                         &block_off,
4070                                                         &new_offset);
4071                 if (ret)
4072                         break;
4073
4074                 name = (const char *)bucket_block(bucket, block_off) +
4075                         new_offset;
4076                 ret = ocfs2_xattr_list_entry(inode->i_sb,
4077                                              xl->buffer,
4078                                              xl->buffer_size,
4079                                              &xl->result,
4080                                              type, name,
4081                                              entry->xe_name_len);
4082                 if (ret)
4083                         break;
4084         }
4085
4086         return ret;
4087 }
4088
4089 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4090                                            struct buffer_head *blk_bh,
4091                                            xattr_tree_rec_func *rec_func,
4092                                            void *para)
4093 {
4094         struct ocfs2_xattr_block *xb =
4095                         (struct ocfs2_xattr_block *)blk_bh->b_data;
4096         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4097         int ret = 0;
4098         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4099         u64 p_blkno = 0;
4100
4101         if (!el->l_next_free_rec || !rec_func)
4102                 return 0;
4103
4104         while (name_hash > 0) {
4105                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4106                                           &e_cpos, &num_clusters, el);
4107                 if (ret) {
4108                         mlog_errno(ret);
4109                         break;
4110                 }
4111
4112                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4113                                num_clusters, para);
4114                 if (ret) {
4115                         if (ret != -ERANGE)
4116                                 mlog_errno(ret);
4117                         break;
4118                 }
4119
4120                 if (e_cpos == 0)
4121                         break;
4122
4123                 name_hash = e_cpos - 1;
4124         }
4125
4126         return ret;
4127
4128 }
4129
4130 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4131                                      struct buffer_head *root_bh,
4132                                      u64 blkno, u32 cpos, u32 len, void *para)
4133 {
4134         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4135                                            ocfs2_list_xattr_bucket, para);
4136 }
4137
4138 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4139                                              struct buffer_head *blk_bh,
4140                                              char *buffer,
4141                                              size_t buffer_size)
4142 {
4143         int ret;
4144         struct ocfs2_xattr_tree_list xl = {
4145                 .buffer = buffer,
4146                 .buffer_size = buffer_size,
4147                 .result = 0,
4148         };
4149
4150         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4151                                               ocfs2_list_xattr_tree_rec, &xl);
4152         if (ret) {
4153                 mlog_errno(ret);
4154                 goto out;
4155         }
4156
4157         ret = xl.result;
4158 out:
4159         return ret;
4160 }
4161
4162 static int cmp_xe(const void *a, const void *b)
4163 {
4164         const struct ocfs2_xattr_entry *l = a, *r = b;
4165         u32 l_hash = le32_to_cpu(l->xe_name_hash);
4166         u32 r_hash = le32_to_cpu(r->xe_name_hash);
4167
4168         if (l_hash > r_hash)
4169                 return 1;
4170         if (l_hash < r_hash)
4171                 return -1;
4172         return 0;
4173 }
4174
4175 static void swap_xe(void *a, void *b, int size)
4176 {
4177         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4178
4179         tmp = *l;
4180         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4181         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4182 }
4183
4184 /*
4185  * When the ocfs2_xattr_block is filled up, new bucket will be created
4186  * and all the xattr entries will be moved to the new bucket.
4187  * The header goes at the start of the bucket, and the names+values are
4188  * filled from the end.  This is why *target starts as the last buffer.
4189  * Note: we need to sort the entries since they are not saved in order
4190  * in the ocfs2_xattr_block.
4191  */
4192 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4193                                            struct buffer_head *xb_bh,
4194                                            struct ocfs2_xattr_bucket *bucket)
4195 {
4196         int i, blocksize = inode->i_sb->s_blocksize;
4197         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4198         u16 offset, size, off_change;
4199         struct ocfs2_xattr_entry *xe;
4200         struct ocfs2_xattr_block *xb =
4201                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
4202         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4203         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4204         u16 count = le16_to_cpu(xb_xh->xh_count);
4205         char *src = xb_bh->b_data;
4206         char *target = bucket_block(bucket, blks - 1);
4207
4208         trace_ocfs2_cp_xattr_block_to_bucket_begin(
4209                                 (unsigned long long)xb_bh->b_blocknr,
4210                                 (unsigned long long)bucket_blkno(bucket));
4211
4212         for (i = 0; i < blks; i++)
4213                 memset(bucket_block(bucket, i), 0, blocksize);
4214
4215         /*
4216          * Since the xe_name_offset is based on ocfs2_xattr_header,
4217          * there is a offset change corresponding to the change of
4218          * ocfs2_xattr_header's position.
4219          */
4220         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4221         xe = &xb_xh->xh_entries[count - 1];
4222         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4223         size = blocksize - offset;
4224
4225         /* copy all the names and values. */
4226         memcpy(target + offset, src + offset, size);
4227
4228         /* Init new header now. */
4229         xh->xh_count = xb_xh->xh_count;
4230         xh->xh_num_buckets = cpu_to_le16(1);
4231         xh->xh_name_value_len = cpu_to_le16(size);
4232         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4233
4234         /* copy all the entries. */
4235         target = bucket_block(bucket, 0);
4236         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4237         size = count * sizeof(struct ocfs2_xattr_entry);
4238         memcpy(target + offset, (char *)xb_xh + offset, size);
4239
4240         /* Change the xe offset for all the xe because of the move. */
4241         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4242                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4243         for (i = 0; i < count; i++)
4244                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4245
4246         trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4247
4248         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4249              cmp_xe, swap_xe);
4250 }
4251
4252 /*
4253  * After we move xattr from block to index btree, we have to
4254  * update ocfs2_xattr_search to the new xe and base.
4255  *
4256  * When the entry is in xattr block, xattr_bh indicates the storage place.
4257  * While if the entry is in index b-tree, "bucket" indicates the
4258  * real place of the xattr.
4259  */
4260 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4261                                             struct ocfs2_xattr_search *xs,
4262                                             struct buffer_head *old_bh)
4263 {
4264         char *buf = old_bh->b_data;
4265         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4266         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4267         int i;
4268
4269         xs->header = bucket_xh(xs->bucket);
4270         xs->base = bucket_block(xs->bucket, 0);
4271         xs->end = xs->base + inode->i_sb->s_blocksize;
4272
4273         if (xs->not_found)
4274                 return;
4275
4276         i = xs->here - old_xh->xh_entries;
4277         xs->here = &xs->header->xh_entries[i];
4278 }
4279
4280 static int ocfs2_xattr_create_index_block(struct inode *inode,
4281                                           struct ocfs2_xattr_search *xs,
4282                                           struct ocfs2_xattr_set_ctxt *ctxt)
4283 {
4284         int ret;
4285         u32 bit_off, len;
4286         u64 blkno;
4287         handle_t *handle = ctxt->handle;
4288         struct ocfs2_inode_info *oi = OCFS2_I(inode);
4289         struct buffer_head *xb_bh = xs->xattr_bh;
4290         struct ocfs2_xattr_block *xb =
4291                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4292         struct ocfs2_xattr_tree_root *xr;
4293         u16 xb_flags = le16_to_cpu(xb->xb_flags);
4294
4295         trace_ocfs2_xattr_create_index_block_begin(
4296                                 (unsigned long long)xb_bh->b_blocknr);
4297
4298         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4299         BUG_ON(!xs->bucket);
4300
4301         /*
4302          * XXX:
4303          * We can use this lock for now, and maybe move to a dedicated mutex
4304          * if performance becomes a problem later.
4305          */
4306         down_write(&oi->ip_alloc_sem);
4307
4308         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4309                                       OCFS2_JOURNAL_ACCESS_WRITE);
4310         if (ret) {
4311                 mlog_errno(ret);
4312                 goto out;
4313         }
4314
4315         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4316                                      1, 1, &bit_off, &len);
4317         if (ret) {
4318                 mlog_errno(ret);
4319                 goto out;
4320         }
4321
4322         /*
4323          * The bucket may spread in many blocks, and
4324          * we will only touch the 1st block and the last block
4325          * in the whole bucket(one for entry and one for data).
4326          */
4327         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4328
4329         trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4330
4331         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4332         if (ret) {
4333                 mlog_errno(ret);
4334                 goto out;
4335         }
4336
4337         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4338                                                 OCFS2_JOURNAL_ACCESS_CREATE);
4339         if (ret) {
4340                 mlog_errno(ret);
4341                 goto out;
4342         }
4343
4344         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4345         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4346
4347         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4348
4349         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4350         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4351                offsetof(struct ocfs2_xattr_block, xb_attrs));
4352
4353         xr = &xb->xb_attrs.xb_root;
4354         xr->xt_clusters = cpu_to_le32(1);
4355         xr->xt_last_eb_blk = 0;
4356         xr->xt_list.l_tree_depth = 0;
4357         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4358         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4359
4360         xr->xt_list.l_recs[0].e_cpos = 0;
4361         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4362         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4363
4364         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4365
4366         ocfs2_journal_dirty(handle, xb_bh);
4367
4368 out:
4369         up_write(&oi->ip_alloc_sem);
4370
4371         return ret;
4372 }
4373
4374 static int cmp_xe_offset(const void *a, const void *b)
4375 {
4376         const struct ocfs2_xattr_entry *l = a, *r = b;
4377         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4378         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4379
4380         if (l_name_offset < r_name_offset)
4381                 return 1;
4382         if (l_name_offset > r_name_offset)
4383                 return -1;
4384         return 0;
4385 }
4386
4387 /*
4388  * defrag a xattr bucket if we find that the bucket has some
4389  * holes beteen name/value pairs.
4390  * We will move all the name/value pairs to the end of the bucket
4391  * so that we can spare some space for insertion.
4392  */
4393 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4394                                      handle_t *handle,
4395                                      struct ocfs2_xattr_bucket *bucket)
4396 {
4397         int ret, i;
4398         size_t end, offset, len;
4399         struct ocfs2_xattr_header *xh;
4400         char *entries, *buf, *bucket_buf = NULL;
4401         u64 blkno = bucket_blkno(bucket);
4402         u16 xh_free_start;
4403         size_t blocksize = inode->i_sb->s_blocksize;
4404         struct ocfs2_xattr_entry *xe;
4405
4406         /*
4407          * In order to make the operation more efficient and generic,
4408          * we copy all the blocks into a contiguous memory and do the
4409          * defragment there, so if anything is error, we will not touch
4410          * the real block.
4411          */
4412         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4413         if (!bucket_buf) {
4414                 ret = -EIO;
4415                 goto out;
4416         }
4417
4418         buf = bucket_buf;
4419         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4420                 memcpy(buf, bucket_block(bucket, i), blocksize);
4421
4422         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4423                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4424         if (ret < 0) {
4425                 mlog_errno(ret);
4426                 goto out;
4427         }
4428
4429         xh = (struct ocfs2_xattr_header *)bucket_buf;
4430         entries = (char *)xh->xh_entries;
4431         xh_free_start = le16_to_cpu(xh->xh_free_start);
4432
4433         trace_ocfs2_defrag_xattr_bucket(
4434              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4435              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4436
4437         /*
4438          * sort all the entries by their offset.
4439          * the largest will be the first, so that we can
4440          * move them to the end one by one.
4441          */
4442         sort(entries, le16_to_cpu(xh->xh_count),
4443              sizeof(struct ocfs2_xattr_entry),
4444              cmp_xe_offset, swap_xe);
4445
4446         /* Move all name/values to the end of the bucket. */
4447         xe = xh->xh_entries;
4448         end = OCFS2_XATTR_BUCKET_SIZE;
4449         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4450                 offset = le16_to_cpu(xe->xe_name_offset);
4451                 len = namevalue_size_xe(xe);
4452
4453                 /*
4454                  * We must make sure that the name/value pair
4455                  * exist in the same block. So adjust end to
4456                  * the previous block end if needed.
4457                  */
4458                 if (((end - len) / blocksize !=
4459                         (end - 1) / blocksize))
4460                         end = end - end % blocksize;
4461
4462                 if (end > offset + len) {
4463                         memmove(bucket_buf + end - len,
4464                                 bucket_buf + offset, len);
4465                         xe->xe_name_offset = cpu_to_le16(end - len);
4466                 }
4467
4468                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4469                                 "bucket %llu\n", (unsigned long long)blkno);
4470
4471                 end -= len;
4472         }
4473
4474         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4475                         "bucket %llu\n", (unsigned long long)blkno);
4476
4477         if (xh_free_start == end)
4478                 goto out;
4479
4480         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4481         xh->xh_free_start = cpu_to_le16(end);
4482
4483         /* sort the entries by their name_hash. */
4484         sort(entries, le16_to_cpu(xh->xh_count),
4485              sizeof(struct ocfs2_xattr_entry),
4486              cmp_xe, swap_xe);
4487
4488         buf = bucket_buf;
4489         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4490                 memcpy(bucket_block(bucket, i), buf, blocksize);
4491         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4492
4493 out:
4494         kfree(bucket_buf);
4495         return ret;
4496 }
4497
4498 /*
4499  * prev_blkno points to the start of an existing extent.  new_blkno
4500  * points to a newly allocated extent.  Because we know each of our
4501  * clusters contains more than bucket, we can easily split one cluster
4502  * at a bucket boundary.  So we take the last cluster of the existing
4503  * extent and split it down the middle.  We move the last half of the
4504  * buckets in the last cluster of the existing extent over to the new
4505  * extent.
4506  *
4507  * first_bh is the buffer at prev_blkno so we can update the existing
4508  * extent's bucket count.  header_bh is the bucket were we were hoping
4509  * to insert our xattr.  If the bucket move places the target in the new
4510  * extent, we'll update first_bh and header_bh after modifying the old
4511  * extent.
4512  *
4513  * first_hash will be set as the 1st xe's name_hash in the new extent.
4514  */
4515 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4516                                                handle_t *handle,
4517                                                struct ocfs2_xattr_bucket *first,
4518                                                struct ocfs2_xattr_bucket *target,
4519                                                u64 new_blkno,
4520                                                u32 num_clusters,
4521                                                u32 *first_hash)
4522 {
4523         int ret;
4524         struct super_block *sb = inode->i_sb;
4525         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4526         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4527         int to_move = num_buckets / 2;
4528         u64 src_blkno;
4529         u64 last_cluster_blkno = bucket_blkno(first) +
4530                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4531
4532         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4533         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4534
4535         trace_ocfs2_mv_xattr_bucket_cross_cluster(
4536                                 (unsigned long long)last_cluster_blkno,
4537                                 (unsigned long long)new_blkno);
4538
4539         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4540                                      last_cluster_blkno, new_blkno,
4541                                      to_move, first_hash);
4542         if (ret) {
4543                 mlog_errno(ret);
4544                 goto out;
4545         }
4546
4547         /* This is the first bucket that got moved */
4548         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4549
4550         /*
4551          * If the target bucket was part of the moved buckets, we need to
4552          * update first and target.
4553          */
4554         if (bucket_blkno(target) >= src_blkno) {
4555                 /* Find the block for the new target bucket */
4556                 src_blkno = new_blkno +
4557                         (bucket_blkno(target) - src_blkno);
4558
4559                 ocfs2_xattr_bucket_relse(first);
4560                 ocfs2_xattr_bucket_relse(target);
4561
4562                 /*
4563                  * These shouldn't fail - the buffers are in the
4564                  * journal from ocfs2_cp_xattr_bucket().
4565                  */
4566                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4567                 if (ret) {
4568                         mlog_errno(ret);
4569                         goto out;
4570                 }
4571                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4572                 if (ret)
4573                         mlog_errno(ret);
4574
4575         }
4576
4577 out:
4578         return ret;
4579 }
4580
4581 /*
4582  * Find the suitable pos when we divide a bucket into 2.
4583  * We have to make sure the xattrs with the same hash value exist
4584  * in the same bucket.
4585  *
4586  * If this ocfs2_xattr_header covers more than one hash value, find a
4587  * place where the hash value changes.  Try to find the most even split.
4588  * The most common case is that all entries have different hash values,
4589  * and the first check we make will find a place to split.
4590  */
4591 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4592 {
4593         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4594         int count = le16_to_cpu(xh->xh_count);
4595         int delta, middle = count / 2;
4596
4597         /*
4598          * We start at the middle.  Each step gets farther away in both
4599          * directions.  We therefore hit the change in hash value
4600          * nearest to the middle.  Note that this loop does not execute for
4601          * count < 2.
4602          */
4603         for (delta = 0; delta < middle; delta++) {
4604                 /* Let's check delta earlier than middle */
4605                 if (cmp_xe(&entries[middle - delta - 1],
4606                            &entries[middle - delta]))
4607                         return middle - delta;
4608
4609                 /* For even counts, don't walk off the end */
4610                 if ((middle + delta + 1) == count)
4611                         continue;
4612
4613                 /* Now try delta past middle */
4614                 if (cmp_xe(&entries[middle + delta],
4615                            &entries[middle + delta + 1]))
4616                         return middle + delta + 1;
4617         }
4618
4619         /* Every entry had the same hash */
4620         return count;
4621 }
4622
4623 /*
4624  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4625  * first_hash will record the 1st hash of the new bucket.
4626  *
4627  * Normally half of the xattrs will be moved.  But we have to make
4628  * sure that the xattrs with the same hash value are stored in the
4629  * same bucket. If all the xattrs in this bucket have the same hash
4630  * value, the new bucket will be initialized as an empty one and the
4631  * first_hash will be initialized as (hash_value+1).
4632  */
4633 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4634                                     handle_t *handle,
4635                                     u64 blk,
4636                                     u64 new_blk,
4637                                     u32 *first_hash,
4638                                     int new_bucket_head)
4639 {
4640         int ret, i;
4641         int count, start, len, name_value_len = 0, name_offset = 0;
4642         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4643         struct ocfs2_xattr_header *xh;
4644         struct ocfs2_xattr_entry *xe;
4645         int blocksize = inode->i_sb->s_blocksize;
4646
4647         trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4648                                               (unsigned long long)new_blk);
4649
4650         s_bucket = ocfs2_xattr_bucket_new(inode);
4651         t_bucket = ocfs2_xattr_bucket_new(inode);
4652         if (!s_bucket || !t_bucket) {
4653                 ret = -ENOMEM;
4654                 mlog_errno(ret);
4655                 goto out;
4656         }
4657
4658         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4659         if (ret) {
4660                 mlog_errno(ret);
4661                 goto out;
4662         }
4663
4664         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4665                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4666         if (ret) {
4667                 mlog_errno(ret);
4668                 goto out;
4669         }
4670
4671         /*
4672          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4673          * there's no need to read it.
4674          */
4675         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4676         if (ret) {
4677                 mlog_errno(ret);
4678                 goto out;
4679         }
4680
4681         /*
4682          * Hey, if we're overwriting t_bucket, what difference does
4683          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4684          * same part of ocfs2_cp_xattr_bucket().
4685          */
4686         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4687                                                 new_bucket_head ?
4688                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4689                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4690         if (ret) {
4691                 mlog_errno(ret);
4692                 goto out;
4693         }
4694
4695         xh = bucket_xh(s_bucket);
4696         count = le16_to_cpu(xh->xh_count);
4697         start = ocfs2_xattr_find_divide_pos(xh);
4698
4699         if (start == count) {
4700                 xe = &xh->xh_entries[start-1];
4701
4702                 /*
4703                  * initialized a new empty bucket here.
4704                  * The hash value is set as one larger than
4705                  * that of the last entry in the previous bucket.
4706                  */
4707                 for (i = 0; i < t_bucket->bu_blocks; i++)
4708                         memset(bucket_block(t_bucket, i), 0, blocksize);
4709
4710                 xh = bucket_xh(t_bucket);
4711                 xh->xh_free_start = cpu_to_le16(blocksize);
4712                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4713                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4714
4715                 goto set_num_buckets;
4716         }
4717
4718         /* copy the whole bucket to the new first. */
4719         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4720
4721         /* update the new bucket. */
4722         xh = bucket_xh(t_bucket);
4723
4724         /*
4725          * Calculate the total name/value len and xh_free_start for
4726          * the old bucket first.
4727          */
4728         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4729         name_value_len = 0;
4730         for (i = 0; i < start; i++) {
4731                 xe = &xh->xh_entries[i];
4732                 name_value_len += namevalue_size_xe(xe);
4733                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4734                         name_offset = le16_to_cpu(xe->xe_name_offset);
4735         }
4736
4737         /*
4738          * Now begin the modification to the new bucket.
4739          *
4740          * In the new bucket, We just move the xattr entry to the beginning
4741          * and don't touch the name/value. So there will be some holes in the
4742          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4743          * called.
4744          */
4745         xe = &xh->xh_entries[start];
4746         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4747         trace_ocfs2_divide_xattr_bucket_move(len,
4748                         (int)((char *)xe - (char *)xh),
4749                         (int)((char *)xh->xh_entries - (char *)xh));
4750         memmove((char *)xh->xh_entries, (char *)xe, len);
4751         xe = &xh->xh_entries[count - start];
4752         len = sizeof(struct ocfs2_xattr_entry) * start;
4753         memset((char *)xe, 0, len);
4754
4755         le16_add_cpu(&xh->xh_count, -start);
4756         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4757
4758         /* Calculate xh_free_start for the new bucket. */
4759         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4760         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4761                 xe = &xh->xh_entries[i];
4762                 if (le16_to_cpu(xe->xe_name_offset) <
4763                     le16_to_cpu(xh->xh_free_start))
4764                         xh->xh_free_start = xe->xe_name_offset;
4765         }
4766
4767 set_num_buckets:
4768         /* set xh->xh_num_buckets for the new xh. */
4769         if (new_bucket_head)
4770                 xh->xh_num_buckets = cpu_to_le16(1);
4771         else
4772                 xh->xh_num_buckets = 0;
4773
4774         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4775
4776         /* store the first_hash of the new bucket. */
4777         if (first_hash)
4778                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4779
4780         /*
4781          * Now only update the 1st block of the old bucket.  If we
4782          * just added a new empty bucket, there is no need to modify
4783          * it.
4784          */
4785         if (start == count)
4786                 goto out;
4787
4788         xh = bucket_xh(s_bucket);
4789         memset(&xh->xh_entries[start], 0,
4790                sizeof(struct ocfs2_xattr_entry) * (count - start));
4791         xh->xh_count = cpu_to_le16(start);
4792         xh->xh_free_start = cpu_to_le16(name_offset);
4793         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4794
4795         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4796
4797 out:
4798         ocfs2_xattr_bucket_free(s_bucket);
4799         ocfs2_xattr_bucket_free(t_bucket);
4800
4801         return ret;
4802 }
4803
4804 /*
4805  * Copy xattr from one bucket to another bucket.
4806  *
4807  * The caller must make sure that the journal transaction
4808  * has enough space for journaling.
4809  */
4810 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4811                                  handle_t *handle,
4812                                  u64 s_blkno,
4813                                  u64 t_blkno,
4814                                  int t_is_new)
4815 {
4816         int ret;
4817         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4818
4819         BUG_ON(s_blkno == t_blkno);
4820
4821         trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4822                                     (unsigned long long)t_blkno,
4823                                     t_is_new);
4824
4825         s_bucket = ocfs2_xattr_bucket_new(inode);
4826         t_bucket = ocfs2_xattr_bucket_new(inode);
4827         if (!s_bucket || !t_bucket) {
4828                 ret = -ENOMEM;
4829                 mlog_errno(ret);
4830                 goto out;
4831         }
4832
4833         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4834         if (ret)
4835                 goto out;
4836
4837         /*
4838          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4839          * there's no need to read it.
4840          */
4841         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4842         if (ret)
4843                 goto out;
4844
4845         /*
4846          * Hey, if we're overwriting t_bucket, what difference does
4847          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4848          * cluster to fill, we came here from
4849          * ocfs2_mv_xattr_buckets(), and it is really new -
4850          * ACCESS_CREATE is required.  But we also might have moved data
4851          * out of t_bucket before extending back into it.
4852          * ocfs2_add_new_xattr_bucket() can do this - its call to
4853          * ocfs2_add_new_xattr_cluster() may have created a new extent
4854          * and copied out the end of the old extent.  Then it re-extends
4855          * the old extent back to create space for new xattrs.  That's
4856          * how we get here, and the bucket isn't really new.
4857          */
4858         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4859                                                 t_is_new ?
4860                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4861                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4862         if (ret)
4863                 goto out;
4864
4865         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4866         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4867
4868 out:
4869         ocfs2_xattr_bucket_free(t_bucket);
4870         ocfs2_xattr_bucket_free(s_bucket);
4871
4872         return ret;
4873 }
4874
4875 /*
4876  * src_blk points to the start of an existing extent.  last_blk points to
4877  * last cluster in that extent.  to_blk points to a newly allocated
4878  * extent.  We copy the buckets from the cluster at last_blk to the new
4879  * extent.  If start_bucket is non-zero, we skip that many buckets before
4880  * we start copying.  The new extent's xh_num_buckets gets set to the
4881  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4882  * by the same amount.
4883  */
4884 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4885                                   u64 src_blk, u64 last_blk, u64 to_blk,
4886                                   unsigned int start_bucket,
4887                                   u32 *first_hash)
4888 {
4889         int i, ret, credits;
4890         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4891         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4892         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4893         struct ocfs2_xattr_bucket *old_first, *new_first;
4894
4895         trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4896                                      (unsigned long long)to_blk);
4897
4898         BUG_ON(start_bucket >= num_buckets);
4899         if (start_bucket) {
4900                 num_buckets -= start_bucket;
4901                 last_blk += (start_bucket * blks_per_bucket);
4902         }
4903
4904         /* The first bucket of the original extent */
4905         old_first = ocfs2_xattr_bucket_new(inode);
4906         /* The first bucket of the new extent */
4907         new_first = ocfs2_xattr_bucket_new(inode);
4908         if (!old_first || !new_first) {
4909                 ret = -ENOMEM;
4910                 mlog_errno(ret);
4911                 goto out;
4912         }
4913
4914         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4915         if (ret) {
4916                 mlog_errno(ret);
4917                 goto out;
4918         }
4919
4920         /*
4921          * We need to update the first bucket of the old extent and all
4922          * the buckets going to the new extent.
4923          */
4924         credits = ((num_buckets + 1) * blks_per_bucket);
4925         ret = ocfs2_extend_trans(handle, credits);
4926         if (ret) {
4927                 mlog_errno(ret);
4928                 goto out;
4929         }
4930
4931         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4932                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4933         if (ret) {
4934                 mlog_errno(ret);
4935                 goto out;
4936         }
4937
4938         for (i = 0; i < num_buckets; i++) {
4939                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4940                                             last_blk + (i * blks_per_bucket),
4941                                             to_blk + (i * blks_per_bucket),
4942                                             1);
4943                 if (ret) {
4944                         mlog_errno(ret);
4945                         goto out;
4946                 }
4947         }
4948
4949         /*
4950          * Get the new bucket ready before we dirty anything
4951          * (This actually shouldn't fail, because we already dirtied
4952          * it once in ocfs2_cp_xattr_bucket()).
4953          */
4954         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4955         if (ret) {
4956                 mlog_errno(ret);
4957                 goto out;
4958         }
4959         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4960                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4961         if (ret) {
4962                 mlog_errno(ret);
4963                 goto out;
4964         }
4965
4966         /* Now update the headers */
4967         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4968         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4969
4970         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4971         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4972
4973         if (first_hash)
4974                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4975
4976 out:
4977         ocfs2_xattr_bucket_free(new_first);
4978         ocfs2_xattr_bucket_free(old_first);
4979         return ret;
4980 }
4981
4982 /*
4983  * Move some xattrs in this cluster to the new cluster.
4984  * This function should only be called when bucket size == cluster size.
4985  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4986  */
4987 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4988                                       handle_t *handle,
4989                                       u64 prev_blk,
4990                                       u64 new_blk,
4991                                       u32 *first_hash)
4992 {
4993         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4994         int ret, credits = 2 * blk_per_bucket;
4995
4996         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4997
4998         ret = ocfs2_extend_trans(handle, credits);
4999         if (ret) {
5000                 mlog_errno(ret);
5001                 return ret;
5002         }
5003
5004         /* Move half of the xattr in start_blk to the next bucket. */
5005         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
5006                                           new_blk, first_hash, 1);
5007 }
5008
5009 /*
5010  * Move some xattrs from the old cluster to the new one since they are not
5011  * contiguous in ocfs2 xattr tree.
5012  *
5013  * new_blk starts a new separate cluster, and we will move some xattrs from
5014  * prev_blk to it. v_start will be set as the first name hash value in this
5015  * new cluster so that it can be used as e_cpos during tree insertion and
5016  * don't collide with our original b-tree operations. first_bh and header_bh
5017  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5018  * to extend the insert bucket.
5019  *
5020  * The problem is how much xattr should we move to the new one and when should
5021  * we update first_bh and header_bh?
5022  * 1. If cluster size > bucket size, that means the previous cluster has more
5023  *    than 1 bucket, so just move half nums of bucket into the new cluster and
5024  *    update the first_bh and header_bh if the insert bucket has been moved
5025  *    to the new cluster.
5026  * 2. If cluster_size == bucket_size:
5027  *    a) If the previous extent rec has more than one cluster and the insert
5028  *       place isn't in the last cluster, copy the entire last cluster to the
5029  *       new one. This time, we don't need to upate the first_bh and header_bh
5030  *       since they will not be moved into the new cluster.
5031  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
5032  *       the new one. And we set the extend flag to zero if the insert place is
5033  *       moved into the new allocated cluster since no extend is needed.
5034  */
5035 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5036                                             handle_t *handle,
5037                                             struct ocfs2_xattr_bucket *first,
5038                                             struct ocfs2_xattr_bucket *target,
5039                                             u64 new_blk,
5040                                             u32 prev_clusters,
5041                                             u32 *v_start,
5042                                             int *extend)
5043 {
5044         int ret;
5045
5046         trace_ocfs2_adjust_xattr_cross_cluster(
5047                         (unsigned long long)bucket_blkno(first),
5048                         (unsigned long long)new_blk, prev_clusters);
5049
5050         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5051                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5052                                                           handle,
5053                                                           first, target,
5054                                                           new_blk,
5055                                                           prev_clusters,
5056                                                           v_start);
5057                 if (ret)
5058                         mlog_errno(ret);
5059         } else {
5060                 /* The start of the last cluster in the first extent */
5061                 u64 last_blk = bucket_blkno(first) +
5062                         ((prev_clusters - 1) *
5063                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
5064
5065                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5066                         ret = ocfs2_mv_xattr_buckets(inode, handle,
5067                                                      bucket_blkno(first),
5068                                                      last_blk, new_blk, 0,
5069                                                      v_start);
5070                         if (ret)
5071                                 mlog_errno(ret);
5072                 } else {
5073                         ret = ocfs2_divide_xattr_cluster(inode, handle,
5074                                                          last_blk, new_blk,
5075                                                          v_start);
5076                         if (ret)
5077                                 mlog_errno(ret);
5078
5079                         if ((bucket_blkno(target) == last_blk) && extend)
5080                                 *extend = 0;
5081                 }
5082         }
5083
5084         return ret;
5085 }
5086
5087 /*
5088  * Add a new cluster for xattr storage.
5089  *
5090  * If the new cluster is contiguous with the previous one, it will be
5091  * appended to the same extent record, and num_clusters will be updated.
5092  * If not, we will insert a new extent for it and move some xattrs in
5093  * the last cluster into the new allocated one.
5094  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5095  * lose the benefits of hashing because we'll have to search large leaves.
5096  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5097  * if it's bigger).
5098  *
5099  * first_bh is the first block of the previous extent rec and header_bh
5100  * indicates the bucket we will insert the new xattrs. They will be updated
5101  * when the header_bh is moved into the new cluster.
5102  */
5103 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5104                                        struct buffer_head *root_bh,
5105                                        struct ocfs2_xattr_bucket *first,
5106                                        struct ocfs2_xattr_bucket *target,
5107                                        u32 *num_clusters,
5108                                        u32 prev_cpos,
5109                                        int *extend,
5110                                        struct ocfs2_xattr_set_ctxt *ctxt)
5111 {
5112         int ret;
5113         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5114         u32 prev_clusters = *num_clusters;
5115         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5116         u64 block;
5117         handle_t *handle = ctxt->handle;
5118         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5119         struct ocfs2_extent_tree et;
5120
5121         trace_ocfs2_add_new_xattr_cluster_begin(
5122                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5123                 (unsigned long long)bucket_blkno(first),
5124                 prev_cpos, prev_clusters);
5125
5126         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5127
5128         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5129                                       OCFS2_JOURNAL_ACCESS_WRITE);
5130         if (ret < 0) {
5131                 mlog_errno(ret);
5132                 goto leave;
5133         }
5134
5135         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5136                                      clusters_to_add, &bit_off, &num_bits);
5137         if (ret < 0) {
5138                 if (ret != -ENOSPC)
5139                         mlog_errno(ret);
5140                 goto leave;
5141         }
5142
5143         BUG_ON(num_bits > clusters_to_add);
5144
5145         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5146         trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5147
5148         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5149             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5150              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5151                 /*
5152                  * If this cluster is contiguous with the old one and
5153                  * adding this new cluster, we don't surpass the limit of
5154                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5155                  * initialized and used like other buckets in the previous
5156                  * cluster.
5157                  * So add it as a contiguous one. The caller will handle
5158                  * its init process.
5159                  */
5160                 v_start = prev_cpos + prev_clusters;
5161                 *num_clusters = prev_clusters + num_bits;
5162         } else {
5163                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5164                                                        handle,
5165                                                        first,
5166                                                        target,
5167                                                        block,
5168                                                        prev_clusters,
5169                                                        &v_start,
5170                                                        extend);
5171                 if (ret) {
5172                         mlog_errno(ret);
5173                         goto leave;
5174                 }
5175         }
5176
5177         trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5178                                                  v_start, num_bits);
5179         ret = ocfs2_insert_extent(handle, &et, v_start, block,
5180                                   num_bits, 0, ctxt->meta_ac);
5181         if (ret < 0) {
5182                 mlog_errno(ret);
5183                 goto leave;
5184         }
5185
5186         ocfs2_journal_dirty(handle, root_bh);
5187
5188 leave:
5189         return ret;
5190 }
5191
5192 /*
5193  * We are given an extent.  'first' is the bucket at the very front of
5194  * the extent.  The extent has space for an additional bucket past
5195  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5196  * of the target bucket.  We wish to shift every bucket past the target
5197  * down one, filling in that additional space.  When we get back to the
5198  * target, we split the target between itself and the now-empty bucket
5199  * at target+1 (aka, target_blkno + blks_per_bucket).
5200  */
5201 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5202                                      handle_t *handle,
5203                                      struct ocfs2_xattr_bucket *first,
5204                                      u64 target_blk,
5205                                      u32 num_clusters)
5206 {
5207         int ret, credits;
5208         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5209         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5210         u64 end_blk;
5211         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5212
5213         trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5214                                         (unsigned long long)bucket_blkno(first),
5215                                         num_clusters, new_bucket);
5216
5217         /* The extent must have room for an additional bucket */
5218         BUG_ON(new_bucket >=
5219                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5220
5221         /* end_blk points to the last existing bucket */
5222         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5223
5224         /*
5225          * end_blk is the start of the last existing bucket.
5226          * Thus, (end_blk - target_blk) covers the target bucket and
5227          * every bucket after it up to, but not including, the last
5228          * existing bucket.  Then we add the last existing bucket, the
5229          * new bucket, and the first bucket (3 * blk_per_bucket).
5230          */
5231         credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5232         ret = ocfs2_extend_trans(handle, credits);
5233         if (ret) {
5234                 mlog_errno(ret);
5235                 goto out;
5236         }
5237
5238         ret = ocfs2_xattr_bucket_journal_access(handle, first,
5239                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5240         if (ret) {
5241                 mlog_errno(ret);
5242                 goto out;
5243         }
5244
5245         while (end_blk != target_blk) {
5246                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5247                                             end_blk + blk_per_bucket, 0);
5248                 if (ret)
5249                         goto out;
5250                 end_blk -= blk_per_bucket;
5251         }
5252
5253         /* Move half of the xattr in target_blkno to the next bucket. */
5254         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5255                                         target_blk + blk_per_bucket, NULL, 0);
5256
5257         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5258         ocfs2_xattr_bucket_journal_dirty(handle, first);
5259
5260 out:
5261         return ret;
5262 }
5263
5264 /*
5265  * Add new xattr bucket in an extent record and adjust the buckets
5266  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5267  * bucket we want to insert into.
5268  *
5269  * In the easy case, we will move all the buckets after target down by
5270  * one. Half of target's xattrs will be moved to the next bucket.
5271  *
5272  * If current cluster is full, we'll allocate a new one.  This may not
5273  * be contiguous.  The underlying calls will make sure that there is
5274  * space for the insert, shifting buckets around if necessary.
5275  * 'target' may be moved by those calls.
5276  */
5277 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5278                                       struct buffer_head *xb_bh,
5279                                       struct ocfs2_xattr_bucket *target,
5280                                       struct ocfs2_xattr_set_ctxt *ctxt)
5281 {
5282         struct ocfs2_xattr_block *xb =
5283                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5284         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5285         struct ocfs2_extent_list *el = &xb_root->xt_list;
5286         u32 name_hash =
5287                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5288         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5289         int ret, num_buckets, extend = 1;
5290         u64 p_blkno;
5291         u32 e_cpos, num_clusters;
5292         /* The bucket at the front of the extent */
5293         struct ocfs2_xattr_bucket *first;
5294
5295         trace_ocfs2_add_new_xattr_bucket(
5296                                 (unsigned long long)bucket_blkno(target));
5297
5298         /* The first bucket of the original extent */
5299         first = ocfs2_xattr_bucket_new(inode);
5300         if (!first) {
5301                 ret = -ENOMEM;
5302                 mlog_errno(ret);
5303                 goto out;
5304         }
5305
5306         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5307                                   &num_clusters, el);
5308         if (ret) {
5309                 mlog_errno(ret);
5310                 goto out;
5311         }
5312
5313         ret = ocfs2_read_xattr_bucket(first, p_blkno);
5314         if (ret) {
5315                 mlog_errno(ret);
5316                 goto out;
5317         }
5318
5319         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5320         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5321                 /*
5322                  * This can move first+target if the target bucket moves
5323                  * to the new extent.
5324                  */
5325                 ret = ocfs2_add_new_xattr_cluster(inode,
5326                                                   xb_bh,
5327                                                   first,
5328                                                   target,
5329                                                   &num_clusters,
5330                                                   e_cpos,
5331                                                   &extend,
5332                                                   ctxt);
5333                 if (ret) {
5334                         mlog_errno(ret);
5335                         goto out;
5336                 }
5337         }
5338
5339         if (extend) {
5340                 ret = ocfs2_extend_xattr_bucket(inode,
5341                                                 ctxt->handle,
5342                                                 first,
5343                                                 bucket_blkno(target),
5344                                                 num_clusters);
5345                 if (ret)
5346                         mlog_errno(ret);
5347         }
5348
5349 out:
5350         ocfs2_xattr_bucket_free(first);
5351
5352         return ret;
5353 }
5354
5355 /*
5356  * Truncate the specified xe_off entry in xattr bucket.
5357  * bucket is indicated by header_bh and len is the new length.
5358  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5359  *
5360  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5361  */
5362 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5363                                              struct ocfs2_xattr_bucket *bucket,
5364                                              int xe_off,
5365                                              int len,
5366                                              struct ocfs2_xattr_set_ctxt *ctxt)
5367 {
5368         int ret, offset;
5369         u64 value_blk;
5370         struct ocfs2_xattr_entry *xe;
5371         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5372         size_t blocksize = inode->i_sb->s_blocksize;
5373         struct ocfs2_xattr_value_buf vb = {
5374                 .vb_access = ocfs2_journal_access,
5375         };
5376
5377         xe = &xh->xh_entries[xe_off];
5378
5379         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5380
5381         offset = le16_to_cpu(xe->xe_name_offset) +
5382                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5383
5384         value_blk = offset / blocksize;
5385
5386         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5387         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5388
5389         vb.vb_bh = bucket->bu_bhs[value_blk];
5390         BUG_ON(!vb.vb_bh);
5391
5392         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5393                 (vb.vb_bh->b_data + offset % blocksize);
5394
5395         /*
5396          * From here on out we have to dirty the bucket.  The generic
5397          * value calls only modify one of the bucket's bhs, but we need
5398          * to send the bucket at once.  So if they error, they *could* have
5399          * modified something.  We have to assume they did, and dirty
5400          * the whole bucket.  This leaves us in a consistent state.
5401          */
5402         trace_ocfs2_xattr_bucket_value_truncate(
5403                         (unsigned long long)bucket_blkno(bucket), xe_off, len);
5404         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5405         if (ret) {
5406                 mlog_errno(ret);
5407                 goto out;
5408         }
5409
5410         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5411                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5412         if (ret) {
5413                 mlog_errno(ret);
5414                 goto out;
5415         }
5416
5417         xe->xe_value_size = cpu_to_le64(len);
5418
5419         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5420
5421 out:
5422         return ret;
5423 }
5424
5425 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5426                                   struct buffer_head *root_bh,
5427                                   u64 blkno,
5428                                   u32 cpos,
5429                                   u32 len,
5430                                   void *para)
5431 {
5432         int ret;
5433         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5434         struct inode *tl_inode = osb->osb_tl_inode;
5435         handle_t *handle;
5436         struct ocfs2_xattr_block *xb =
5437                         (struct ocfs2_xattr_block *)root_bh->b_data;
5438         struct ocfs2_alloc_context *meta_ac = NULL;
5439         struct ocfs2_cached_dealloc_ctxt dealloc;
5440         struct ocfs2_extent_tree et;
5441
5442         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5443                                           ocfs2_delete_xattr_in_bucket, para);
5444         if (ret) {
5445                 mlog_errno(ret);
5446                 return ret;
5447         }
5448
5449         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5450
5451         ocfs2_init_dealloc_ctxt(&dealloc);
5452
5453         trace_ocfs2_rm_xattr_cluster(
5454                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
5455                         (unsigned long long)blkno, cpos, len);
5456
5457         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5458                                                len);
5459
5460         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5461         if (ret) {
5462                 mlog_errno(ret);
5463                 return ret;
5464         }
5465
5466         inode_lock(tl_inode);
5467
5468         if (ocfs2_truncate_log_needs_flush(osb)) {
5469                 ret = __ocfs2_flush_truncate_log(osb);
5470                 if (ret < 0) {
5471                         mlog_errno(ret);
5472                         goto out;
5473                 }
5474         }
5475
5476         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5477         if (IS_ERR(handle)) {
5478                 ret = -ENOMEM;
5479                 mlog_errno(ret);
5480                 goto out;
5481         }
5482
5483         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5484                                       OCFS2_JOURNAL_ACCESS_WRITE);
5485         if (ret) {
5486                 mlog_errno(ret);
5487                 goto out_commit;
5488         }
5489
5490         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5491                                   &dealloc);
5492         if (ret) {
5493                 mlog_errno(ret);
5494                 goto out_commit;
5495         }
5496
5497         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5498         ocfs2_journal_dirty(handle, root_bh);
5499
5500         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5501         if (ret)
5502                 mlog_errno(ret);
5503         ocfs2_update_inode_fsync_trans(handle, inode, 0);
5504
5505 out_commit:
5506         ocfs2_commit_trans(osb, handle);
5507 out:
5508         ocfs2_schedule_truncate_log_flush(osb, 1);
5509
5510         inode_unlock(tl_inode);
5511
5512         if (meta_ac)
5513                 ocfs2_free_alloc_context(meta_ac);
5514
5515         ocfs2_run_deallocs(osb, &dealloc);
5516
5517         return ret;
5518 }
5519
5520 /*
5521  * check whether the xattr bucket is filled up with the same hash value.
5522  * If we want to insert the xattr with the same hash, return -ENOSPC.
5523  * If we want to insert a xattr with different hash value, go ahead
5524  * and ocfs2_divide_xattr_bucket will handle this.
5525  */
5526 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5527                                               struct ocfs2_xattr_bucket *bucket,
5528                                               const char *name)
5529 {
5530         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5531         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5532
5533         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5534                 return 0;
5535
5536         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5537             xh->xh_entries[0].xe_name_hash) {
5538                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5539                      "hash = %u\n",
5540                      (unsigned long long)bucket_blkno(bucket),
5541                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5542                 return -ENOSPC;
5543         }
5544
5545         return 0;
5546 }
5547
5548 /*
5549  * Try to set the entry in the current bucket.  If we fail, the caller
5550  * will handle getting us another bucket.
5551  */
5552 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5553                                         struct ocfs2_xattr_info *xi,
5554                                         struct ocfs2_xattr_search *xs,
5555                                         struct ocfs2_xattr_set_ctxt *ctxt)
5556 {
5557         int ret;
5558         struct ocfs2_xa_loc loc;
5559
5560         trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5561
5562         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5563                                        xs->not_found ? NULL : xs->here);
5564         ret = ocfs2_xa_set(&loc, xi, ctxt);
5565         if (!ret) {
5566                 xs->here = loc.xl_entry;
5567                 goto out;
5568         }
5569         if (ret != -ENOSPC) {
5570                 mlog_errno(ret);
5571                 goto out;
5572         }
5573
5574         /* Ok, we need space.  Let's try defragmenting the bucket. */
5575         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5576                                         xs->bucket);
5577         if (ret) {
5578                 mlog_errno(ret);
5579                 goto out;
5580         }
5581
5582         ret = ocfs2_xa_set(&loc, xi, ctxt);
5583         if (!ret) {
5584                 xs->here = loc.xl_entry;
5585                 goto out;
5586         }
5587         if (ret != -ENOSPC)
5588                 mlog_errno(ret);
5589
5590
5591 out:
5592         return ret;
5593 }
5594
5595 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5596                                              struct ocfs2_xattr_info *xi,
5597                                              struct ocfs2_xattr_search *xs,
5598                                              struct ocfs2_xattr_set_ctxt *ctxt)
5599 {
5600         int ret;
5601
5602         trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5603
5604         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5605         if (!ret)
5606                 goto out;
5607         if (ret != -ENOSPC) {
5608                 mlog_errno(ret);
5609                 goto out;
5610         }
5611
5612         /* Ack, need more space.  Let's try to get another bucket! */
5613
5614         /*
5615          * We do not allow for overlapping ranges between buckets. And
5616          * the maximum number of collisions we will allow for then is
5617          * one bucket's worth, so check it here whether we need to
5618          * add a new bucket for the insert.
5619          */
5620         ret = ocfs2_check_xattr_bucket_collision(inode,
5621                                                  xs->bucket,
5622                                                  xi->xi_name);
5623         if (ret) {
5624                 mlog_errno(ret);
5625                 goto out;
5626         }
5627
5628         ret = ocfs2_add_new_xattr_bucket(inode,
5629                                          xs->xattr_bh,
5630                                          xs->bucket,
5631                                          ctxt);
5632         if (ret) {
5633                 mlog_errno(ret);
5634                 goto out;
5635         }
5636
5637         /*
5638          * ocfs2_add_new_xattr_bucket() will have updated
5639          * xs->bucket if it moved, but it will not have updated
5640          * any of the other search fields.  Thus, we drop it and
5641          * re-search.  Everything should be cached, so it'll be
5642          * quick.
5643          */
5644         ocfs2_xattr_bucket_relse(xs->bucket);
5645         ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5646                                            xi->xi_name_index,
5647                                            xi->xi_name, xs);
5648         if (ret && ret != -ENODATA)
5649                 goto out;
5650         xs->not_found = ret;
5651
5652         /* Ok, we have a new bucket, let's try again */
5653         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5654         if (ret && (ret != -ENOSPC))
5655                 mlog_errno(ret);
5656
5657 out:
5658         return ret;
5659 }
5660
5661 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5662                                         struct ocfs2_xattr_bucket *bucket,
5663                                         void *para)
5664 {
5665         int ret = 0, ref_credits;
5666         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5667         u16 i;
5668         struct ocfs2_xattr_entry *xe;
5669         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5670         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5671         int credits = ocfs2_remove_extent_credits(osb->sb) +
5672                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5673         struct ocfs2_xattr_value_root *xv;
5674         struct ocfs2_rm_xattr_bucket_para *args =
5675                         (struct ocfs2_rm_xattr_bucket_para *)para;
5676
5677         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5678
5679         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5680                 xe = &xh->xh_entries[i];
5681                 if (ocfs2_xattr_is_local(xe))
5682                         continue;
5683
5684                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5685                                                       i, &xv, NULL);
5686                 if (ret) {
5687                         mlog_errno(ret);
5688                         break;
5689                 }
5690
5691                 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5692                                                          args->ref_ci,
5693                                                          args->ref_root_bh,
5694                                                          &ctxt.meta_ac,
5695                                                          &ref_credits);
5696
5697                 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5698                 if (IS_ERR(ctxt.handle)) {
5699                         ret = PTR_ERR(ctxt.handle);
5700                         mlog_errno(ret);
5701                         break;
5702                 }
5703
5704                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5705                                                         i, 0, &ctxt);
5706
5707                 ocfs2_commit_trans(osb, ctxt.handle);
5708                 if (ctxt.meta_ac) {
5709                         ocfs2_free_alloc_context(ctxt.meta_ac);
5710                         ctxt.meta_ac = NULL;
5711                 }
5712                 if (ret) {
5713                         mlog_errno(ret);
5714                         break;
5715                 }
5716         }
5717
5718         if (ctxt.meta_ac)
5719                 ocfs2_free_alloc_context(ctxt.meta_ac);
5720         ocfs2_schedule_truncate_log_flush(osb, 1);
5721         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5722         return ret;
5723 }
5724
5725 /*
5726  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5727  * or change the extent record flag), we need to recalculate
5728  * the metaecc for the whole bucket. So it is done here.
5729  *
5730  * Note:
5731  * We have to give the extra credits for the caller.
5732  */
5733 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5734                                             handle_t *handle,
5735                                             void *para)
5736 {
5737         int ret;
5738         struct ocfs2_xattr_bucket *bucket =
5739                         (struct ocfs2_xattr_bucket *)para;
5740
5741         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5742                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5743         if (ret) {
5744                 mlog_errno(ret);
5745                 return ret;
5746         }
5747
5748         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5749
5750         return 0;
5751 }
5752
5753 /*
5754  * Special action we need if the xattr value is refcounted.
5755  *
5756  * 1. If the xattr is refcounted, lock the tree.
5757  * 2. CoW the xattr if we are setting the new value and the value
5758  *    will be stored outside.
5759  * 3. In other case, decrease_refcount will work for us, so just
5760  *    lock the refcount tree, calculate the meta and credits is OK.
5761  *
5762  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5763  * currently CoW is a completed transaction, while this function
5764  * will also lock the allocators and let us deadlock. So we will
5765  * CoW the whole xattr value.
5766  */
5767 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5768                                         struct ocfs2_dinode *di,
5769                                         struct ocfs2_xattr_info *xi,
5770                                         struct ocfs2_xattr_search *xis,
5771                                         struct ocfs2_xattr_search *xbs,
5772                                         struct ocfs2_refcount_tree **ref_tree,
5773                                         int *meta_add,
5774                                         int *credits)
5775 {
5776         int ret = 0;
5777         struct ocfs2_xattr_block *xb;
5778         struct ocfs2_xattr_entry *xe;
5779         char *base;
5780         u32 p_cluster, num_clusters;
5781         unsigned int ext_flags;
5782         int name_offset, name_len;
5783         struct ocfs2_xattr_value_buf vb;
5784         struct ocfs2_xattr_bucket *bucket = NULL;
5785         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5786         struct ocfs2_post_refcount refcount;
5787         struct ocfs2_post_refcount *p = NULL;
5788         struct buffer_head *ref_root_bh = NULL;
5789
5790         if (!xis->not_found) {
5791                 xe = xis->here;
5792                 name_offset = le16_to_cpu(xe->xe_name_offset);
5793                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5794                 base = xis->base;
5795                 vb.vb_bh = xis->inode_bh;
5796                 vb.vb_access = ocfs2_journal_access_di;
5797         } else {
5798                 int i, block_off = 0;
5799                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5800                 xe = xbs->here;
5801                 name_offset = le16_to_cpu(xe->xe_name_offset);
5802                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5803                 i = xbs->here - xbs->header->xh_entries;
5804
5805                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5806                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5807                                                         bucket_xh(xbs->bucket),
5808                                                         i, &block_off,
5809                                                         &name_offset);
5810                         if (ret) {
5811                                 mlog_errno(ret);
5812                                 goto out;
5813                         }
5814                         base = bucket_block(xbs->bucket, block_off);
5815                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5816                         vb.vb_access = ocfs2_journal_access;
5817
5818                         if (ocfs2_meta_ecc(osb)) {
5819                                 /*create parameters for ocfs2_post_refcount. */
5820                                 bucket = xbs->bucket;
5821                                 refcount.credits = bucket->bu_blocks;
5822                                 refcount.para = bucket;
5823                                 refcount.func =
5824                                         ocfs2_xattr_bucket_post_refcount;
5825                                 p = &refcount;
5826                         }
5827                 } else {
5828                         base = xbs->base;
5829                         vb.vb_bh = xbs->xattr_bh;
5830                         vb.vb_access = ocfs2_journal_access_xb;
5831                 }
5832         }
5833
5834         if (ocfs2_xattr_is_local(xe))
5835                 goto out;
5836
5837         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5838                                 (base + name_offset + name_len);
5839
5840         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5841                                        &num_clusters, &vb.vb_xv->xr_list,
5842                                        &ext_flags);
5843         if (ret) {
5844                 mlog_errno(ret);
5845                 goto out;
5846         }
5847
5848         /*
5849          * We just need to check the 1st extent record, since we always
5850          * CoW the whole xattr. So there shouldn't be a xattr with
5851          * some REFCOUNT extent recs after the 1st one.
5852          */
5853         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5854                 goto out;
5855
5856         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5857                                        1, ref_tree, &ref_root_bh);
5858         if (ret) {
5859                 mlog_errno(ret);
5860                 goto out;
5861         }
5862
5863         /*
5864          * If we are deleting the xattr or the new size will be stored inside,
5865          * cool, leave it there, the xattr truncate process will remove them
5866          * for us(it still needs the refcount tree lock and the meta, credits).
5867          * And the worse case is that every cluster truncate will split the
5868          * refcount tree, and make the original extent become 3. So we will need
5869          * 2 * cluster more extent recs at most.
5870          */
5871         if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5872
5873                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5874                                                          &(*ref_tree)->rf_ci,
5875                                                          ref_root_bh, vb.vb_xv,
5876                                                          meta_add, credits);
5877                 if (ret)
5878                         mlog_errno(ret);
5879                 goto out;
5880         }
5881
5882         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5883                                        *ref_tree, ref_root_bh, 0,
5884                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5885         if (ret)
5886                 mlog_errno(ret);
5887
5888 out:
5889         brelse(ref_root_bh);
5890         return ret;
5891 }
5892
5893 /*
5894  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5895  * The physical clusters will be added to refcount tree.
5896  */
5897 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5898                                 struct ocfs2_xattr_value_root *xv,
5899                                 struct ocfs2_extent_tree *value_et,
5900                                 struct ocfs2_caching_info *ref_ci,
5901                                 struct buffer_head *ref_root_bh,
5902                                 struct ocfs2_cached_dealloc_ctxt *dealloc,
5903                                 struct ocfs2_post_refcount *refcount)
5904 {
5905         int ret = 0;
5906         u32 clusters = le32_to_cpu(xv->xr_clusters);
5907         u32 cpos, p_cluster, num_clusters;
5908         struct ocfs2_extent_list *el = &xv->xr_list;
5909         unsigned int ext_flags;
5910
5911         cpos = 0;
5912         while (cpos < clusters) {
5913                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5914                                                &num_clusters, el, &ext_flags);
5915                 if (ret) {
5916                         mlog_errno(ret);
5917                         break;
5918                 }
5919
5920                 cpos += num_clusters;
5921                 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5922                         continue;
5923
5924                 BUG_ON(!p_cluster);
5925
5926                 ret = ocfs2_add_refcount_flag(inode, value_et,
5927                                               ref_ci, ref_root_bh,
5928                                               cpos - num_clusters,
5929                                               p_cluster, num_clusters,
5930                                               dealloc, refcount);
5931                 if (ret) {
5932                         mlog_errno(ret);
5933                         break;
5934                 }
5935         }
5936
5937         return ret;
5938 }
5939
5940 /*
5941  * Given a normal ocfs2_xattr_header, refcount all the entries which
5942  * have value stored outside.
5943  * Used for xattrs stored in inode and ocfs2_xattr_block.
5944  */
5945 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5946                                 struct ocfs2_xattr_value_buf *vb,
5947                                 struct ocfs2_xattr_header *header,
5948                                 struct ocfs2_caching_info *ref_ci,
5949                                 struct buffer_head *ref_root_bh,
5950                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5951 {
5952
5953         struct ocfs2_xattr_entry *xe;
5954         struct ocfs2_xattr_value_root *xv;
5955         struct ocfs2_extent_tree et;
5956         int i, ret = 0;
5957
5958         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5959                 xe = &header->xh_entries[i];
5960
5961                 if (ocfs2_xattr_is_local(xe))
5962                         continue;
5963
5964                 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5965                         le16_to_cpu(xe->xe_name_offset) +
5966                         OCFS2_XATTR_SIZE(xe->xe_name_len));
5967
5968                 vb->vb_xv = xv;
5969                 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5970
5971                 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5972                                                         ref_ci, ref_root_bh,
5973                                                         dealloc, NULL);
5974                 if (ret) {
5975                         mlog_errno(ret);
5976                         break;
5977                 }
5978         }
5979
5980         return ret;
5981 }
5982
5983 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5984                                 struct buffer_head *fe_bh,
5985                                 struct ocfs2_caching_info *ref_ci,
5986                                 struct buffer_head *ref_root_bh,
5987                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5988 {
5989         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5990         struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5991                                 (fe_bh->b_data + inode->i_sb->s_blocksize -
5992                                 le16_to_cpu(di->i_xattr_inline_size));
5993         struct ocfs2_xattr_value_buf vb = {
5994                 .vb_bh = fe_bh,
5995                 .vb_access = ocfs2_journal_access_di,
5996         };
5997
5998         return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5999                                                   ref_ci, ref_root_bh, dealloc);
6000 }
6001
6002 struct ocfs2_xattr_tree_value_refcount_para {
6003         struct ocfs2_caching_info *ref_ci;
6004         struct buffer_head *ref_root_bh;
6005         struct ocfs2_cached_dealloc_ctxt *dealloc;
6006 };
6007
6008 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
6009                                            struct ocfs2_xattr_bucket *bucket,
6010                                            int offset,
6011                                            struct ocfs2_xattr_value_root **xv,
6012                                            struct buffer_head **bh)
6013 {
6014         int ret, block_off, name_offset;
6015         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6016         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6017         void *base;
6018
6019         ret = ocfs2_xattr_bucket_get_name_value(sb,
6020                                                 bucket_xh(bucket),
6021                                                 offset,
6022                                                 &block_off,
6023                                                 &name_offset);
6024         if (ret) {
6025                 mlog_errno(ret);
6026                 goto out;
6027         }
6028
6029         base = bucket_block(bucket, block_off);
6030
6031         *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6032                          OCFS2_XATTR_SIZE(xe->xe_name_len));
6033
6034         if (bh)
6035                 *bh = bucket->bu_bhs[block_off];
6036 out:
6037         return ret;
6038 }
6039
6040 /*
6041  * For a given xattr bucket, refcount all the entries which
6042  * have value stored outside.
6043  */
6044 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6045                                              struct ocfs2_xattr_bucket *bucket,
6046                                              void *para)
6047 {
6048         int i, ret = 0;
6049         struct ocfs2_extent_tree et;
6050         struct ocfs2_xattr_tree_value_refcount_para *ref =
6051                         (struct ocfs2_xattr_tree_value_refcount_para *)para;
6052         struct ocfs2_xattr_header *xh =
6053                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6054         struct ocfs2_xattr_entry *xe;
6055         struct ocfs2_xattr_value_buf vb = {
6056                 .vb_access = ocfs2_journal_access,
6057         };
6058         struct ocfs2_post_refcount refcount = {
6059                 .credits = bucket->bu_blocks,
6060                 .para = bucket,
6061                 .func = ocfs2_xattr_bucket_post_refcount,
6062         };
6063         struct ocfs2_post_refcount *p = NULL;
6064
6065         /* We only need post_refcount if we support metaecc. */
6066         if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6067                 p = &refcount;
6068
6069         trace_ocfs2_xattr_bucket_value_refcount(
6070                                 (unsigned long long)bucket_blkno(bucket),
6071                                 le16_to_cpu(xh->xh_count));
6072         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6073                 xe = &xh->xh_entries[i];
6074
6075                 if (ocfs2_xattr_is_local(xe))
6076                         continue;
6077
6078                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6079                                                       &vb.vb_xv, &vb.vb_bh);
6080                 if (ret) {
6081                         mlog_errno(ret);
6082                         break;
6083                 }
6084
6085                 ocfs2_init_xattr_value_extent_tree(&et,
6086                                                    INODE_CACHE(inode), &vb);
6087
6088                 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6089                                                         &et, ref->ref_ci,
6090                                                         ref->ref_root_bh,
6091                                                         ref->dealloc, p);
6092                 if (ret) {
6093                         mlog_errno(ret);
6094                         break;
6095                 }
6096         }
6097
6098         return ret;
6099
6100 }
6101
6102 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6103                                      struct buffer_head *root_bh,
6104                                      u64 blkno, u32 cpos, u32 len, void *para)
6105 {
6106         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6107                                            ocfs2_xattr_bucket_value_refcount,
6108                                            para);
6109 }
6110
6111 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6112                                 struct buffer_head *blk_bh,
6113                                 struct ocfs2_caching_info *ref_ci,
6114                                 struct buffer_head *ref_root_bh,
6115                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
6116 {
6117         int ret = 0;
6118         struct ocfs2_xattr_block *xb =
6119                                 (struct ocfs2_xattr_block *)blk_bh->b_data;
6120
6121         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6122                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6123                 struct ocfs2_xattr_value_buf vb = {
6124                         .vb_bh = blk_bh,
6125                         .vb_access = ocfs2_journal_access_xb,
6126                 };
6127
6128                 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6129                                                          ref_ci, ref_root_bh,
6130                                                          dealloc);
6131         } else {
6132                 struct ocfs2_xattr_tree_value_refcount_para para = {
6133                         .ref_ci = ref_ci,
6134                         .ref_root_bh = ref_root_bh,
6135                         .dealloc = dealloc,
6136                 };
6137
6138                 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6139                                                 ocfs2_refcount_xattr_tree_rec,
6140                                                 &para);
6141         }
6142
6143         return ret;
6144 }
6145
6146 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6147                                      struct buffer_head *fe_bh,
6148                                      struct ocfs2_caching_info *ref_ci,
6149                                      struct buffer_head *ref_root_bh,
6150                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
6151 {
6152         int ret = 0;
6153         struct ocfs2_inode_info *oi = OCFS2_I(inode);
6154         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6155         struct buffer_head *blk_bh = NULL;
6156
6157         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6158                 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6159                                                          ref_ci, ref_root_bh,
6160                                                          dealloc);
6161                 if (ret) {
6162                         mlog_errno(ret);
6163                         goto out;
6164                 }
6165         }
6166
6167         if (!di->i_xattr_loc)
6168                 goto out;
6169
6170         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6171                                      &blk_bh);
6172         if (ret < 0) {
6173                 mlog_errno(ret);
6174                 goto out;
6175         }
6176
6177         ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6178                                                 ref_root_bh, dealloc);
6179         if (ret)
6180                 mlog_errno(ret);
6181
6182         brelse(blk_bh);
6183 out:
6184
6185         return ret;
6186 }
6187
6188 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6189 /*
6190  * Store the information we need in xattr reflink.
6191  * old_bh and new_bh are inode bh for the old and new inode.
6192  */
6193 struct ocfs2_xattr_reflink {
6194         struct inode *old_inode;
6195         struct inode *new_inode;
6196         struct buffer_head *old_bh;
6197         struct buffer_head *new_bh;
6198         struct ocfs2_caching_info *ref_ci;
6199         struct buffer_head *ref_root_bh;
6200         struct ocfs2_cached_dealloc_ctxt *dealloc;
6201         should_xattr_reflinked *xattr_reflinked;
6202 };
6203
6204 /*
6205  * Given a xattr header and xe offset,
6206  * return the proper xv and the corresponding bh.
6207  * xattr in inode, block and xattr tree have different implementaions.
6208  */
6209 typedef int (get_xattr_value_root)(struct super_block *sb,
6210                                    struct buffer_head *bh,
6211                                    struct ocfs2_xattr_header *xh,
6212                                    int offset,
6213                                    struct ocfs2_xattr_value_root **xv,
6214                                    struct buffer_head **ret_bh,
6215                                    void *para);
6216
6217 /*
6218  * Calculate all the xattr value root metadata stored in this xattr header and
6219  * credits we need if we create them from the scratch.
6220  * We use get_xattr_value_root so that all types of xattr container can use it.
6221  */
6222 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6223                                              struct buffer_head *bh,
6224                                              struct ocfs2_xattr_header *xh,
6225                                              int *metas, int *credits,
6226                                              int *num_recs,
6227                                              get_xattr_value_root *func,
6228                                              void *para)
6229 {
6230         int i, ret = 0;
6231         struct ocfs2_xattr_value_root *xv;
6232         struct ocfs2_xattr_entry *xe;
6233
6234         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6235                 xe = &xh->xh_entries[i];
6236                 if (ocfs2_xattr_is_local(xe))
6237                         continue;
6238
6239                 ret = func(sb, bh, xh, i, &xv, NULL, para);
6240                 if (ret) {
6241                         mlog_errno(ret);
6242                         break;
6243                 }
6244
6245                 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6246                           le16_to_cpu(xv->xr_list.l_next_free_rec);
6247
6248                 *credits += ocfs2_calc_extend_credits(sb,
6249                                                 &def_xv.xv.xr_list);
6250
6251                 /*
6252                  * If the value is a tree with depth > 1, We don't go deep
6253                  * to the extent block, so just calculate a maximum record num.
6254                  */
6255                 if (!xv->xr_list.l_tree_depth)
6256                         *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6257                 else
6258                         *num_recs += ocfs2_clusters_for_bytes(sb,
6259                                                               XATTR_SIZE_MAX);
6260         }
6261
6262         return ret;
6263 }
6264
6265 /* Used by xattr inode and block to return the right xv and buffer_head. */
6266 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6267                                       struct buffer_head *bh,
6268                                       struct ocfs2_xattr_header *xh,
6269                                       int offset,
6270                                       struct ocfs2_xattr_value_root **xv,
6271                                       struct buffer_head **ret_bh,
6272                                       void *para)
6273 {
6274         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6275
6276         *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6277                 le16_to_cpu(xe->xe_name_offset) +
6278                 OCFS2_XATTR_SIZE(xe->xe_name_len));
6279
6280         if (ret_bh)
6281                 *ret_bh = bh;
6282
6283         return 0;
6284 }
6285
6286 /*
6287  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6288  * It is only used for inline xattr and xattr block.
6289  */
6290 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6291                                         struct ocfs2_xattr_header *xh,
6292                                         struct buffer_head *ref_root_bh,
6293                                         int *credits,
6294                                         struct ocfs2_alloc_context **meta_ac)
6295 {
6296         int ret, meta_add = 0, num_recs = 0;
6297         struct ocfs2_refcount_block *rb =
6298                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6299
6300         *credits = 0;
6301
6302         ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6303                                                 &meta_add, credits, &num_recs,
6304                                                 ocfs2_get_xattr_value_root,
6305                                                 NULL);
6306         if (ret) {
6307                 mlog_errno(ret);
6308                 goto out;
6309         }
6310
6311         /*
6312          * We need to add/modify num_recs in refcount tree, so just calculate
6313          * an approximate number we need for refcount tree change.
6314          * Sometimes we need to split the tree, and after split,  half recs
6315          * will be moved to the new block, and a new block can only provide
6316          * half number of recs. So we multiple new blocks by 2.
6317          */
6318         num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6319         meta_add += num_recs;
6320         *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6321         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6322                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6323                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6324         else
6325                 *credits += 1;
6326
6327         ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6328         if (ret)
6329                 mlog_errno(ret);
6330
6331 out:
6332         return ret;
6333 }
6334
6335 /*
6336  * Given a xattr header, reflink all the xattrs in this container.
6337  * It can be used for inode, block and bucket.
6338  *
6339  * NOTE:
6340  * Before we call this function, the caller has memcpy the xattr in
6341  * old_xh to the new_xh.
6342  *
6343  * If args.xattr_reflinked is set, call it to decide whether the xe should
6344  * be reflinked or not. If not, remove it from the new xattr header.
6345  */
6346 static int ocfs2_reflink_xattr_header(handle_t *handle,
6347                                       struct ocfs2_xattr_reflink *args,
6348                                       struct buffer_head *old_bh,
6349                                       struct ocfs2_xattr_header *xh,
6350                                       struct buffer_head *new_bh,
6351                                       struct ocfs2_xattr_header *new_xh,
6352                                       struct ocfs2_xattr_value_buf *vb,
6353                                       struct ocfs2_alloc_context *meta_ac,
6354                                       get_xattr_value_root *func,
6355                                       void *para)
6356 {
6357         int ret = 0, i, j;
6358         struct super_block *sb = args->old_inode->i_sb;
6359         struct buffer_head *value_bh;
6360         struct ocfs2_xattr_entry *xe, *last;
6361         struct ocfs2_xattr_value_root *xv, *new_xv;
6362         struct ocfs2_extent_tree data_et;
6363         u32 clusters, cpos, p_cluster, num_clusters;
6364         unsigned int ext_flags = 0;
6365
6366         trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6367                                          le16_to_cpu(xh->xh_count));
6368
6369         last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6370         for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6371                 xe = &xh->xh_entries[i];
6372
6373                 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6374                         xe = &new_xh->xh_entries[j];
6375
6376                         le16_add_cpu(&new_xh->xh_count, -1);
6377                         if (new_xh->xh_count) {
6378                                 memmove(xe, xe + 1,
6379                                         (void *)last - (void *)xe);
6380                                 memset(last, 0,
6381                                        sizeof(struct ocfs2_xattr_entry));
6382                         }
6383
6384                         /*
6385                          * We don't want j to increase in the next round since
6386                          * it is already moved ahead.
6387                          */
6388                         j--;
6389                         continue;
6390                 }
6391
6392                 if (ocfs2_xattr_is_local(xe))
6393                         continue;
6394
6395                 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6396                 if (ret) {
6397                         mlog_errno(ret);
6398                         break;
6399                 }
6400
6401                 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6402                 if (ret) {
6403                         mlog_errno(ret);
6404                         break;
6405                 }
6406
6407                 /*
6408                  * For the xattr which has l_tree_depth = 0, all the extent
6409                  * recs have already be copied to the new xh with the
6410                  * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6411                  * increase the refount count int the refcount tree.
6412                  *
6413                  * For the xattr which has l_tree_depth > 0, we need
6414                  * to initialize it to the empty default value root,
6415                  * and then insert the extents one by one.
6416                  */
6417                 if (xv->xr_list.l_tree_depth) {
6418                         memcpy(new_xv, &def_xv, sizeof(def_xv));
6419                         vb->vb_xv = new_xv;
6420                         vb->vb_bh = value_bh;
6421                         ocfs2_init_xattr_value_extent_tree(&data_et,
6422                                         INODE_CACHE(args->new_inode), vb);
6423                 }
6424
6425                 clusters = le32_to_cpu(xv->xr_clusters);
6426                 cpos = 0;
6427                 while (cpos < clusters) {
6428                         ret = ocfs2_xattr_get_clusters(args->old_inode,
6429                                                        cpos,
6430                                                        &p_cluster,
6431                                                        &num_clusters,
6432                                                        &xv->xr_list,
6433                                                        &ext_flags);
6434                         if (ret) {
6435                                 mlog_errno(ret);
6436                                 goto out;
6437                         }
6438
6439                         BUG_ON(!p_cluster);
6440
6441                         if (xv->xr_list.l_tree_depth) {
6442                                 ret = ocfs2_insert_extent(handle,
6443                                                 &data_et, cpos,
6444                                                 ocfs2_clusters_to_blocks(
6445                                                         args->old_inode->i_sb,
6446                                                         p_cluster),
6447                                                 num_clusters, ext_flags,
6448                                                 meta_ac);
6449                                 if (ret) {
6450                                         mlog_errno(ret);
6451                                         goto out;
6452                                 }
6453                         }
6454
6455                         ret = ocfs2_increase_refcount(handle, args->ref_ci,
6456                                                       args->ref_root_bh,
6457                                                       p_cluster, num_clusters,
6458                                                       meta_ac, args->dealloc);
6459                         if (ret) {
6460                                 mlog_errno(ret);
6461                                 goto out;
6462                         }
6463
6464                         cpos += num_clusters;
6465                 }
6466         }
6467
6468 out:
6469         return ret;
6470 }
6471
6472 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6473 {
6474         int ret = 0, credits = 0;
6475         handle_t *handle;
6476         struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6477         struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6478         int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6479         int header_off = osb->sb->s_blocksize - inline_size;
6480         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6481                                         (args->old_bh->b_data + header_off);
6482         struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6483                                         (args->new_bh->b_data + header_off);
6484         struct ocfs2_alloc_context *meta_ac = NULL;
6485         struct ocfs2_inode_info *new_oi;
6486         struct ocfs2_dinode *new_di;
6487         struct ocfs2_xattr_value_buf vb = {
6488                 .vb_bh = args->new_bh,
6489                 .vb_access = ocfs2_journal_access_di,
6490         };
6491
6492         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6493                                                   &credits, &meta_ac);
6494         if (ret) {
6495                 mlog_errno(ret);
6496                 goto out;
6497         }
6498
6499         handle = ocfs2_start_trans(osb, credits);
6500         if (IS_ERR(handle)) {
6501                 ret = PTR_ERR(handle);
6502                 mlog_errno(ret);
6503                 goto out;
6504         }
6505
6506         ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6507                                       args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6508         if (ret) {
6509                 mlog_errno(ret);
6510                 goto out_commit;
6511         }
6512
6513         memcpy(args->new_bh->b_data + header_off,
6514                args->old_bh->b_data + header_off, inline_size);
6515
6516         new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6517         new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6518
6519         ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6520                                          args->new_bh, new_xh, &vb, meta_ac,
6521                                          ocfs2_get_xattr_value_root, NULL);
6522         if (ret) {
6523                 mlog_errno(ret);
6524                 goto out_commit;
6525         }
6526
6527         new_oi = OCFS2_I(args->new_inode);
6528         /*
6529          * Adjust extent record count to reserve space for extended attribute.
6530          * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6531          */
6532         if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6533             !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6534                 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6535                 le16_add_cpu(&el->l_count, -(inline_size /
6536                                         sizeof(struct ocfs2_extent_rec)));
6537         }
6538         spin_lock(&new_oi->ip_lock);
6539         new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6540         new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6541         spin_unlock(&new_oi->ip_lock);
6542
6543         ocfs2_journal_dirty(handle, args->new_bh);
6544
6545 out_commit:
6546         ocfs2_commit_trans(osb, handle);
6547
6548 out:
6549         if (meta_ac)
6550                 ocfs2_free_alloc_context(meta_ac);
6551         return ret;
6552 }
6553
6554 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6555                                           struct buffer_head *fe_bh,
6556                                           struct buffer_head **ret_bh,
6557                                           int indexed)
6558 {
6559         int ret;
6560         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6561         struct ocfs2_xattr_set_ctxt ctxt;
6562
6563         memset(&ctxt, 0, sizeof(ctxt));
6564         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6565         if (ret < 0) {
6566                 mlog_errno(ret);
6567                 return ret;
6568         }
6569
6570         ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6571         if (IS_ERR(ctxt.handle)) {
6572                 ret = PTR_ERR(ctxt.handle);
6573                 mlog_errno(ret);
6574                 goto out;
6575         }
6576
6577         trace_ocfs2_create_empty_xattr_block(
6578                                 (unsigned long long)fe_bh->b_blocknr, indexed);
6579         ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6580                                        ret_bh);
6581         if (ret)
6582                 mlog_errno(ret);
6583
6584         ocfs2_commit_trans(osb, ctxt.handle);
6585 out:
6586         ocfs2_free_alloc_context(ctxt.meta_ac);
6587         return ret;
6588 }
6589
6590 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6591                                      struct buffer_head *blk_bh,
6592                                      struct buffer_head *new_blk_bh)
6593 {
6594         int ret = 0, credits = 0;
6595         handle_t *handle;
6596         struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6597         struct ocfs2_dinode *new_di;
6598         struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6599         int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6600         struct ocfs2_xattr_block *xb =
6601                         (struct ocfs2_xattr_block *)blk_bh->b_data;
6602         struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6603         struct ocfs2_xattr_block *new_xb =
6604                         (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6605         struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6606         struct ocfs2_alloc_context *meta_ac;
6607         struct ocfs2_xattr_value_buf vb = {
6608                 .vb_bh = new_blk_bh,
6609                 .vb_access = ocfs2_journal_access_xb,
6610         };
6611
6612         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6613                                                   &credits, &meta_ac);
6614         if (ret) {
6615                 mlog_errno(ret);
6616                 return ret;
6617         }
6618
6619         /* One more credits in case we need to add xattr flags in new inode. */
6620         handle = ocfs2_start_trans(osb, credits + 1);
6621         if (IS_ERR(handle)) {
6622                 ret = PTR_ERR(handle);
6623                 mlog_errno(ret);
6624                 goto out;
6625         }
6626
6627         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6628                 ret = ocfs2_journal_access_di(handle,
6629                                               INODE_CACHE(args->new_inode),
6630                                               args->new_bh,
6631                                               OCFS2_JOURNAL_ACCESS_WRITE);
6632                 if (ret) {
6633                         mlog_errno(ret);
6634                         goto out_commit;
6635                 }
6636         }
6637
6638         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6639                                       new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6640         if (ret) {
6641                 mlog_errno(ret);
6642                 goto out_commit;
6643         }
6644
6645         memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6646                osb->sb->s_blocksize - header_off);
6647
6648         ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6649                                          new_blk_bh, new_xh, &vb, meta_ac,
6650                                          ocfs2_get_xattr_value_root, NULL);
6651         if (ret) {
6652                 mlog_errno(ret);
6653                 goto out_commit;
6654         }
6655
6656         ocfs2_journal_dirty(handle, new_blk_bh);
6657
6658         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6659                 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6660                 spin_lock(&new_oi->ip_lock);
6661                 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6662                 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6663                 spin_unlock(&new_oi->ip_lock);
6664
6665                 ocfs2_journal_dirty(handle, args->new_bh);
6666         }
6667
6668 out_commit:
6669         ocfs2_commit_trans(osb, handle);
6670
6671 out:
6672         ocfs2_free_alloc_context(meta_ac);
6673         return ret;
6674 }
6675
6676 struct ocfs2_reflink_xattr_tree_args {
6677         struct ocfs2_xattr_reflink *reflink;
6678         struct buffer_head *old_blk_bh;
6679         struct buffer_head *new_blk_bh;
6680         struct ocfs2_xattr_bucket *old_bucket;
6681         struct ocfs2_xattr_bucket *new_bucket;
6682 };
6683
6684 /*
6685  * NOTE:
6686  * We have to handle the case that both old bucket and new bucket
6687  * will call this function to get the right ret_bh.
6688  * So The caller must give us the right bh.
6689  */
6690 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6691                                         struct buffer_head *bh,
6692                                         struct ocfs2_xattr_header *xh,
6693                                         int offset,
6694                                         struct ocfs2_xattr_value_root **xv,
6695                                         struct buffer_head **ret_bh,
6696                                         void *para)
6697 {
6698         struct ocfs2_reflink_xattr_tree_args *args =
6699                         (struct ocfs2_reflink_xattr_tree_args *)para;
6700         struct ocfs2_xattr_bucket *bucket;
6701
6702         if (bh == args->old_bucket->bu_bhs[0])
6703                 bucket = args->old_bucket;
6704         else
6705                 bucket = args->new_bucket;
6706
6707         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6708                                                xv, ret_bh);
6709 }
6710
6711 struct ocfs2_value_tree_metas {
6712         int num_metas;
6713         int credits;
6714         int num_recs;
6715 };
6716
6717 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6718                                         struct buffer_head *bh,
6719                                         struct ocfs2_xattr_header *xh,
6720                                         int offset,
6721                                         struct ocfs2_xattr_value_root **xv,
6722                                         struct buffer_head **ret_bh,
6723                                         void *para)
6724 {
6725         struct ocfs2_xattr_bucket *bucket =
6726                                 (struct ocfs2_xattr_bucket *)para;
6727
6728         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6729                                                xv, ret_bh);
6730 }
6731
6732 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6733                                       struct ocfs2_xattr_bucket *bucket,
6734                                       void *para)
6735 {
6736         struct ocfs2_value_tree_metas *metas =
6737                         (struct ocfs2_value_tree_metas *)para;
6738         struct ocfs2_xattr_header *xh =
6739                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6740
6741         /* Add the credits for this bucket first. */
6742         metas->credits += bucket->bu_blocks;
6743         return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6744                                         xh, &metas->num_metas,
6745                                         &metas->credits, &metas->num_recs,
6746                                         ocfs2_value_tree_metas_in_bucket,
6747                                         bucket);
6748 }
6749
6750 /*
6751  * Given a xattr extent rec starting from blkno and having len clusters,
6752  * iterate all the buckets calculate how much metadata we need for reflinking
6753  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6754  */
6755 static int ocfs2_lock_reflink_xattr_rec_allocators(
6756                                 struct ocfs2_reflink_xattr_tree_args *args,
6757                                 struct ocfs2_extent_tree *xt_et,
6758                                 u64 blkno, u32 len, int *credits,
6759                                 struct ocfs2_alloc_context **meta_ac,
6760                                 struct ocfs2_alloc_context **data_ac)
6761 {
6762         int ret, num_free_extents;
6763         struct ocfs2_value_tree_metas metas;
6764         struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6765         struct ocfs2_refcount_block *rb;
6766
6767         memset(&metas, 0, sizeof(metas));
6768
6769         ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6770                                           ocfs2_calc_value_tree_metas, &metas);
6771         if (ret) {
6772                 mlog_errno(ret);
6773                 goto out;
6774         }
6775
6776         *credits = metas.credits;
6777
6778         /*
6779          * Calculate we need for refcount tree change.
6780          *
6781          * We need to add/modify num_recs in refcount tree, so just calculate
6782          * an approximate number we need for refcount tree change.
6783          * Sometimes we need to split the tree, and after split,  half recs
6784          * will be moved to the new block, and a new block can only provide
6785          * half number of recs. So we multiple new blocks by 2.
6786          * In the end, we have to add credits for modifying the already
6787          * existed refcount block.
6788          */
6789         rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6790         metas.num_recs =
6791                 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6792                  ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6793         metas.num_metas += metas.num_recs;
6794         *credits += metas.num_recs +
6795                     metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6796         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6797                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6798                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6799         else
6800                 *credits += 1;
6801
6802         /* count in the xattr tree change. */
6803         num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6804         if (num_free_extents < 0) {
6805                 ret = num_free_extents;
6806                 mlog_errno(ret);
6807                 goto out;
6808         }
6809
6810         if (num_free_extents < len)
6811                 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6812
6813         *credits += ocfs2_calc_extend_credits(osb->sb,
6814                                               xt_et->et_root_el);
6815
6816         if (metas.num_metas) {
6817                 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6818                                                         meta_ac);
6819                 if (ret) {
6820                         mlog_errno(ret);
6821                         goto out;
6822                 }
6823         }
6824
6825         if (len) {
6826                 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6827                 if (ret)
6828                         mlog_errno(ret);
6829         }
6830 out:
6831         if (ret) {
6832                 if (*meta_ac) {
6833                         ocfs2_free_alloc_context(*meta_ac);
6834                         *meta_ac = NULL;
6835                 }
6836         }
6837
6838         return ret;
6839 }
6840
6841 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6842                                 u64 blkno, u64 new_blkno, u32 clusters,
6843                                 u32 *cpos, int num_buckets,
6844                                 struct ocfs2_alloc_context *meta_ac,
6845                                 struct ocfs2_alloc_context *data_ac,
6846                                 struct ocfs2_reflink_xattr_tree_args *args)
6847 {
6848         int i, j, ret = 0;
6849         struct super_block *sb = args->reflink->old_inode->i_sb;
6850         int bpb = args->old_bucket->bu_blocks;
6851         struct ocfs2_xattr_value_buf vb = {
6852                 .vb_access = ocfs2_journal_access,
6853         };
6854
6855         for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6856                 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6857                 if (ret) {
6858                         mlog_errno(ret);
6859                         break;
6860                 }
6861
6862                 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6863                 if (ret) {
6864                         mlog_errno(ret);
6865                         break;
6866                 }
6867
6868                 ret = ocfs2_xattr_bucket_journal_access(handle,
6869                                                 args->new_bucket,
6870                                                 OCFS2_JOURNAL_ACCESS_CREATE);
6871                 if (ret) {
6872                         mlog_errno(ret);
6873                         break;
6874                 }
6875
6876                 for (j = 0; j < bpb; j++)
6877                         memcpy(bucket_block(args->new_bucket, j),
6878                                bucket_block(args->old_bucket, j),
6879                                sb->s_blocksize);
6880
6881                 /*
6882                  * Record the start cpos so that we can use it to initialize
6883                  * our xattr tree we also set the xh_num_bucket for the new
6884                  * bucket.
6885                  */
6886                 if (i == 0) {
6887                         *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6888                                             xh_entries[0].xe_name_hash);
6889                         bucket_xh(args->new_bucket)->xh_num_buckets =
6890                                 cpu_to_le16(num_buckets);
6891                 }
6892
6893                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6894
6895                 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6896                                         args->old_bucket->bu_bhs[0],
6897                                         bucket_xh(args->old_bucket),
6898                                         args->new_bucket->bu_bhs[0],
6899                                         bucket_xh(args->new_bucket),
6900                                         &vb, meta_ac,
6901                                         ocfs2_get_reflink_xattr_value_root,
6902                                         args);
6903                 if (ret) {
6904                         mlog_errno(ret);
6905                         break;
6906                 }
6907
6908                 /*
6909                  * Re-access and dirty the bucket to calculate metaecc.
6910                  * Because we may extend the transaction in reflink_xattr_header
6911                  * which will let the already accessed block gone.
6912                  */
6913                 ret = ocfs2_xattr_bucket_journal_access(handle,
6914                                                 args->new_bucket,
6915                                                 OCFS2_JOURNAL_ACCESS_WRITE);
6916                 if (ret) {
6917                         mlog_errno(ret);
6918                         break;
6919                 }
6920
6921                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6922
6923                 ocfs2_xattr_bucket_relse(args->old_bucket);
6924                 ocfs2_xattr_bucket_relse(args->new_bucket);
6925         }
6926
6927         ocfs2_xattr_bucket_relse(args->old_bucket);
6928         ocfs2_xattr_bucket_relse(args->new_bucket);
6929         return ret;
6930 }
6931
6932 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6933                                 struct inode *inode,
6934                                 struct ocfs2_reflink_xattr_tree_args *args,
6935                                 struct ocfs2_extent_tree *et,
6936                                 struct ocfs2_alloc_context *meta_ac,
6937                                 struct ocfs2_alloc_context *data_ac,
6938                                 u64 blkno, u32 cpos, u32 len)
6939 {
6940         int ret, first_inserted = 0;
6941         u32 p_cluster, num_clusters, reflink_cpos = 0;
6942         u64 new_blkno;
6943         unsigned int num_buckets, reflink_buckets;
6944         unsigned int bpc =
6945                 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6946
6947         ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6948         if (ret) {
6949                 mlog_errno(ret);
6950                 goto out;
6951         }
6952         num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6953         ocfs2_xattr_bucket_relse(args->old_bucket);
6954
6955         while (len && num_buckets) {
6956                 ret = ocfs2_claim_clusters(handle, data_ac,
6957                                            1, &p_cluster, &num_clusters);
6958                 if (ret) {
6959                         mlog_errno(ret);
6960                         goto out;
6961                 }
6962
6963                 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6964                 reflink_buckets = min(num_buckets, bpc * num_clusters);
6965
6966                 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6967                                                  new_blkno, num_clusters,
6968                                                  &reflink_cpos, reflink_buckets,
6969                                                  meta_ac, data_ac, args);
6970                 if (ret) {
6971                         mlog_errno(ret);
6972                         goto out;
6973                 }
6974
6975                 /*
6976                  * For the 1st allocated cluster, we make it use the same cpos
6977                  * so that the xattr tree looks the same as the original one
6978                  * in the most case.
6979                  */
6980                 if (!first_inserted) {
6981                         reflink_cpos = cpos;
6982                         first_inserted = 1;
6983                 }
6984                 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6985                                           num_clusters, 0, meta_ac);
6986                 if (ret)
6987                         mlog_errno(ret);
6988
6989                 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6990                                                   num_clusters, reflink_cpos);
6991
6992                 len -= num_clusters;
6993                 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6994                 num_buckets -= reflink_buckets;
6995         }
6996 out:
6997         return ret;
6998 }
6999
7000 /*
7001  * Create the same xattr extent record in the new inode's xattr tree.
7002  */
7003 static int ocfs2_reflink_xattr_rec(struct inode *inode,
7004                                    struct buffer_head *root_bh,
7005                                    u64 blkno,
7006                                    u32 cpos,
7007                                    u32 len,
7008                                    void *para)
7009 {
7010         int ret, credits = 0;
7011         handle_t *handle;
7012         struct ocfs2_reflink_xattr_tree_args *args =
7013                         (struct ocfs2_reflink_xattr_tree_args *)para;
7014         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7015         struct ocfs2_alloc_context *meta_ac = NULL;
7016         struct ocfs2_alloc_context *data_ac = NULL;
7017         struct ocfs2_extent_tree et;
7018
7019         trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
7020
7021         ocfs2_init_xattr_tree_extent_tree(&et,
7022                                           INODE_CACHE(args->reflink->new_inode),
7023                                           args->new_blk_bh);
7024
7025         ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7026                                                       len, &credits,
7027                                                       &meta_ac, &data_ac);
7028         if (ret) {
7029                 mlog_errno(ret);
7030                 goto out;
7031         }
7032
7033         handle = ocfs2_start_trans(osb, credits);
7034         if (IS_ERR(handle)) {
7035                 ret = PTR_ERR(handle);
7036                 mlog_errno(ret);
7037                 goto out;
7038         }
7039
7040         ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7041                                           meta_ac, data_ac,
7042                                           blkno, cpos, len);
7043         if (ret)
7044                 mlog_errno(ret);
7045
7046         ocfs2_commit_trans(osb, handle);
7047
7048 out:
7049         if (meta_ac)
7050                 ocfs2_free_alloc_context(meta_ac);
7051         if (data_ac)
7052                 ocfs2_free_alloc_context(data_ac);
7053         return ret;
7054 }
7055
7056 /*
7057  * Create reflinked xattr buckets.
7058  * We will add bucket one by one, and refcount all the xattrs in the bucket
7059  * if they are stored outside.
7060  */
7061 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7062                                     struct buffer_head *blk_bh,
7063                                     struct buffer_head *new_blk_bh)
7064 {
7065         int ret;
7066         struct ocfs2_reflink_xattr_tree_args para;
7067
7068         memset(&para, 0, sizeof(para));
7069         para.reflink = args;
7070         para.old_blk_bh = blk_bh;
7071         para.new_blk_bh = new_blk_bh;
7072
7073         para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7074         if (!para.old_bucket) {
7075                 mlog_errno(-ENOMEM);
7076                 return -ENOMEM;
7077         }
7078
7079         para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7080         if (!para.new_bucket) {
7081                 ret = -ENOMEM;
7082                 mlog_errno(ret);
7083                 goto out;
7084         }
7085
7086         ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7087                                               ocfs2_reflink_xattr_rec,
7088                                               &para);
7089         if (ret)
7090                 mlog_errno(ret);
7091
7092 out:
7093         ocfs2_xattr_bucket_free(para.old_bucket);
7094         ocfs2_xattr_bucket_free(para.new_bucket);
7095         return ret;
7096 }
7097
7098 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7099                                         struct buffer_head *blk_bh)
7100 {
7101         int ret, indexed = 0;
7102         struct buffer_head *new_blk_bh = NULL;
7103         struct ocfs2_xattr_block *xb =
7104                         (struct ocfs2_xattr_block *)blk_bh->b_data;
7105
7106
7107         if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7108                 indexed = 1;
7109
7110         ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7111                                              &new_blk_bh, indexed);
7112         if (ret) {
7113                 mlog_errno(ret);
7114                 goto out;
7115         }
7116
7117         if (!indexed)
7118                 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7119         else
7120                 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7121         if (ret)
7122                 mlog_errno(ret);
7123
7124 out:
7125         brelse(new_blk_bh);
7126         return ret;
7127 }
7128
7129 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7130 {
7131         int type = ocfs2_xattr_get_type(xe);
7132
7133         return type != OCFS2_XATTR_INDEX_SECURITY &&
7134                type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7135                type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7136 }
7137
7138 int ocfs2_reflink_xattrs(struct inode *old_inode,
7139                          struct buffer_head *old_bh,
7140                          struct inode *new_inode,
7141                          struct buffer_head *new_bh,
7142                          bool preserve_security)
7143 {
7144         int ret;
7145         struct ocfs2_xattr_reflink args;
7146         struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7147         struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7148         struct buffer_head *blk_bh = NULL;
7149         struct ocfs2_cached_dealloc_ctxt dealloc;
7150         struct ocfs2_refcount_tree *ref_tree;
7151         struct buffer_head *ref_root_bh = NULL;
7152
7153         ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7154                                        le64_to_cpu(di->i_refcount_loc),
7155                                        1, &ref_tree, &ref_root_bh);
7156         if (ret) {
7157                 mlog_errno(ret);
7158                 goto out;
7159         }
7160
7161         ocfs2_init_dealloc_ctxt(&dealloc);
7162
7163         args.old_inode = old_inode;
7164         args.new_inode = new_inode;
7165         args.old_bh = old_bh;
7166         args.new_bh = new_bh;
7167         args.ref_ci = &ref_tree->rf_ci;
7168         args.ref_root_bh = ref_root_bh;
7169         args.dealloc = &dealloc;
7170         if (preserve_security)
7171                 args.xattr_reflinked = NULL;
7172         else
7173                 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7174
7175         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7176                 ret = ocfs2_reflink_xattr_inline(&args);
7177                 if (ret) {
7178                         mlog_errno(ret);
7179                         goto out_unlock;
7180                 }
7181         }
7182
7183         if (!di->i_xattr_loc)
7184                 goto out_unlock;
7185
7186         ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7187                                      &blk_bh);
7188         if (ret < 0) {
7189                 mlog_errno(ret);
7190                 goto out_unlock;
7191         }
7192
7193         ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7194         if (ret)
7195                 mlog_errno(ret);
7196
7197         brelse(blk_bh);
7198
7199 out_unlock:
7200         ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7201                                    ref_tree, 1);
7202         brelse(ref_root_bh);
7203
7204         if (ocfs2_dealloc_has_cluster(&dealloc)) {
7205                 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7206                 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7207         }
7208
7209 out:
7210         return ret;
7211 }
7212
7213 /*
7214  * Initialize security and acl for a already created inode.
7215  * Used for reflink a non-preserve-security file.
7216  *
7217  * It uses common api like ocfs2_xattr_set, so the caller
7218  * must not hold any lock expect i_mutex.
7219  */
7220 int ocfs2_init_security_and_acl(struct inode *dir,
7221                                 struct inode *inode,
7222                                 const struct qstr *qstr)
7223 {
7224         int ret = 0;
7225         struct buffer_head *dir_bh = NULL;
7226
7227         ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7228         if (ret) {
7229                 mlog_errno(ret);
7230                 goto leave;
7231         }
7232
7233         ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7234         if (ret) {
7235                 mlog_errno(ret);
7236                 goto leave;
7237         }
7238         ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7239         if (ret)
7240                 mlog_errno(ret);
7241
7242         ocfs2_inode_unlock(dir, 0);
7243         brelse(dir_bh);
7244 leave:
7245         return ret;
7246 }
7247
7248 /*
7249  * 'security' attributes support
7250  */
7251 static int ocfs2_xattr_security_get(const struct xattr_handler *handler,
7252                                     struct dentry *unused, struct inode *inode,
7253                                     const char *name, void *buffer, size_t size)
7254 {
7255         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY,
7256                                name, buffer, size);
7257 }
7258
7259 static int ocfs2_xattr_security_set(const struct xattr_handler *handler,
7260                                     struct dentry *unused, struct inode *inode,
7261                                     const char *name, const void *value,
7262                                     size_t size, int flags)
7263 {
7264         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7265                                name, value, size, flags);
7266 }
7267
7268 static int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7269                      void *fs_info)
7270 {
7271         const struct xattr *xattr;
7272         int err = 0;
7273
7274         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7275                 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7276                                       xattr->name, xattr->value,
7277                                       xattr->value_len, XATTR_CREATE);
7278                 if (err)
7279                         break;
7280         }
7281         return err;
7282 }
7283
7284 int ocfs2_init_security_get(struct inode *inode,
7285                             struct inode *dir,
7286                             const struct qstr *qstr,
7287                             struct ocfs2_security_xattr_info *si)
7288 {
7289         /* check whether ocfs2 support feature xattr */
7290         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7291                 return -EOPNOTSUPP;
7292         if (si)
7293                 return security_old_inode_init_security(inode, dir, qstr,
7294                                                         &si->name, &si->value,
7295                                                         &si->value_len);
7296
7297         return security_inode_init_security(inode, dir, qstr,
7298                                             &ocfs2_initxattrs, NULL);
7299 }
7300
7301 int ocfs2_init_security_set(handle_t *handle,
7302                             struct inode *inode,
7303                             struct buffer_head *di_bh,
7304                             struct ocfs2_security_xattr_info *si,
7305                             struct ocfs2_alloc_context *xattr_ac,
7306                             struct ocfs2_alloc_context *data_ac)
7307 {
7308         return ocfs2_xattr_set_handle(handle, inode, di_bh,
7309                                      OCFS2_XATTR_INDEX_SECURITY,
7310                                      si->name, si->value, si->value_len, 0,
7311                                      xattr_ac, data_ac);
7312 }
7313
7314 const struct xattr_handler ocfs2_xattr_security_handler = {
7315         .prefix = XATTR_SECURITY_PREFIX,
7316         .get    = ocfs2_xattr_security_get,
7317         .set    = ocfs2_xattr_security_set,
7318 };
7319
7320 /*
7321  * 'trusted' attributes support
7322  */
7323 static int ocfs2_xattr_trusted_get(const struct xattr_handler *handler,
7324                                    struct dentry *unused, struct inode *inode,
7325                                    const char *name, void *buffer, size_t size)
7326 {
7327         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED,
7328                                name, buffer, size);
7329 }
7330
7331 static int ocfs2_xattr_trusted_set(const struct xattr_handler *handler,
7332                                    struct dentry *unused, struct inode *inode,
7333                                    const char *name, const void *value,
7334                                    size_t size, int flags)
7335 {
7336         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED,
7337                                name, value, size, flags);
7338 }
7339
7340 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7341         .prefix = XATTR_TRUSTED_PREFIX,
7342         .get    = ocfs2_xattr_trusted_get,
7343         .set    = ocfs2_xattr_trusted_set,
7344 };
7345
7346 /*
7347  * 'user' attributes support
7348  */
7349 static int ocfs2_xattr_user_get(const struct xattr_handler *handler,
7350                                 struct dentry *unused, struct inode *inode,
7351                                 const char *name, void *buffer, size_t size)
7352 {
7353         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7354
7355         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7356                 return -EOPNOTSUPP;
7357         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
7358                                buffer, size);
7359 }
7360
7361 static int ocfs2_xattr_user_set(const struct xattr_handler *handler,
7362                                 struct dentry *unused, struct inode *inode,
7363                                 const char *name, const void *value,
7364                                 size_t size, int flags)
7365 {
7366         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7367
7368         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7369                 return -EOPNOTSUPP;
7370
7371         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER,
7372                                name, value, size, flags);
7373 }
7374
7375 const struct xattr_handler ocfs2_xattr_user_handler = {
7376         .prefix = XATTR_USER_PREFIX,
7377         .get    = ocfs2_xattr_user_get,
7378         .set    = ocfs2_xattr_user_set,
7379 };