fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  11  *
  12  * This program is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU General Public
  14  * License version 2 as published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  */
  21
  22 #include <linux/capability.h>
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/uio.h>
  29 #include <linux/sched.h>
  30 #include <linux/splice.h>
  31 #include <linux/mount.h>
  32 #include <linux/writeback.h>
  33 #include <linux/falloc.h>
  34 #include <linux/sort.h>
  35 #include <linux/init.h>
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/security.h>
  39
  40 #include <cluster/masklog.h>
  41
  42 #include "ocfs2.h"
  43 #include "alloc.h"
  44 #include "blockcheck.h"
  45 #include "dlmglue.h"
  46 #include "file.h"
  47 #include "symlink.h"
  48 #include "sysfile.h"
  49 #include "inode.h"
  50 #include "journal.h"
  51 #include "ocfs2_fs.h"
  52 #include "suballoc.h"
  53 #include "uptodate.h"
  54 #include "buffer_head_io.h"
  55 #include "super.h"
  56 #include "xattr.h"
  57 #include "refcounttree.h"
  58 #include "acl.h"
  59 #include "ocfs2_trace.h"
  60
  61 struct ocfs2_xattr_def_value_root {
  62         struct ocfs2_xattr_value_root   xv;
  63         struct ocfs2_extent_rec         er;
  64 };
  65
  66 struct ocfs2_xattr_bucket {
  67         /* The inode these xattrs are associated with */
  68         struct inode *bu_inode;
  69
  70         /* The actual buffers that make up the bucket */
  71         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  72
  73         /* How many blocks make up one bucket for this filesystem */
  74         int bu_blocks;
  75 };
  76
  77 struct ocfs2_xattr_set_ctxt {
  78         handle_t *handle;
  79         struct ocfs2_alloc_context *meta_ac;
  80         struct ocfs2_alloc_context *data_ac;
  81         struct ocfs2_cached_dealloc_ctxt dealloc;
  82         int set_abort;
  83 };
  84
  85 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  86 #define OCFS2_XATTR_INLINE_SIZE 80
  87 #define OCFS2_XATTR_HEADER_GAP  4
  88 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  89                                          - sizeof(struct ocfs2_xattr_header) \
  90                                          - OCFS2_XATTR_HEADER_GAP)
  91 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  92                                          - sizeof(struct ocfs2_xattr_block) \
  93                                          - sizeof(struct ocfs2_xattr_header) \
  94                                          - OCFS2_XATTR_HEADER_GAP)
  95
  96 static struct ocfs2_xattr_def_value_root def_xv = {
  97         .xv.xr_list.l_count = cpu_to_le16(1),
  98 };
  99
 100 const struct xattr_handler *ocfs2_xattr_handlers[] = {
 101         &ocfs2_xattr_user_handler,
 102         &ocfs2_xattr_acl_access_handler,
 103         &ocfs2_xattr_acl_default_handler,
 104         &ocfs2_xattr_trusted_handler,
 105         &ocfs2_xattr_security_handler,
 106         NULL
 107 };
 108
 109 static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 110         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 111         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 112                                         = &ocfs2_xattr_acl_access_handler,
 113         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 114                                         = &ocfs2_xattr_acl_default_handler,
 115         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 116         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 117 };
 118
 119 struct ocfs2_xattr_info {
 120         int             xi_name_index;
 121         const char      *xi_name;
 122         int             xi_name_len;
 123         const void      *xi_value;
 124         size_t          xi_value_len;
 125 };
 126
 127 struct ocfs2_xattr_search {
 128         struct buffer_head *inode_bh;
 129         /*
 130          * xattr_bh point to the block buffer head which has extended attribute
 131          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 132          */
 133         struct buffer_head *xattr_bh;
 134         struct ocfs2_xattr_header *header;
 135         struct ocfs2_xattr_bucket *bucket;
 136         void *base;
 137         void *end;
 138         struct ocfs2_xattr_entry *here;
 139         int not_found;
 140 };
 141
 142 /* Operations on struct ocfs2_xa_entry */
 143 struct ocfs2_xa_loc;
 144 struct ocfs2_xa_loc_operations {
 145         /*
 146          * Journal functions
 147          */
 148         int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
 149                                   int type);
 150         void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
 151
 152         /*
 153          * Return a pointer to the appropriate buffer in loc->xl_storage
 154          * at the given offset from loc->xl_header.
 155          */
 156         void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
 157
 158         /* Can we reuse the existing entry for the new value? */
 159         int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
 160                              struct ocfs2_xattr_info *xi);
 161
 162         /* How much space is needed for the new value? */
 163         int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
 164                                struct ocfs2_xattr_info *xi);
 165
 166         /*
 167          * Return the offset of the first name+value pair.  This is
 168          * the start of our downward-filling free space.
 169          */
 170         int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
 171
 172         /*
 173          * Remove the name+value at this location.  Do whatever is
 174          * appropriate with the remaining name+value pairs.
 175          */
 176         void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
 177
 178         /* Fill xl_entry with a new entry */
 179         void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
 180
 181         /* Add name+value storage to an entry */
 182         void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
 183
 184         /*
 185          * Initialize the value buf's access and bh fields for this entry.
 186          * ocfs2_xa_fill_value_buf() will handle the xv pointer.
 187          */
 188         void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
 189                                    struct ocfs2_xattr_value_buf *vb);
 190 };
 191
 192 /*
 193  * Describes an xattr entry location.  This is a memory structure
 194  * tracking the on-disk structure.
 195  */
 196 struct ocfs2_xa_loc {
 197         /* This xattr belongs to this inode */
 198         struct inode *xl_inode;
 199
 200         /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
 201         struct ocfs2_xattr_header *xl_header;
 202
 203         /* Bytes from xl_header to the end of the storage */
 204         int xl_size;
 205
 206         /*
 207          * The ocfs2_xattr_entry this location describes.  If this is
 208          * NULL, this location describes the on-disk structure where it
 209          * would have been.
 210          */
 211         struct ocfs2_xattr_entry *xl_entry;
 212
 213         /*
 214          * Internal housekeeping
 215          */
 216
 217         /* Buffer(s) containing this entry */
 218         void *xl_storage;
 219
 220         /* Operations on the storage backing this location */
 221         const struct ocfs2_xa_loc_operations *xl_ops;
 222 };
 223
 224 /*
 225  * Convenience functions to calculate how much space is needed for a
 226  * given name+value pair
 227  */
 228 static int namevalue_size(int name_len, uint64_t value_len)
 229 {
 230         if (value_len > OCFS2_XATTR_INLINE_SIZE)
 231                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 232         else
 233                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 234 }
 235
 236 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
 237 {
 238         return namevalue_size(xi->xi_name_len, xi->xi_value_len);
 239 }
 240
 241 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
 242 {
 243         u64 value_len = le64_to_cpu(xe->xe_value_size);
 244
 245         BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
 246                ocfs2_xattr_is_local(xe));
 247         return namevalue_size(xe->xe_name_len, value_len);
 248 }
 249
 250
 251 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 252                                              struct ocfs2_xattr_header *xh,
 253                                              int index,
 254                                              int *block_off,
 255                                              int *new_offset);
 256
 257 static int ocfs2_xattr_block_find(struct inode *inode,
 258                                   int name_index,
 259                                   const char *name,
 260                                   struct ocfs2_xattr_search *xs);
 261 static int ocfs2_xattr_index_block_find(struct inode *inode,
 262                                         struct buffer_head *root_bh,
 263                                         int name_index,
 264                                         const char *name,
 265                                         struct ocfs2_xattr_search *xs);
 266
 267 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 268                                         struct buffer_head *blk_bh,
 269                                         char *buffer,
 270                                         size_t buffer_size);
 271
 272 static int ocfs2_xattr_create_index_block(struct inode *inode,
 273                                           struct ocfs2_xattr_search *xs,
 274                                           struct ocfs2_xattr_set_ctxt *ctxt);
 275
 276 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 277                                              struct ocfs2_xattr_info *xi,
 278                                              struct ocfs2_xattr_search *xs,
 279                                              struct ocfs2_xattr_set_ctxt *ctxt);
 280
 281 typedef int (xattr_tree_rec_func)(struct inode *inode,
 282                                   struct buffer_head *root_bh,
 283                                   u64 blkno, u32 cpos, u32 len, void *para);
 284 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
 285                                            struct buffer_head *root_bh,
 286                                            xattr_tree_rec_func *rec_func,
 287                                            void *para);
 288 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
 289                                         struct ocfs2_xattr_bucket *bucket,
 290                                         void *para);
 291 static int ocfs2_rm_xattr_cluster(struct inode *inode,
 292                                   struct buffer_head *root_bh,
 293                                   u64 blkno,
 294                                   u32 cpos,
 295                                   u32 len,
 296                                   void *para);
 297
 298 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 299                                   u64 src_blk, u64 last_blk, u64 to_blk,
 300                                   unsigned int start_bucket,
 301                                   u32 *first_hash);
 302 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
 303                                         struct ocfs2_dinode *di,
 304                                         struct ocfs2_xattr_info *xi,
 305                                         struct ocfs2_xattr_search *xis,
 306                                         struct ocfs2_xattr_search *xbs,
 307                                         struct ocfs2_refcount_tree **ref_tree,
 308                                         int *meta_need,
 309                                         int *credits);
 310 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
 311                                            struct ocfs2_xattr_bucket *bucket,
 312                                            int offset,
 313                                            struct ocfs2_xattr_value_root **xv,
 314                                            struct buffer_head **bh);
 315
 316 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 317 {
 318         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 319 }
 320
 321 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 322 {
 323         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 324 }
 325
 326 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 327 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 328 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 329
 330 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 331 {
 332         struct ocfs2_xattr_bucket *bucket;
 333         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 334
 335         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 336
 337         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 338         if (bucket) {
 339                 bucket->bu_inode = inode;
 340                 bucket->bu_blocks = blks;
 341         }
 342
 343         return bucket;
 344 }
 345
 346 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 347 {
 348         int i;
 349
 350         for (i = 0; i < bucket->bu_blocks; i++) {
 351                 brelse(bucket->bu_bhs[i]);
 352                 bucket->bu_bhs[i] = NULL;
 353         }
 354 }
 355
 356 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 357 {
 358         if (bucket) {
 359                 ocfs2_xattr_bucket_relse(bucket);
 360                 bucket->bu_inode = NULL;
 361                 kfree(bucket);
 362         }
 363 }
 364
 365 /*
 366  * A bucket that has never been written to disk doesn't need to be
 367  * read.  We just need the buffer_heads.  Don't call this for
 368  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 369  * them fully.
 370  */
 371 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 372                                    u64 xb_blkno)
 373 {
 374         int i, rc = 0;
 375
 376         for (i = 0; i < bucket->bu_blocks; i++) {
 377                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 378                                               xb_blkno + i);
 379                 if (!bucket->bu_bhs[i]) {
 380                         rc = -EIO;
 381                         mlog_errno(rc);
 382                         break;
 383                 }
 384
 385                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 386                                            bucket->bu_bhs[i]))
 387                         ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 388                                                       bucket->bu_bhs[i]);
 389         }
 390
 391         if (rc)
 392                 ocfs2_xattr_bucket_relse(bucket);
 393         return rc;
 394 }
 395
 396 /* Read the xattr bucket at xb_blkno */
 397 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 398                                    u64 xb_blkno)
 399 {
 400         int rc;
 401
 402         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 403                                bucket->bu_blocks, bucket->bu_bhs, 0,
 404                                NULL);
 405         if (!rc) {
 406                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 407                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 408                                                  bucket->bu_bhs,
 409                                                  bucket->bu_blocks,
 410                                                  &bucket_xh(bucket)->xh_check);
 411                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 412                 if (rc)
 413                         mlog_errno(rc);
 414         }
 415
 416         if (rc)
 417                 ocfs2_xattr_bucket_relse(bucket);
 418         return rc;
 419 }
 420
 421 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 422                                              struct ocfs2_xattr_bucket *bucket,
 423                                              int type)
 424 {
 425         int i, rc = 0;
 426
 427         for (i = 0; i < bucket->bu_blocks; i++) {
 428                 rc = ocfs2_journal_access(handle,
 429                                           INODE_CACHE(bucket->bu_inode),
 430                                           bucket->bu_bhs[i], type);
 431                 if (rc) {
 432                         mlog_errno(rc);
 433                         break;
 434                 }
 435         }
 436
 437         return rc;
 438 }
 439
 440 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 441                                              struct ocfs2_xattr_bucket *bucket)
 442 {
 443         int i;
 444
 445         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 446         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 447                                    bucket->bu_bhs, bucket->bu_blocks,
 448                                    &bucket_xh(bucket)->xh_check);
 449         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 450
 451         for (i = 0; i < bucket->bu_blocks; i++)
 452                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 453 }
 454
 455 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 456                                          struct ocfs2_xattr_bucket *src)
 457 {
 458         int i;
 459         int blocksize = src->bu_inode->i_sb->s_blocksize;
 460
 461         BUG_ON(dest->bu_blocks != src->bu_blocks);
 462         BUG_ON(dest->bu_inode != src->bu_inode);
 463
 464         for (i = 0; i < src->bu_blocks; i++) {
 465                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 466                        blocksize);
 467         }
 468 }
 469
 470 static int ocfs2_validate_xattr_block(struct super_block *sb,
 471                                       struct buffer_head *bh)
 472 {
 473         int rc;
 474         struct ocfs2_xattr_block *xb =
 475                 (struct ocfs2_xattr_block *)bh->b_data;
 476
 477         trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
 478
 479         BUG_ON(!buffer_uptodate(bh));
 480
 481         /*
 482          * If the ecc fails, we return the error but otherwise
 483          * leave the filesystem running.  We know any error is
 484          * local to this block.
 485          */
 486         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 487         if (rc)
 488                 return rc;
 489
 490         /*
 491          * Errors after here are fatal
 492          */
 493
 494         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 495                 ocfs2_error(sb,
 496                             "Extended attribute block #%llu has bad "
 497                             "signature %.*s",
 498                             (unsigned long long)bh->b_blocknr, 7,
 499                             xb->xb_signature);
 500                 return -EINVAL;
 501         }
 502
 503         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 504                 ocfs2_error(sb,
 505                             "Extended attribute block #%llu has an "
 506                             "invalid xb_blkno of %llu",
 507                             (unsigned long long)bh->b_blocknr,
 508                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 509                 return -EINVAL;
 510         }
 511
 512         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 513                 ocfs2_error(sb,
 514                             "Extended attribute block #%llu has an invalid "
 515                             "xb_fs_generation of #%u",
 516                             (unsigned long long)bh->b_blocknr,
 517                             le32_to_cpu(xb->xb_fs_generation));
 518                 return -EINVAL;
 519         }
 520
 521         return 0;
 522 }
 523
 524 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 525                                   struct buffer_head **bh)
 526 {
 527         int rc;
 528         struct buffer_head *tmp = *bh;
 529
 530         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 531                               ocfs2_validate_xattr_block);
 532
 533         /* If ocfs2_read_block() got us a new bh, pass it up. */
 534         if (!rc && !*bh)
 535                 *bh = tmp;
 536
 537         return rc;
 538 }
 539
 540 static inline const char *ocfs2_xattr_prefix(int name_index)
 541 {
 542         const struct xattr_handler *handler = NULL;
 543
 544         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 545                 handler = ocfs2_xattr_handler_map[name_index];
 546
 547         return handler ? handler->prefix : NULL;
 548 }
 549
 550 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 551                                  const char *name,
 552                                  int name_len)
 553 {
 554         /* Get hash value of uuid from super block */
 555         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 556         int i;
 557
 558         /* hash extended attribute name */
 559         for (i = 0; i < name_len; i++) {
 560                 hash = (hash << OCFS2_HASH_SHIFT) ^
 561                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 562                        *name++;
 563         }
 564
 565         return hash;
 566 }
 567
 568 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 569 {
 570         return namevalue_size(name_len, value_len) +
 571                 sizeof(struct ocfs2_xattr_entry);
 572 }
 573
 574 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 575 {
 576         return namevalue_size_xi(xi) +
 577                 sizeof(struct ocfs2_xattr_entry);
 578 }
 579
 580 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 581 {
 582         return namevalue_size_xe(xe) +
 583                 sizeof(struct ocfs2_xattr_entry);
 584 }
 585
 586 int ocfs2_calc_security_init(struct inode *dir,
 587                              struct ocfs2_security_xattr_info *si,
 588                              int *want_clusters,
 589                              int *xattr_credits,
 590                              struct ocfs2_alloc_context **xattr_ac)
 591 {
 592         int ret = 0;
 593         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 594         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 595                                                  si->value_len);
 596
 597         /*
 598          * The max space of security xattr taken inline is
 599          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 600          * So reserve one metadata block for it is ok.
 601          */
 602         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 603             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 604                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 605                 if (ret) {
 606                         mlog_errno(ret);
 607                         return ret;
 608                 }
 609                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 610         }
 611
 612         /* reserve clusters for xattr value which will be set in B tree*/
 613         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 614                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 615                                                             si->value_len);
 616
 617                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 618                                                            new_clusters);
 619                 *want_clusters += new_clusters;
 620         }
 621         return ret;
 622 }
 623
 624 int ocfs2_calc_xattr_init(struct inode *dir,
 625                           struct buffer_head *dir_bh,
 626                           umode_t mode,
 627                           struct ocfs2_security_xattr_info *si,
 628                           int *want_clusters,
 629                           int *xattr_credits,
 630                           int *want_meta)
 631 {
 632         int ret = 0;
 633         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 634         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 635
 636         if (si->enable)
 637                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 638                                                      si->value_len);
 639
 640         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 641                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 642                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 643                                         "", NULL, 0);
 644                 if (acl_len > 0) {
 645                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 646                         if (S_ISDIR(mode))
 647                                 a_size <<= 1;
 648                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 649                         mlog_errno(ret);
 650                         return ret;
 651                 }
 652         }
 653
 654         if (!(s_size + a_size))
 655                 return ret;
 656
 657         /*
 658          * The max space of security xattr taken inline is
 659          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 660          * The max space of acl xattr taken inline is
 661          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 662          * when blocksize = 512, may reserve one more cluser for
 663          * xattr bucket, otherwise reserve one metadata block
 664          * for them is ok.
 665          * If this is a new directory with inline data,
 666          * we choose to reserve the entire inline area for
 667          * directory contents and force an external xattr block.
 668          */
 669         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 670             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 671             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 672                 *want_meta = *want_meta + 1;
 673                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 674         }
 675
 676         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 677             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 678                 *want_clusters += 1;
 679                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 680         }
 681
 682         /*
 683          * reserve credits and clusters for xattrs which has large value
 684          * and have to be set outside
 685          */
 686         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 687                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 688                                                         si->value_len);
 689                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 690                                                            new_clusters);
 691                 *want_clusters += new_clusters;
 692         }
 693         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 694             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 695                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 696                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 697                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 698                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 699                                                            new_clusters);
 700                 *want_clusters += new_clusters;
 701         }
 702
 703         return ret;
 704 }
 705
 706 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 707                                          u32 clusters_to_add,
 708                                          struct ocfs2_xattr_value_buf *vb,
 709                                          struct ocfs2_xattr_set_ctxt *ctxt)
 710 {
 711         int status = 0, credits;
 712         handle_t *handle = ctxt->handle;
 713         enum ocfs2_alloc_restarted why;
 714         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 715         struct ocfs2_extent_tree et;
 716
 717         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 718
 719         while (clusters_to_add) {
 720                 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 721
 722                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 723                                        OCFS2_JOURNAL_ACCESS_WRITE);
 724                 if (status < 0) {
 725                         mlog_errno(status);
 726                         break;
 727                 }
 728
 729                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 730                 status = ocfs2_add_clusters_in_btree(handle,
 731                                                      &et,
 732                                                      &logical_start,
 733                                                      clusters_to_add,
 734                                                      0,
 735                                                      ctxt->data_ac,
 736                                                      ctxt->meta_ac,
 737                                                      &why);
 738                 if ((status < 0) && (status != -EAGAIN)) {
 739                         if (status != -ENOSPC)
 740                                 mlog_errno(status);
 741                         break;
 742                 }
 743
 744                 ocfs2_journal_dirty(handle, vb->vb_bh);
 745
 746                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 747                                          prev_clusters;
 748
 749                 if (why != RESTART_NONE && clusters_to_add) {
 750                         /*
 751                          * We can only fail in case the alloc file doesn't give
 752                          * up enough clusters.
 753                          */
 754                         BUG_ON(why == RESTART_META);
 755
 756                         credits = ocfs2_calc_extend_credits(inode->i_sb,
 757                                                             &vb->vb_xv->xr_list,
 758                                                             clusters_to_add);
 759                         status = ocfs2_extend_trans(handle, credits);
 760                         if (status < 0) {
 761                                 status = -ENOMEM;
 762                                 mlog_errno(status);
 763                                 break;
 764                         }
 765                 }
 766         }
 767
 768         return status;
 769 }
 770
 771 static int __ocfs2_remove_xattr_range(struct inode *inode,
 772                                       struct ocfs2_xattr_value_buf *vb,
 773                                       u32 cpos, u32 phys_cpos, u32 len,
 774                                       unsigned int ext_flags,
 775                                       struct ocfs2_xattr_set_ctxt *ctxt)
 776 {
 777         int ret;
 778         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 779         handle_t *handle = ctxt->handle;
 780         struct ocfs2_extent_tree et;
 781
 782         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 783
 784         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 785                             OCFS2_JOURNAL_ACCESS_WRITE);
 786         if (ret) {
 787                 mlog_errno(ret);
 788                 goto out;
 789         }
 790
 791         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 792                                   &ctxt->dealloc);
 793         if (ret) {
 794                 mlog_errno(ret);
 795                 goto out;
 796         }
 797
 798         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 799         ocfs2_journal_dirty(handle, vb->vb_bh);
 800
 801         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 802                 ret = ocfs2_decrease_refcount(inode, handle,
 803                                         ocfs2_blocks_to_clusters(inode->i_sb,
 804                                                                  phys_blkno),
 805                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 806         else
 807                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 808                                                   phys_blkno, len);
 809         if (ret)
 810                 mlog_errno(ret);
 811
 812 out:
 813         return ret;
 814 }
 815
 816 static int ocfs2_xattr_shrink_size(struct inode *inode,
 817                                    u32 old_clusters,
 818                                    u32 new_clusters,
 819                                    struct ocfs2_xattr_value_buf *vb,
 820                                    struct ocfs2_xattr_set_ctxt *ctxt)
 821 {
 822         int ret = 0;
 823         unsigned int ext_flags;
 824         u32 trunc_len, cpos, phys_cpos, alloc_size;
 825         u64 block;
 826
 827         if (old_clusters <= new_clusters)
 828                 return 0;
 829
 830         cpos = new_clusters;
 831         trunc_len = old_clusters - new_clusters;
 832         while (trunc_len) {
 833                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 834                                                &alloc_size,
 835                                                &vb->vb_xv->xr_list, &ext_flags);
 836                 if (ret) {
 837                         mlog_errno(ret);
 838                         goto out;
 839                 }
 840
 841                 if (alloc_size > trunc_len)
 842                         alloc_size = trunc_len;
 843
 844                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 845                                                  phys_cpos, alloc_size,
 846                                                  ext_flags, ctxt);
 847                 if (ret) {
 848                         mlog_errno(ret);
 849                         goto out;
 850                 }
 851
 852                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 853                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 854                                                        block, alloc_size);
 855                 cpos += alloc_size;
 856                 trunc_len -= alloc_size;
 857         }
 858
 859 out:
 860         return ret;
 861 }
 862
 863 static int ocfs2_xattr_value_truncate(struct inode *inode,
 864                                       struct ocfs2_xattr_value_buf *vb,
 865                                       int len,
 866                                       struct ocfs2_xattr_set_ctxt *ctxt)
 867 {
 868         int ret;
 869         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 870         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 871
 872         if (new_clusters == old_clusters)
 873                 return 0;
 874
 875         if (new_clusters > old_clusters)
 876                 ret = ocfs2_xattr_extend_allocation(inode,
 877                                                     new_clusters - old_clusters,
 878                                                     vb, ctxt);
 879         else
 880                 ret = ocfs2_xattr_shrink_size(inode,
 881                                               old_clusters, new_clusters,
 882                                               vb, ctxt);
 883
 884         return ret;
 885 }
 886
 887 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 888                                   size_t *result, const char *prefix,
 889                                   const char *name, int name_len)
 890 {
 891         char *p = buffer + *result;
 892         int prefix_len = strlen(prefix);
 893         int total_len = prefix_len + name_len + 1;
 894
 895         *result += total_len;
 896
 897         /* we are just looking for how big our buffer needs to be */
 898         if (!size)
 899                 return 0;
 900
 901         if (*result > size)
 902                 return -ERANGE;
 903
 904         memcpy(p, prefix, prefix_len);
 905         memcpy(p + prefix_len, name, name_len);
 906         p[prefix_len + name_len] = '\0';
 907
 908         return 0;
 909 }
 910
 911 static int ocfs2_xattr_list_entries(struct inode *inode,
 912                                     struct ocfs2_xattr_header *header,
 913                                     char *buffer, size_t buffer_size)
 914 {
 915         size_t result = 0;
 916         int i, type, ret;
 917         const char *prefix, *name;
 918
 919         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 920                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 921                 type = ocfs2_xattr_get_type(entry);
 922                 prefix = ocfs2_xattr_prefix(type);
 923
 924                 if (prefix) {
 925                         name = (const char *)header +
 926                                 le16_to_cpu(entry->xe_name_offset);
 927
 928                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 929                                                      &result, prefix, name,
 930                                                      entry->xe_name_len);
 931                         if (ret)
 932                                 return ret;
 933                 }
 934         }
 935
 936         return result;
 937 }
 938
 939 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 940                                          struct ocfs2_dinode *di)
 941 {
 942         struct ocfs2_xattr_header *xh;
 943         int i;
 944
 945         xh = (struct ocfs2_xattr_header *)
 946                  ((void *)di + inode->i_sb->s_blocksize -
 947                  le16_to_cpu(di->i_xattr_inline_size));
 948
 949         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 950                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 951                         return 1;
 952
 953         return 0;
 954 }
 955
 956 static int ocfs2_xattr_ibody_list(struct inode *inode,
 957                                   struct ocfs2_dinode *di,
 958                                   char *buffer,
 959                                   size_t buffer_size)
 960 {
 961         struct ocfs2_xattr_header *header = NULL;
 962         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 963         int ret = 0;
 964
 965         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 966                 return ret;
 967
 968         header = (struct ocfs2_xattr_header *)
 969                  ((void *)di + inode->i_sb->s_blocksize -
 970                  le16_to_cpu(di->i_xattr_inline_size));
 971
 972         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 973
 974         return ret;
 975 }
 976
 977 static int ocfs2_xattr_block_list(struct inode *inode,
 978                                   struct ocfs2_dinode *di,
 979                                   char *buffer,
 980                                   size_t buffer_size)
 981 {
 982         struct buffer_head *blk_bh = NULL;
 983         struct ocfs2_xattr_block *xb;
 984         int ret = 0;
 985
 986         if (!di->i_xattr_loc)
 987                 return ret;
 988
 989         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 990                                      &blk_bh);
 991         if (ret < 0) {
 992                 mlog_errno(ret);
 993                 return ret;
 994         }
 995
 996         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 997         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 998                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 999                 ret = ocfs2_xattr_list_entries(inode, header,
1000                                                buffer, buffer_size);
1001         } else
1002                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1003                                                    buffer, buffer_size);
1004
1005         brelse(blk_bh);
1006
1007         return ret;
1008 }
1009
1010 ssize_t ocfs2_listxattr(struct dentry *dentry,
1011                         char *buffer,
1012                         size_t size)
1013 {
1014         int ret = 0, i_ret = 0, b_ret = 0;
1015         struct buffer_head *di_bh = NULL;
1016         struct ocfs2_dinode *di = NULL;
1017         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1018
1019         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1020                 return -EOPNOTSUPP;
1021
1022         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1023                 return ret;
1024
1025         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1026         if (ret < 0) {
1027                 mlog_errno(ret);
1028                 return ret;
1029         }
1030
1031         di = (struct ocfs2_dinode *)di_bh->b_data;
1032
1033         down_read(&oi->ip_xattr_sem);
1034         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1035         if (i_ret < 0)
1036                 b_ret = 0;
1037         else {
1038                 if (buffer) {
1039                         buffer += i_ret;
1040                         size -= i_ret;
1041                 }
1042                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1043                                                buffer, size);
1044                 if (b_ret < 0)
1045                         i_ret = 0;
1046         }
1047         up_read(&oi->ip_xattr_sem);
1048         ocfs2_inode_unlock(dentry->d_inode, 0);
1049
1050         brelse(di_bh);
1051
1052         return i_ret + b_ret;
1053 }
1054
1055 static int ocfs2_xattr_find_entry(int name_index,
1056                                   const char *name,
1057                                   struct ocfs2_xattr_search *xs)
1058 {
1059         struct ocfs2_xattr_entry *entry;
1060         size_t name_len;
1061         int i, cmp = 1;
1062
1063         if (name == NULL)
1064                 return -EINVAL;
1065
1066         name_len = strlen(name);
1067         entry = xs->here;
1068         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1069                 cmp = name_index - ocfs2_xattr_get_type(entry);
1070                 if (!cmp)
1071                         cmp = name_len - entry->xe_name_len;
1072                 if (!cmp)
1073                         cmp = memcmp(name, (xs->base +
1074                                      le16_to_cpu(entry->xe_name_offset)),
1075                                      name_len);
1076                 if (cmp == 0)
1077                         break;
1078                 entry += 1;
1079         }
1080         xs->here = entry;
1081
1082         return cmp ? -ENODATA : 0;
1083 }
1084
1085 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1086                                          struct ocfs2_xattr_value_root *xv,
1087                                          void *buffer,
1088                                          size_t len)
1089 {
1090         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1091         u64 blkno;
1092         int i, ret = 0;
1093         size_t cplen, blocksize;
1094         struct buffer_head *bh = NULL;
1095         struct ocfs2_extent_list *el;
1096
1097         el = &xv->xr_list;
1098         clusters = le32_to_cpu(xv->xr_clusters);
1099         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1100         blocksize = inode->i_sb->s_blocksize;
1101
1102         cpos = 0;
1103         while (cpos < clusters) {
1104                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1105                                                &num_clusters, el, NULL);
1106                 if (ret) {
1107                         mlog_errno(ret);
1108                         goto out;
1109                 }
1110
1111                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1112                 /* Copy ocfs2_xattr_value */
1113                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1114                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1115                                                &bh, NULL);
1116                         if (ret) {
1117                                 mlog_errno(ret);
1118                                 goto out;
1119                         }
1120
1121                         cplen = len >= blocksize ? blocksize : len;
1122                         memcpy(buffer, bh->b_data, cplen);
1123                         len -= cplen;
1124                         buffer += cplen;
1125
1126                         brelse(bh);
1127                         bh = NULL;
1128                         if (len == 0)
1129                                 break;
1130                 }
1131                 cpos += num_clusters;
1132         }
1133 out:
1134         return ret;
1135 }
1136
1137 static int ocfs2_xattr_ibody_get(struct inode *inode,
1138                                  int name_index,
1139                                  const char *name,
1140                                  void *buffer,
1141                                  size_t buffer_size,
1142                                  struct ocfs2_xattr_search *xs)
1143 {
1144         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1145         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1146         struct ocfs2_xattr_value_root *xv;
1147         size_t size;
1148         int ret = 0;
1149
1150         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1151                 return -ENODATA;
1152
1153         xs->end = (void *)di + inode->i_sb->s_blocksize;
1154         xs->header = (struct ocfs2_xattr_header *)
1155                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1156         xs->base = (void *)xs->header;
1157         xs->here = xs->header->xh_entries;
1158
1159         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1160         if (ret)
1161                 return ret;
1162         size = le64_to_cpu(xs->here->xe_value_size);
1163         if (buffer) {
1164                 if (size > buffer_size)
1165                         return -ERANGE;
1166                 if (ocfs2_xattr_is_local(xs->here)) {
1167                         memcpy(buffer, (void *)xs->base +
1168                                le16_to_cpu(xs->here->xe_name_offset) +
1169                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1170                 } else {
1171                         xv = (struct ocfs2_xattr_value_root *)
1172                                 (xs->base + le16_to_cpu(
1173                                  xs->here->xe_name_offset) +
1174                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1175                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1176                                                             buffer, size);
1177                         if (ret < 0) {
1178                                 mlog_errno(ret);
1179                                 return ret;
1180                         }
1181                 }
1182         }
1183
1184         return size;
1185 }
1186
1187 static int ocfs2_xattr_block_get(struct inode *inode,
1188                                  int name_index,
1189                                  const char *name,
1190                                  void *buffer,
1191                                  size_t buffer_size,
1192                                  struct ocfs2_xattr_search *xs)
1193 {
1194         struct ocfs2_xattr_block *xb;
1195         struct ocfs2_xattr_value_root *xv;
1196         size_t size;
1197         int ret = -ENODATA, name_offset, name_len, i;
1198         int uninitialized_var(block_off);
1199
1200         xs->bucket = ocfs2_xattr_bucket_new(inode);
1201         if (!xs->bucket) {
1202                 ret = -ENOMEM;
1203                 mlog_errno(ret);
1204                 goto cleanup;
1205         }
1206
1207         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1208         if (ret) {
1209                 mlog_errno(ret);
1210                 goto cleanup;
1211         }
1212
1213         if (xs->not_found) {
1214                 ret = -ENODATA;
1215                 goto cleanup;
1216         }
1217
1218         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1219         size = le64_to_cpu(xs->here->xe_value_size);
1220         if (buffer) {
1221                 ret = -ERANGE;
1222                 if (size > buffer_size)
1223                         goto cleanup;
1224
1225                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1226                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1227                 i = xs->here - xs->header->xh_entries;
1228
1229                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1230                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1231                                                                 bucket_xh(xs->bucket),
1232                                                                 i,
1233                                                                 &block_off,
1234                                                                 &name_offset);
1235                         xs->base = bucket_block(xs->bucket, block_off);
1236                 }
1237                 if (ocfs2_xattr_is_local(xs->here)) {
1238                         memcpy(buffer, (void *)xs->base +
1239                                name_offset + name_len, size);
1240                 } else {
1241                         xv = (struct ocfs2_xattr_value_root *)
1242                                 (xs->base + name_offset + name_len);
1243                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1244                                                             buffer, size);
1245                         if (ret < 0) {
1246                                 mlog_errno(ret);
1247                                 goto cleanup;
1248                         }
1249                 }
1250         }
1251         ret = size;
1252 cleanup:
1253         ocfs2_xattr_bucket_free(xs->bucket);
1254
1255         brelse(xs->xattr_bh);
1256         xs->xattr_bh = NULL;
1257         return ret;
1258 }
1259
1260 int ocfs2_xattr_get_nolock(struct inode *inode,
1261                            struct buffer_head *di_bh,
1262                            int name_index,
1263                            const char *name,
1264                            void *buffer,
1265                            size_t buffer_size)
1266 {
1267         int ret;
1268         struct ocfs2_dinode *di = NULL;
1269         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1270         struct ocfs2_xattr_search xis = {
1271                 .not_found = -ENODATA,
1272         };
1273         struct ocfs2_xattr_search xbs = {
1274                 .not_found = -ENODATA,
1275         };
1276
1277         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1278                 return -EOPNOTSUPP;
1279
1280         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1281                 ret = -ENODATA;
1282
1283         xis.inode_bh = xbs.inode_bh = di_bh;
1284         di = (struct ocfs2_dinode *)di_bh->b_data;
1285
1286         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1287                                     buffer_size, &xis);
1288         if (ret == -ENODATA && di->i_xattr_loc)
1289                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1290                                             buffer_size, &xbs);
1291
1292         return ret;
1293 }
1294
1295 /* ocfs2_xattr_get()
1296  *
1297  * Copy an extended attribute into the buffer provided.
1298  * Buffer is NULL to compute the size of buffer required.
1299  */
1300 static int ocfs2_xattr_get(struct inode *inode,
1301                            int name_index,
1302                            const char *name,
1303                            void *buffer,
1304                            size_t buffer_size)
1305 {
1306         int ret;
1307         struct buffer_head *di_bh = NULL;
1308
1309         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1310         if (ret < 0) {
1311                 mlog_errno(ret);
1312                 return ret;
1313         }
1314         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1315         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1316                                      name, buffer, buffer_size);
1317         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1318
1319         ocfs2_inode_unlock(inode, 0);
1320
1321         brelse(di_bh);
1322
1323         return ret;
1324 }
1325
1326 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1327                                            handle_t *handle,
1328                                            struct ocfs2_xattr_value_buf *vb,
1329                                            const void *value,
1330                                            int value_len)
1331 {
1332         int ret = 0, i, cp_len;
1333         u16 blocksize = inode->i_sb->s_blocksize;
1334         u32 p_cluster, num_clusters;
1335         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1336         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1337         u64 blkno;
1338         struct buffer_head *bh = NULL;
1339         unsigned int ext_flags;
1340         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1341
1342         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1343
1344         while (cpos < clusters) {
1345                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1346                                                &num_clusters, &xv->xr_list,
1347                                                &ext_flags);
1348                 if (ret) {
1349                         mlog_errno(ret);
1350                         goto out;
1351                 }
1352
1353                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1354
1355                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1356
1357                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1358                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1359                                                &bh, NULL);
1360                         if (ret) {
1361                                 mlog_errno(ret);
1362                                 goto out;
1363                         }
1364
1365                         ret = ocfs2_journal_access(handle,
1366                                                    INODE_CACHE(inode),
1367                                                    bh,
1368                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1369                         if (ret < 0) {
1370                                 mlog_errno(ret);
1371                                 goto out;
1372                         }
1373
1374                         cp_len = value_len > blocksize ? blocksize : value_len;
1375                         memcpy(bh->b_data, value, cp_len);
1376                         value_len -= cp_len;
1377                         value += cp_len;
1378                         if (cp_len < blocksize)
1379                                 memset(bh->b_data + cp_len, 0,
1380                                        blocksize - cp_len);
1381
1382                         ocfs2_journal_dirty(handle, bh);
1383                         brelse(bh);
1384                         bh = NULL;
1385
1386                         /*
1387                          * XXX: do we need to empty all the following
1388                          * blocks in this cluster?
1389                          */
1390                         if (!value_len)
1391                                 break;
1392                 }
1393                 cpos += num_clusters;
1394         }
1395 out:
1396         brelse(bh);
1397
1398         return ret;
1399 }
1400
1401 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1402                                        int num_entries)
1403 {
1404         int free_space;
1405
1406         if (!needed_space)
1407                 return 0;
1408
1409         free_space = free_start -
1410                 sizeof(struct ocfs2_xattr_header) -
1411                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1412                 OCFS2_XATTR_HEADER_GAP;
1413         if (free_space < 0)
1414                 return -EIO;
1415         if (free_space < needed_space)
1416                 return -ENOSPC;
1417
1418         return 0;
1419 }
1420
1421 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1422                                    int type)
1423 {
1424         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1425 }
1426
1427 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1428 {
1429         loc->xl_ops->xlo_journal_dirty(handle, loc);
1430 }
1431
1432 /* Give a pointer into the storage for the given offset */
1433 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1434 {
1435         BUG_ON(offset >= loc->xl_size);
1436         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1437 }
1438
1439 /*
1440  * Wipe the name+value pair and allow the storage to reclaim it.  This
1441  * must be followed by either removal of the entry or a call to
1442  * ocfs2_xa_add_namevalue().
1443  */
1444 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1445 {
1446         loc->xl_ops->xlo_wipe_namevalue(loc);
1447 }
1448
1449 /*
1450  * Find lowest offset to a name+value pair.  This is the start of our
1451  * downward-growing free space.
1452  */
1453 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1454 {
1455         return loc->xl_ops->xlo_get_free_start(loc);
1456 }
1457
1458 /* Can we reuse loc->xl_entry for xi? */
1459 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1460                                     struct ocfs2_xattr_info *xi)
1461 {
1462         return loc->xl_ops->xlo_can_reuse(loc, xi);
1463 }
1464
1465 /* How much free space is needed to set the new value */
1466 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1467                                 struct ocfs2_xattr_info *xi)
1468 {
1469         return loc->xl_ops->xlo_check_space(loc, xi);
1470 }
1471
1472 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1473 {
1474         loc->xl_ops->xlo_add_entry(loc, name_hash);
1475         loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1476         /*
1477          * We can't leave the new entry's xe_name_offset at zero or
1478          * add_namevalue() will go nuts.  We set it to the size of our
1479          * storage so that it can never be less than any other entry.
1480          */
1481         loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1482 }
1483
1484 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1485                                    struct ocfs2_xattr_info *xi)
1486 {
1487         int size = namevalue_size_xi(xi);
1488         int nameval_offset;
1489         char *nameval_buf;
1490
1491         loc->xl_ops->xlo_add_namevalue(loc, size);
1492         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1493         loc->xl_entry->xe_name_len = xi->xi_name_len;
1494         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1495         ocfs2_xattr_set_local(loc->xl_entry,
1496                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1497
1498         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1499         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1500         memset(nameval_buf, 0, size);
1501         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1502 }
1503
1504 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1505                                     struct ocfs2_xattr_value_buf *vb)
1506 {
1507         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1508         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1509
1510         /* Value bufs are for value trees */
1511         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1512         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1513                (name_size + OCFS2_XATTR_ROOT_SIZE));
1514
1515         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1516         vb->vb_xv =
1517                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1518                                                         nameval_offset +
1519                                                         name_size);
1520 }
1521
1522 static int ocfs2_xa_block_journal_access(handle_t *handle,
1523                                          struct ocfs2_xa_loc *loc, int type)
1524 {
1525         struct buffer_head *bh = loc->xl_storage;
1526         ocfs2_journal_access_func access;
1527
1528         if (loc->xl_size == (bh->b_size -
1529                              offsetof(struct ocfs2_xattr_block,
1530                                       xb_attrs.xb_header)))
1531                 access = ocfs2_journal_access_xb;
1532         else
1533                 access = ocfs2_journal_access_di;
1534         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1535 }
1536
1537 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1538                                          struct ocfs2_xa_loc *loc)
1539 {
1540         struct buffer_head *bh = loc->xl_storage;
1541
1542         ocfs2_journal_dirty(handle, bh);
1543 }
1544
1545 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1546                                            int offset)
1547 {
1548         return (char *)loc->xl_header + offset;
1549 }
1550
1551 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1552                                     struct ocfs2_xattr_info *xi)
1553 {
1554         /*
1555          * Block storage is strict.  If the sizes aren't exact, we will
1556          * remove the old one and reinsert the new.
1557          */
1558         return namevalue_size_xe(loc->xl_entry) ==
1559                 namevalue_size_xi(xi);
1560 }
1561
1562 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1563 {
1564         struct ocfs2_xattr_header *xh = loc->xl_header;
1565         int i, count = le16_to_cpu(xh->xh_count);
1566         int offset, free_start = loc->xl_size;
1567
1568         for (i = 0; i < count; i++) {
1569                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1570                 if (offset < free_start)
1571                         free_start = offset;
1572         }
1573
1574         return free_start;
1575 }
1576
1577 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1578                                       struct ocfs2_xattr_info *xi)
1579 {
1580         int count = le16_to_cpu(loc->xl_header->xh_count);
1581         int free_start = ocfs2_xa_get_free_start(loc);
1582         int needed_space = ocfs2_xi_entry_usage(xi);
1583
1584         /*
1585          * Block storage will reclaim the original entry before inserting
1586          * the new value, so we only need the difference.  If the new
1587          * entry is smaller than the old one, we don't need anything.
1588          */
1589         if (loc->xl_entry) {
1590                 /* Don't need space if we're reusing! */
1591                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1592                         needed_space = 0;
1593                 else
1594                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1595         }
1596         if (needed_space < 0)
1597                 needed_space = 0;
1598         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1599 }
1600
1601 /*
1602  * Block storage for xattrs keeps the name+value pairs compacted.  When
1603  * we remove one, we have to shift any that preceded it towards the end.
1604  */
1605 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1606 {
1607         int i, offset;
1608         int namevalue_offset, first_namevalue_offset, namevalue_size;
1609         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1610         struct ocfs2_xattr_header *xh = loc->xl_header;
1611         int count = le16_to_cpu(xh->xh_count);
1612
1613         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1614         namevalue_size = namevalue_size_xe(entry);
1615         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1616
1617         /* Shift the name+value pairs */
1618         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1619                 (char *)xh + first_namevalue_offset,
1620                 namevalue_offset - first_namevalue_offset);
1621         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1622
1623         /* Now tell xh->xh_entries about it */
1624         for (i = 0; i < count; i++) {
1625                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1626                 if (offset <= namevalue_offset)
1627                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1628                                      namevalue_size);
1629         }
1630
1631         /*
1632          * Note that we don't update xh_free_start or xh_name_value_len
1633          * because they're not used in block-stored xattrs.
1634          */
1635 }
1636
1637 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1638 {
1639         int count = le16_to_cpu(loc->xl_header->xh_count);
1640         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1641         le16_add_cpu(&loc->xl_header->xh_count, 1);
1642         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1643 }
1644
1645 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1646 {
1647         int free_start = ocfs2_xa_get_free_start(loc);
1648
1649         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1650 }
1651
1652 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1653                                           struct ocfs2_xattr_value_buf *vb)
1654 {
1655         struct buffer_head *bh = loc->xl_storage;
1656
1657         if (loc->xl_size == (bh->b_size -
1658                              offsetof(struct ocfs2_xattr_block,
1659                                       xb_attrs.xb_header)))
1660                 vb->vb_access = ocfs2_journal_access_xb;
1661         else
1662                 vb->vb_access = ocfs2_journal_access_di;
1663         vb->vb_bh = bh;
1664 }
1665
1666 /*
1667  * Operations for xattrs stored in blocks.  This includes inline inode
1668  * storage and unindexed ocfs2_xattr_blocks.
1669  */
1670 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1671         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1672         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1673         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1674         .xlo_check_space        = ocfs2_xa_block_check_space,
1675         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1676         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1677         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1678         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1679         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1680         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1681 };
1682
1683 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1684                                           struct ocfs2_xa_loc *loc, int type)
1685 {
1686         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1687
1688         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1689 }
1690
1691 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1692                                           struct ocfs2_xa_loc *loc)
1693 {
1694         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1695
1696         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1697 }
1698
1699 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1700                                             int offset)
1701 {
1702         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1703         int block, block_offset;
1704
1705         /* The header is at the front of the bucket */
1706         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1707         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1708
1709         return bucket_block(bucket, block) + block_offset;
1710 }
1711
1712 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1713                                      struct ocfs2_xattr_info *xi)
1714 {
1715         return namevalue_size_xe(loc->xl_entry) >=
1716                 namevalue_size_xi(xi);
1717 }
1718
1719 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1720 {
1721         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1722         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1723 }
1724
1725 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1726                                          int free_start, int size)
1727 {
1728         /*
1729          * We need to make sure that the name+value pair fits within
1730          * one block.
1731          */
1732         if (((free_start - size) >> sb->s_blocksize_bits) !=
1733             ((free_start - 1) >> sb->s_blocksize_bits))
1734                 free_start -= free_start % sb->s_blocksize;
1735
1736         return free_start;
1737 }
1738
1739 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1740                                        struct ocfs2_xattr_info *xi)
1741 {
1742         int rc;
1743         int count = le16_to_cpu(loc->xl_header->xh_count);
1744         int free_start = ocfs2_xa_get_free_start(loc);
1745         int needed_space = ocfs2_xi_entry_usage(xi);
1746         int size = namevalue_size_xi(xi);
1747         struct super_block *sb = loc->xl_inode->i_sb;
1748
1749         /*
1750          * Bucket storage does not reclaim name+value pairs it cannot
1751          * reuse.  They live as holes until the bucket fills, and then
1752          * the bucket is defragmented.  However, the bucket can reclaim
1753          * the ocfs2_xattr_entry.
1754          */
1755         if (loc->xl_entry) {
1756                 /* Don't need space if we're reusing! */
1757                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1758                         needed_space = 0;
1759                 else
1760                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1761         }
1762         BUG_ON(needed_space < 0);
1763
1764         if (free_start < size) {
1765                 if (needed_space)
1766                         return -ENOSPC;
1767         } else {
1768                 /*
1769                  * First we check if it would fit in the first place.
1770                  * Below, we align the free start to a block.  This may
1771                  * slide us below the minimum gap.  By checking unaligned
1772                  * first, we avoid that error.
1773                  */
1774                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1775                                                  count);
1776                 if (rc)
1777                         return rc;
1778                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1779                                                            size);
1780         }
1781         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1782 }
1783
1784 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1785 {
1786         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1787                      -namevalue_size_xe(loc->xl_entry));
1788 }
1789
1790 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1791 {
1792         struct ocfs2_xattr_header *xh = loc->xl_header;
1793         int count = le16_to_cpu(xh->xh_count);
1794         int low = 0, high = count - 1, tmp;
1795         struct ocfs2_xattr_entry *tmp_xe;
1796
1797         /*
1798          * We keep buckets sorted by name_hash, so we need to find
1799          * our insert place.
1800          */
1801         while (low <= high && count) {
1802                 tmp = (low + high) / 2;
1803                 tmp_xe = &xh->xh_entries[tmp];
1804
1805                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1806                         low = tmp + 1;
1807                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1808                         high = tmp - 1;
1809                 else {
1810                         low = tmp;
1811                         break;
1812                 }
1813         }
1814
1815         if (low != count)
1816                 memmove(&xh->xh_entries[low + 1],
1817                         &xh->xh_entries[low],
1818                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1819
1820         le16_add_cpu(&xh->xh_count, 1);
1821         loc->xl_entry = &xh->xh_entries[low];
1822         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1823 }
1824
1825 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1826 {
1827         int free_start = ocfs2_xa_get_free_start(loc);
1828         struct ocfs2_xattr_header *xh = loc->xl_header;
1829         struct super_block *sb = loc->xl_inode->i_sb;
1830         int nameval_offset;
1831
1832         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1833         nameval_offset = free_start - size;
1834         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1835         xh->xh_free_start = cpu_to_le16(nameval_offset);
1836         le16_add_cpu(&xh->xh_name_value_len, size);
1837
1838 }
1839
1840 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1841                                            struct ocfs2_xattr_value_buf *vb)
1842 {
1843         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1844         struct super_block *sb = loc->xl_inode->i_sb;
1845         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1846         int size = namevalue_size_xe(loc->xl_entry);
1847         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1848
1849         /* Values are not allowed to straddle block boundaries */
1850         BUG_ON(block_offset !=
1851                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1852         /* We expect the bucket to be filled in */
1853         BUG_ON(!bucket->bu_bhs[block_offset]);
1854
1855         vb->vb_access = ocfs2_journal_access;
1856         vb->vb_bh = bucket->bu_bhs[block_offset];
1857 }
1858
1859 /* Operations for xattrs stored in buckets. */
1860 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1861         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1862         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1863         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1864         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1865         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1866         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1867         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1868         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1869         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1870         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1871 };
1872
1873 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1874 {
1875         struct ocfs2_xattr_value_buf vb;
1876
1877         if (ocfs2_xattr_is_local(loc->xl_entry))
1878                 return 0;
1879
1880         ocfs2_xa_fill_value_buf(loc, &vb);
1881         return le32_to_cpu(vb.vb_xv->xr_clusters);
1882 }
1883
1884 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1885                                    struct ocfs2_xattr_set_ctxt *ctxt)
1886 {
1887         int trunc_rc, access_rc;
1888         struct ocfs2_xattr_value_buf vb;
1889
1890         ocfs2_xa_fill_value_buf(loc, &vb);
1891         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1892                                               ctxt);
1893
1894         /*
1895          * The caller of ocfs2_xa_value_truncate() has already called
1896          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1897          * calls ocfs2_extend_trans().  This may commit the previous
1898          * transaction and open a new one.  If this is a bucket, truncate
1899          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1900          * the caller is expecting to dirty the entire bucket.  So we must
1901          * reset the journal work.  We do this even if truncate has failed,
1902          * as it could have failed after committing the extend.
1903          */
1904         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1905                                             OCFS2_JOURNAL_ACCESS_WRITE);
1906
1907         /* Errors in truncate take precedence */
1908         return trunc_rc ? trunc_rc : access_rc;
1909 }
1910
1911 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1912 {
1913         int index, count;
1914         struct ocfs2_xattr_header *xh = loc->xl_header;
1915         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1916
1917         ocfs2_xa_wipe_namevalue(loc);
1918         loc->xl_entry = NULL;
1919
1920         le16_add_cpu(&xh->xh_count, -1);
1921         count = le16_to_cpu(xh->xh_count);
1922
1923         /*
1924          * Only zero out the entry if there are more remaining.  This is
1925          * important for an empty bucket, as it keeps track of the
1926          * bucket's hash value.  It doesn't hurt empty block storage.
1927          */
1928         if (count) {
1929                 index = ((char *)entry - (char *)&xh->xh_entries) /
1930                         sizeof(struct ocfs2_xattr_entry);
1931                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1932                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1933                 memset(&xh->xh_entries[count], 0,
1934                        sizeof(struct ocfs2_xattr_entry));
1935         }
1936 }
1937
1938 /*
1939  * If we have a problem adjusting the size of an external value during
1940  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1941  * in an intermediate state.  For example, the value may be partially
1942  * truncated.
1943  *
1944  * If the value tree hasn't changed, the extend/truncate went nowhere.
1945  * We have nothing to do.  The caller can treat it as a straight error.
1946  *
1947  * If the value tree got partially truncated, we now have a corrupted
1948  * extended attribute.  We're going to wipe its entry and leak the
1949  * clusters.  Better to leak some storage than leave a corrupt entry.
1950  *
1951  * If the value tree grew, it obviously didn't grow enough for the
1952  * new entry.  We're not going to try and reclaim those clusters either.
1953  * If there was already an external value there (orig_clusters != 0),
1954  * the new clusters are attached safely and we can just leave the old
1955  * value in place.  If there was no external value there, we remove
1956  * the entry.
1957  *
1958  * This way, the xattr block we store in the journal will be consistent.
1959  * If the size change broke because of the journal, no changes will hit
1960  * disk anyway.
1961  */
1962 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1963                                             const char *what,
1964                                             unsigned int orig_clusters)
1965 {
1966         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1967         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1968                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1969
1970         if (new_clusters < orig_clusters) {
1971                 mlog(ML_ERROR,
1972                      "Partial truncate while %s xattr %.*s.  Leaking "
1973                      "%u clusters and removing the entry\n",
1974                      what, loc->xl_entry->xe_name_len, nameval_buf,
1975                      orig_clusters - new_clusters);
1976                 ocfs2_xa_remove_entry(loc);
1977         } else if (!orig_clusters) {
1978                 mlog(ML_ERROR,
1979                      "Unable to allocate an external value for xattr "
1980                      "%.*s safely.  Leaking %u clusters and removing the "
1981                      "entry\n",
1982                      loc->xl_entry->xe_name_len, nameval_buf,
1983                      new_clusters - orig_clusters);
1984                 ocfs2_xa_remove_entry(loc);
1985         } else if (new_clusters > orig_clusters)
1986                 mlog(ML_ERROR,
1987                      "Unable to grow xattr %.*s safely.  %u new clusters "
1988                      "have been added, but the value will not be "
1989                      "modified\n",
1990                      loc->xl_entry->xe_name_len, nameval_buf,
1991                      new_clusters - orig_clusters);
1992 }
1993
1994 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
1995                            struct ocfs2_xattr_set_ctxt *ctxt)
1996 {
1997         int rc = 0;
1998         unsigned int orig_clusters;
1999
2000         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2001                 orig_clusters = ocfs2_xa_value_clusters(loc);
2002                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2003                 if (rc) {
2004                         mlog_errno(rc);
2005                         /*
2006                          * Since this is remove, we can return 0 if
2007                          * ocfs2_xa_cleanup_value_truncate() is going to
2008                          * wipe the entry anyway.  So we check the
2009                          * cluster count as well.
2010                          */
2011                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2012                                 rc = 0;
2013                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2014                                                         orig_clusters);
2015                         if (rc)
2016                                 goto out;
2017                 }
2018         }
2019
2020         ocfs2_xa_remove_entry(loc);
2021
2022 out:
2023         return rc;
2024 }
2025
2026 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2027 {
2028         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2029         char *nameval_buf;
2030
2031         nameval_buf = ocfs2_xa_offset_pointer(loc,
2032                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2033         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2034 }
2035
2036 /*
2037  * Take an existing entry and make it ready for the new value.  This
2038  * won't allocate space, but it may free space.  It should be ready for
2039  * ocfs2_xa_prepare_entry() to finish the work.
2040  */
2041 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2042                                 struct ocfs2_xattr_info *xi,
2043                                 struct ocfs2_xattr_set_ctxt *ctxt)
2044 {
2045         int rc = 0;
2046         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2047         unsigned int orig_clusters;
2048         char *nameval_buf;
2049         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2050         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2051
2052         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2053                name_size);
2054
2055         nameval_buf = ocfs2_xa_offset_pointer(loc,
2056                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2057         if (xe_local) {
2058                 memset(nameval_buf + name_size, 0,
2059                        namevalue_size_xe(loc->xl_entry) - name_size);
2060                 if (!xi_local)
2061                         ocfs2_xa_install_value_root(loc);
2062         } else {
2063                 orig_clusters = ocfs2_xa_value_clusters(loc);
2064                 if (xi_local) {
2065                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2066                         if (rc < 0)
2067                                 mlog_errno(rc);
2068                         else
2069                                 memset(nameval_buf + name_size, 0,
2070                                        namevalue_size_xe(loc->xl_entry) -
2071                                        name_size);
2072                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2073                            xi->xi_value_len) {
2074                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2075                                                      ctxt);
2076                         if (rc < 0)
2077                                 mlog_errno(rc);
2078                 }
2079
2080                 if (rc) {
2081                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2082                                                         orig_clusters);
2083                         goto out;
2084                 }
2085         }
2086
2087         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2088         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2089
2090 out:
2091         return rc;
2092 }
2093
2094 /*
2095  * Prepares loc->xl_entry to receive the new xattr.  This includes
2096  * properly setting up the name+value pair region.  If loc->xl_entry
2097  * already exists, it will take care of modifying it appropriately.
2098  *
2099  * Note that this modifies the data.  You did journal_access already,
2100  * right?
2101  */
2102 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2103                                   struct ocfs2_xattr_info *xi,
2104                                   u32 name_hash,
2105                                   struct ocfs2_xattr_set_ctxt *ctxt)
2106 {
2107         int rc = 0;
2108         unsigned int orig_clusters;
2109         __le64 orig_value_size = 0;
2110
2111         rc = ocfs2_xa_check_space(loc, xi);
2112         if (rc)
2113                 goto out;
2114
2115         if (loc->xl_entry) {
2116                 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2117                         orig_value_size = loc->xl_entry->xe_value_size;
2118                         rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2119                         if (rc)
2120                                 goto out;
2121                         goto alloc_value;
2122                 }
2123
2124                 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2125                         orig_clusters = ocfs2_xa_value_clusters(loc);
2126                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2127                         if (rc) {
2128                                 mlog_errno(rc);
2129                                 ocfs2_xa_cleanup_value_truncate(loc,
2130                                                                 "overwriting",
2131                                                                 orig_clusters);
2132                                 goto out;
2133                         }
2134                 }
2135                 ocfs2_xa_wipe_namevalue(loc);
2136         } else
2137                 ocfs2_xa_add_entry(loc, name_hash);
2138
2139         /*
2140          * If we get here, we have a blank entry.  Fill it.  We grow our
2141          * name+value pair back from the end.
2142          */
2143         ocfs2_xa_add_namevalue(loc, xi);
2144         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2145                 ocfs2_xa_install_value_root(loc);
2146
2147 alloc_value:
2148         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2149                 orig_clusters = ocfs2_xa_value_clusters(loc);
2150                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2151                 if (rc < 0) {
2152                         ctxt->set_abort = 1;
2153                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2154                                                         orig_clusters);
2155                         /*
2156                          * If we were growing an existing value,
2157                          * ocfs2_xa_cleanup_value_truncate() won't remove
2158                          * the entry. We need to restore the original value
2159                          * size.
2160                          */
2161                         if (loc->xl_entry) {
2162                                 BUG_ON(!orig_value_size);
2163                                 loc->xl_entry->xe_value_size = orig_value_size;
2164                         }
2165                         mlog_errno(rc);
2166                 }
2167         }
2168
2169 out:
2170         return rc;
2171 }
2172
2173 /*
2174  * Store the value portion of the name+value pair.  This will skip
2175  * values that are stored externally.  Their tree roots were set up
2176  * by ocfs2_xa_prepare_entry().
2177  */
2178 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2179                                 struct ocfs2_xattr_info *xi,
2180                                 struct ocfs2_xattr_set_ctxt *ctxt)
2181 {
2182         int rc = 0;
2183         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2184         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2185         char *nameval_buf;
2186         struct ocfs2_xattr_value_buf vb;
2187
2188         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2189         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2190                 ocfs2_xa_fill_value_buf(loc, &vb);
2191                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2192                                                      ctxt->handle, &vb,
2193                                                      xi->xi_value,
2194                                                      xi->xi_value_len);
2195         } else
2196                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2197
2198         return rc;
2199 }
2200
2201 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2202                         struct ocfs2_xattr_info *xi,
2203                         struct ocfs2_xattr_set_ctxt *ctxt)
2204 {
2205         int ret;
2206         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2207                                               xi->xi_name_len);
2208
2209         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2210                                       OCFS2_JOURNAL_ACCESS_WRITE);
2211         if (ret) {
2212                 mlog_errno(ret);
2213                 goto out;
2214         }
2215
2216         /*
2217          * From here on out, everything is going to modify the buffer a
2218          * little.  Errors are going to leave the xattr header in a
2219          * sane state.  Thus, even with errors we dirty the sucker.
2220          */
2221
2222         /* Don't worry, we are never called with !xi_value and !xl_entry */
2223         if (!xi->xi_value) {
2224                 ret = ocfs2_xa_remove(loc, ctxt);
2225                 goto out_dirty;
2226         }
2227
2228         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2229         if (ret) {
2230                 if (ret != -ENOSPC)
2231                         mlog_errno(ret);
2232                 goto out_dirty;
2233         }
2234
2235         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2236         if (ret)
2237                 mlog_errno(ret);
2238
2239 out_dirty:
2240         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2241
2242 out:
2243         return ret;
2244 }
2245
2246 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2247                                      struct inode *inode,
2248                                      struct buffer_head *bh,
2249                                      struct ocfs2_xattr_entry *entry)
2250 {
2251         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2252
2253         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2254
2255         loc->xl_inode = inode;
2256         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2257         loc->xl_storage = bh;
2258         loc->xl_entry = entry;
2259         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2260         loc->xl_header =
2261                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2262                                               loc->xl_size);
2263 }
2264
2265 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2266                                           struct inode *inode,
2267                                           struct buffer_head *bh,
2268                                           struct ocfs2_xattr_entry *entry)
2269 {
2270         struct ocfs2_xattr_block *xb =
2271                 (struct ocfs2_xattr_block *)bh->b_data;
2272
2273         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2274
2275         loc->xl_inode = inode;
2276         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2277         loc->xl_storage = bh;
2278         loc->xl_header = &(xb->xb_attrs.xb_header);
2279         loc->xl_entry = entry;
2280         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2281                                              xb_attrs.xb_header);
2282 }
2283
2284 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2285                                            struct ocfs2_xattr_bucket *bucket,
2286                                            struct ocfs2_xattr_entry *entry)
2287 {
2288         loc->xl_inode = bucket->bu_inode;
2289         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2290         loc->xl_storage = bucket;
2291         loc->xl_header = bucket_xh(bucket);
2292         loc->xl_entry = entry;
2293         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2294 }
2295
2296 /*
2297  * In xattr remove, if it is stored outside and refcounted, we may have
2298  * the chance to split the refcount tree. So need the allocators.
2299  */
2300 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2301                                         struct ocfs2_xattr_value_root *xv,
2302                                         struct ocfs2_caching_info *ref_ci,
2303                                         struct buffer_head *ref_root_bh,
2304                                         struct ocfs2_alloc_context **meta_ac,
2305                                         int *ref_credits)
2306 {
2307         int ret, meta_add = 0;
2308         u32 p_cluster, num_clusters;
2309         unsigned int ext_flags;
2310
2311         *ref_credits = 0;
2312         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2313                                        &num_clusters,
2314                                        &xv->xr_list,
2315                                        &ext_flags);
2316         if (ret) {
2317                 mlog_errno(ret);
2318                 goto out;
2319         }
2320
2321         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2322                 goto out;
2323
2324         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2325                                                  ref_root_bh, xv,
2326                                                  &meta_add, ref_credits);
2327         if (ret) {
2328                 mlog_errno(ret);
2329                 goto out;
2330         }
2331
2332         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2333                                                 meta_add, meta_ac);
2334         if (ret)
2335                 mlog_errno(ret);
2336
2337 out:
2338         return ret;
2339 }
2340
2341 static int ocfs2_remove_value_outside(struct inode*inode,
2342                                       struct ocfs2_xattr_value_buf *vb,
2343                                       struct ocfs2_xattr_header *header,
2344                                       struct ocfs2_caching_info *ref_ci,
2345                                       struct buffer_head *ref_root_bh)
2346 {
2347         int ret = 0, i, ref_credits;
2348         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2349         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2350         void *val;
2351
2352         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2353
2354         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2355                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2356
2357                 if (ocfs2_xattr_is_local(entry))
2358                         continue;
2359
2360                 val = (void *)header +
2361                         le16_to_cpu(entry->xe_name_offset);
2362                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2363                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2364
2365                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2366                                                          ref_ci, ref_root_bh,
2367                                                          &ctxt.meta_ac,
2368                                                          &ref_credits);
2369
2370                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2371                                         ocfs2_remove_extent_credits(osb->sb));
2372                 if (IS_ERR(ctxt.handle)) {
2373                         ret = PTR_ERR(ctxt.handle);
2374                         mlog_errno(ret);
2375                         break;
2376                 }
2377
2378                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2379
2380                 ocfs2_commit_trans(osb, ctxt.handle);
2381                 if (ctxt.meta_ac) {
2382                         ocfs2_free_alloc_context(ctxt.meta_ac);
2383                         ctxt.meta_ac = NULL;
2384                 }
2385
2386                 if (ret < 0) {
2387                         mlog_errno(ret);
2388                         break;
2389                 }
2390
2391         }
2392
2393         if (ctxt.meta_ac)
2394                 ocfs2_free_alloc_context(ctxt.meta_ac);
2395         ocfs2_schedule_truncate_log_flush(osb, 1);
2396         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2397         return ret;
2398 }
2399
2400 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2401                                     struct buffer_head *di_bh,
2402                                     struct ocfs2_caching_info *ref_ci,
2403                                     struct buffer_head *ref_root_bh)
2404 {
2405
2406         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2407         struct ocfs2_xattr_header *header;
2408         int ret;
2409         struct ocfs2_xattr_value_buf vb = {
2410                 .vb_bh = di_bh,
2411                 .vb_access = ocfs2_journal_access_di,
2412         };
2413
2414         header = (struct ocfs2_xattr_header *)
2415                  ((void *)di + inode->i_sb->s_blocksize -
2416                  le16_to_cpu(di->i_xattr_inline_size));
2417
2418         ret = ocfs2_remove_value_outside(inode, &vb, header,
2419                                          ref_ci, ref_root_bh);
2420
2421         return ret;
2422 }
2423
2424 struct ocfs2_rm_xattr_bucket_para {
2425         struct ocfs2_caching_info *ref_ci;
2426         struct buffer_head *ref_root_bh;
2427 };
2428
2429 static int ocfs2_xattr_block_remove(struct inode *inode,
2430                                     struct buffer_head *blk_bh,
2431                                     struct ocfs2_caching_info *ref_ci,
2432                                     struct buffer_head *ref_root_bh)
2433 {
2434         struct ocfs2_xattr_block *xb;
2435         int ret = 0;
2436         struct ocfs2_xattr_value_buf vb = {
2437                 .vb_bh = blk_bh,
2438                 .vb_access = ocfs2_journal_access_xb,
2439         };
2440         struct ocfs2_rm_xattr_bucket_para args = {
2441                 .ref_ci = ref_ci,
2442                 .ref_root_bh = ref_root_bh,
2443         };
2444
2445         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2446         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2447                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2448                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2449                                                  ref_ci, ref_root_bh);
2450         } else
2451                 ret = ocfs2_iterate_xattr_index_block(inode,
2452                                                 blk_bh,
2453                                                 ocfs2_rm_xattr_cluster,
2454                                                 &args);
2455
2456         return ret;
2457 }
2458
2459 static int ocfs2_xattr_free_block(struct inode *inode,
2460                                   u64 block,
2461                                   struct ocfs2_caching_info *ref_ci,
2462                                   struct buffer_head *ref_root_bh)
2463 {
2464         struct inode *xb_alloc_inode;
2465         struct buffer_head *xb_alloc_bh = NULL;
2466         struct buffer_head *blk_bh = NULL;
2467         struct ocfs2_xattr_block *xb;
2468         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2469         handle_t *handle;
2470         int ret = 0;
2471         u64 blk, bg_blkno;
2472         u16 bit;
2473
2474         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2475         if (ret < 0) {
2476                 mlog_errno(ret);
2477                 goto out;
2478         }
2479
2480         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2481         if (ret < 0) {
2482                 mlog_errno(ret);
2483                 goto out;
2484         }
2485
2486         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2487         blk = le64_to_cpu(xb->xb_blkno);
2488         bit = le16_to_cpu(xb->xb_suballoc_bit);
2489         if (xb->xb_suballoc_loc)
2490                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2491         else
2492                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2493
2494         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2495                                 EXTENT_ALLOC_SYSTEM_INODE,
2496                                 le16_to_cpu(xb->xb_suballoc_slot));
2497         if (!xb_alloc_inode) {
2498                 ret = -ENOMEM;
2499                 mlog_errno(ret);
2500                 goto out;
2501         }
2502         mutex_lock(&xb_alloc_inode->i_mutex);
2503
2504         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2505         if (ret < 0) {
2506                 mlog_errno(ret);
2507                 goto out_mutex;
2508         }
2509
2510         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2511         if (IS_ERR(handle)) {
2512                 ret = PTR_ERR(handle);
2513                 mlog_errno(ret);
2514                 goto out_unlock;
2515         }
2516
2517         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2518                                        bit, bg_blkno, 1);
2519         if (ret < 0)
2520                 mlog_errno(ret);
2521
2522         ocfs2_commit_trans(osb, handle);
2523 out_unlock:
2524         ocfs2_inode_unlock(xb_alloc_inode, 1);
2525         brelse(xb_alloc_bh);
2526 out_mutex:
2527         mutex_unlock(&xb_alloc_inode->i_mutex);
2528         iput(xb_alloc_inode);
2529 out:
2530         brelse(blk_bh);
2531         return ret;
2532 }
2533
2534 /*
2535  * ocfs2_xattr_remove()
2536  *
2537  * Free extended attribute resources associated with this inode.
2538  */
2539 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2540 {
2541         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2542         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2543         struct ocfs2_refcount_tree *ref_tree = NULL;
2544         struct buffer_head *ref_root_bh = NULL;
2545         struct ocfs2_caching_info *ref_ci = NULL;
2546         handle_t *handle;
2547         int ret;
2548
2549         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2550                 return 0;
2551
2552         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2553                 return 0;
2554
2555         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2556                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2557                                                le64_to_cpu(di->i_refcount_loc),
2558                                                1, &ref_tree, &ref_root_bh);
2559                 if (ret) {
2560                         mlog_errno(ret);
2561                         goto out;
2562                 }
2563                 ref_ci = &ref_tree->rf_ci;
2564
2565         }
2566
2567         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2568                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2569                                                ref_ci, ref_root_bh);
2570                 if (ret < 0) {
2571                         mlog_errno(ret);
2572                         goto out;
2573                 }
2574         }
2575
2576         if (di->i_xattr_loc) {
2577                 ret = ocfs2_xattr_free_block(inode,
2578                                              le64_to_cpu(di->i_xattr_loc),
2579                                              ref_ci, ref_root_bh);
2580                 if (ret < 0) {
2581                         mlog_errno(ret);
2582                         goto out;
2583                 }
2584         }
2585
2586         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2587                                    OCFS2_INODE_UPDATE_CREDITS);
2588         if (IS_ERR(handle)) {
2589                 ret = PTR_ERR(handle);
2590                 mlog_errno(ret);
2591                 goto out;
2592         }
2593         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2594                                       OCFS2_JOURNAL_ACCESS_WRITE);
2595         if (ret) {
2596                 mlog_errno(ret);
2597                 goto out_commit;
2598         }
2599
2600         di->i_xattr_loc = 0;
2601
2602         spin_lock(&oi->ip_lock);
2603         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2604         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2605         spin_unlock(&oi->ip_lock);
2606
2607         ocfs2_journal_dirty(handle, di_bh);
2608 out_commit:
2609         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2610 out:
2611         if (ref_tree)
2612                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2613         brelse(ref_root_bh);
2614         return ret;
2615 }
2616
2617 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2618                                         struct ocfs2_dinode *di)
2619 {
2620         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2621         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2622         int free;
2623
2624         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2625                 return 0;
2626
2627         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2628                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2629                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2630         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2631                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2632                         le64_to_cpu(di->i_size);
2633         } else {
2634                 struct ocfs2_extent_list *el = &di->id2.i_list;
2635                 free = (le16_to_cpu(el->l_count) -
2636                         le16_to_cpu(el->l_next_free_rec)) *
2637                         sizeof(struct ocfs2_extent_rec);
2638         }
2639         if (free >= xattrsize)
2640                 return 1;
2641
2642         return 0;
2643 }
2644
2645 /*
2646  * ocfs2_xattr_ibody_find()
2647  *
2648  * Find extended attribute in inode block and
2649  * fill search info into struct ocfs2_xattr_search.
2650  */
2651 static int ocfs2_xattr_ibody_find(struct inode *inode,
2652                                   int name_index,
2653                                   const char *name,
2654                                   struct ocfs2_xattr_search *xs)
2655 {
2656         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2657         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2658         int ret;
2659         int has_space = 0;
2660
2661         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2662                 return 0;
2663
2664         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2665                 down_read(&oi->ip_alloc_sem);
2666                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2667                 up_read(&oi->ip_alloc_sem);
2668                 if (!has_space)
2669                         return 0;
2670         }
2671
2672         xs->xattr_bh = xs->inode_bh;
2673         xs->end = (void *)di + inode->i_sb->s_blocksize;
2674         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2675                 xs->header = (struct ocfs2_xattr_header *)
2676                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2677         else
2678                 xs->header = (struct ocfs2_xattr_header *)
2679                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2680         xs->base = (void *)xs->header;
2681         xs->here = xs->header->xh_entries;
2682
2683         /* Find the named attribute. */
2684         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2685                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2686                 if (ret && ret != -ENODATA)
2687                         return ret;
2688                 xs->not_found = ret;
2689         }
2690
2691         return 0;
2692 }
2693
2694 static int ocfs2_xattr_ibody_init(struct inode *inode,
2695                                   struct buffer_head *di_bh,
2696                                   struct ocfs2_xattr_set_ctxt *ctxt)
2697 {
2698         int ret;
2699         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2700         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2701         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2702         unsigned int xattrsize = osb->s_xattr_inline_size;
2703
2704         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2705                 ret = -ENOSPC;
2706                 goto out;
2707         }
2708
2709         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2710                                       OCFS2_JOURNAL_ACCESS_WRITE);
2711         if (ret) {
2712                 mlog_errno(ret);
2713                 goto out;
2714         }
2715
2716         /*
2717          * Adjust extent record count or inline data size
2718          * to reserve space for extended attribute.
2719          */
2720         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2721                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2722                 le16_add_cpu(&idata->id_count, -xattrsize);
2723         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2724                 struct ocfs2_extent_list *el = &di->id2.i_list;
2725                 le16_add_cpu(&el->l_count, -(xattrsize /
2726                                              sizeof(struct ocfs2_extent_rec)));
2727         }
2728         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2729
2730         spin_lock(&oi->ip_lock);
2731         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2732         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2733         spin_unlock(&oi->ip_lock);
2734
2735         ocfs2_journal_dirty(ctxt->handle, di_bh);
2736
2737 out:
2738         return ret;
2739 }
2740
2741 /*
2742  * ocfs2_xattr_ibody_set()
2743  *
2744  * Set, replace or remove an extended attribute into inode block.
2745  *
2746  */
2747 static int ocfs2_xattr_ibody_set(struct inode *inode,
2748                                  struct ocfs2_xattr_info *xi,
2749                                  struct ocfs2_xattr_search *xs,
2750                                  struct ocfs2_xattr_set_ctxt *ctxt)
2751 {
2752         int ret;
2753         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2754         struct ocfs2_xa_loc loc;
2755
2756         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2757                 return -ENOSPC;
2758
2759         down_write(&oi->ip_alloc_sem);
2760         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2761                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2762                 if (ret) {
2763                         if (ret != -ENOSPC)
2764                                 mlog_errno(ret);
2765                         goto out;
2766                 }
2767         }
2768
2769         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2770                                  xs->not_found ? NULL : xs->here);
2771         ret = ocfs2_xa_set(&loc, xi, ctxt);
2772         if (ret) {
2773                 if (ret != -ENOSPC)
2774                         mlog_errno(ret);
2775                 goto out;
2776         }
2777         xs->here = loc.xl_entry;
2778
2779 out:
2780         up_write(&oi->ip_alloc_sem);
2781
2782         return ret;
2783 }
2784
2785 /*
2786  * ocfs2_xattr_block_find()
2787  *
2788  * Find extended attribute in external block and
2789  * fill search info into struct ocfs2_xattr_search.
2790  */
2791 static int ocfs2_xattr_block_find(struct inode *inode,
2792                                   int name_index,
2793                                   const char *name,
2794                                   struct ocfs2_xattr_search *xs)
2795 {
2796         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2797         struct buffer_head *blk_bh = NULL;
2798         struct ocfs2_xattr_block *xb;
2799         int ret = 0;
2800
2801         if (!di->i_xattr_loc)
2802                 return ret;
2803
2804         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2805                                      &blk_bh);
2806         if (ret < 0) {
2807                 mlog_errno(ret);
2808                 return ret;
2809         }
2810
2811         xs->xattr_bh = blk_bh;
2812         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2813
2814         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2815                 xs->header = &xb->xb_attrs.xb_header;
2816                 xs->base = (void *)xs->header;
2817                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2818                 xs->here = xs->header->xh_entries;
2819
2820                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2821         } else
2822                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2823                                                    name_index,
2824                                                    name, xs);
2825
2826         if (ret && ret != -ENODATA) {
2827                 xs->xattr_bh = NULL;
2828                 goto cleanup;
2829         }
2830         xs->not_found = ret;
2831         return 0;
2832 cleanup:
2833         brelse(blk_bh);
2834
2835         return ret;
2836 }
2837
2838 static int ocfs2_create_xattr_block(struct inode *inode,
2839                                     struct buffer_head *inode_bh,
2840                                     struct ocfs2_xattr_set_ctxt *ctxt,
2841                                     int indexed,
2842                                     struct buffer_head **ret_bh)
2843 {
2844         int ret;
2845         u16 suballoc_bit_start;
2846         u32 num_got;
2847         u64 suballoc_loc, first_blkno;
2848         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2849         struct buffer_head *new_bh = NULL;
2850         struct ocfs2_xattr_block *xblk;
2851
2852         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2853                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2854         if (ret < 0) {
2855                 mlog_errno(ret);
2856                 goto end;
2857         }
2858
2859         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2860                                    &suballoc_loc, &suballoc_bit_start,
2861                                    &num_got, &first_blkno);
2862         if (ret < 0) {
2863                 mlog_errno(ret);
2864                 goto end;
2865         }
2866
2867         new_bh = sb_getblk(inode->i_sb, first_blkno);
2868         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2869
2870         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2871                                       new_bh,
2872                                       OCFS2_JOURNAL_ACCESS_CREATE);
2873         if (ret < 0) {
2874                 mlog_errno(ret);
2875                 goto end;
2876         }
2877
2878         /* Initialize ocfs2_xattr_block */
2879         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2880         memset(xblk, 0, inode->i_sb->s_blocksize);
2881         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2882         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2883         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2884         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2885         xblk->xb_fs_generation =
2886                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2887         xblk->xb_blkno = cpu_to_le64(first_blkno);
2888         if (indexed) {
2889                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2890                 xr->xt_clusters = cpu_to_le32(1);
2891                 xr->xt_last_eb_blk = 0;
2892                 xr->xt_list.l_tree_depth = 0;
2893                 xr->xt_list.l_count = cpu_to_le16(
2894                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2895                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2896                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2897         }
2898         ocfs2_journal_dirty(ctxt->handle, new_bh);
2899
2900         /* Add it to the inode */
2901         di->i_xattr_loc = cpu_to_le64(first_blkno);
2902
2903         spin_lock(&OCFS2_I(inode)->ip_lock);
2904         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2905         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2906         spin_unlock(&OCFS2_I(inode)->ip_lock);
2907
2908         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2909
2910         *ret_bh = new_bh;
2911         new_bh = NULL;
2912
2913 end:
2914         brelse(new_bh);
2915         return ret;
2916 }
2917
2918 /*
2919  * ocfs2_xattr_block_set()
2920  *
2921  * Set, replace or remove an extended attribute into external block.
2922  *
2923  */
2924 static int ocfs2_xattr_block_set(struct inode *inode,
2925                                  struct ocfs2_xattr_info *xi,
2926                                  struct ocfs2_xattr_search *xs,
2927                                  struct ocfs2_xattr_set_ctxt *ctxt)
2928 {
2929         struct buffer_head *new_bh = NULL;
2930         struct ocfs2_xattr_block *xblk = NULL;
2931         int ret;
2932         struct ocfs2_xa_loc loc;
2933
2934         if (!xs->xattr_bh) {
2935                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2936                                                0, &new_bh);
2937                 if (ret) {
2938                         mlog_errno(ret);
2939                         goto end;
2940                 }
2941
2942                 xs->xattr_bh = new_bh;
2943                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2944                 xs->header = &xblk->xb_attrs.xb_header;
2945                 xs->base = (void *)xs->header;
2946                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2947                 xs->here = xs->header->xh_entries;
2948         } else
2949                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2950
2951         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2952                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2953                                               xs->not_found ? NULL : xs->here);
2954
2955                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2956                 if (!ret)
2957                         xs->here = loc.xl_entry;
2958                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2959                         goto end;
2960                 else {
2961                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2962                         if (ret)
2963                                 goto end;
2964                 }
2965         }
2966
2967         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2968                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2969
2970 end:
2971         return ret;
2972 }
2973
2974 /* Check whether the new xattr can be inserted into the inode. */
2975 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2976                                        struct ocfs2_xattr_info *xi,
2977                                        struct ocfs2_xattr_search *xs)
2978 {
2979         struct ocfs2_xattr_entry *last;
2980         int free, i;
2981         size_t min_offs = xs->end - xs->base;
2982
2983         if (!xs->header)
2984                 return 0;
2985
2986         last = xs->header->xh_entries;
2987
2988         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2989                 size_t offs = le16_to_cpu(last->xe_name_offset);
2990                 if (offs < min_offs)
2991                         min_offs = offs;
2992                 last += 1;
2993         }
2994
2995         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2996         if (free < 0)
2997                 return 0;
2998
2999         BUG_ON(!xs->not_found);
3000
3001         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3002                 return 1;
3003
3004         return 0;
3005 }
3006
3007 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3008                                      struct ocfs2_dinode *di,
3009                                      struct ocfs2_xattr_info *xi,
3010                                      struct ocfs2_xattr_search *xis,
3011                                      struct ocfs2_xattr_search *xbs,
3012                                      int *clusters_need,
3013                                      int *meta_need,
3014                                      int *credits_need)
3015 {
3016         int ret = 0, old_in_xb = 0;
3017         int clusters_add = 0, meta_add = 0, credits = 0;
3018         struct buffer_head *bh = NULL;
3019         struct ocfs2_xattr_block *xb = NULL;
3020         struct ocfs2_xattr_entry *xe = NULL;
3021         struct ocfs2_xattr_value_root *xv = NULL;
3022         char *base = NULL;
3023         int name_offset, name_len = 0;
3024         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3025                                                     xi->xi_value_len);
3026         u64 value_size;
3027
3028         /*
3029          * Calculate the clusters we need to write.
3030          * No matter whether we replace an old one or add a new one,
3031          * we need this for writing.
3032          */
3033         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3034                 credits += new_clusters *
3035                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3036
3037         if (xis->not_found && xbs->not_found) {
3038                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3039
3040                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3041                         clusters_add += new_clusters;
3042                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3043                                                         &def_xv.xv.xr_list,
3044                                                         new_clusters);
3045                 }
3046
3047                 goto meta_guess;
3048         }
3049
3050         if (!xis->not_found) {
3051                 xe = xis->here;
3052                 name_offset = le16_to_cpu(xe->xe_name_offset);
3053                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3054                 base = xis->base;
3055                 credits += OCFS2_INODE_UPDATE_CREDITS;
3056         } else {
3057                 int i, block_off = 0;
3058                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3059                 xe = xbs->here;
3060                 name_offset = le16_to_cpu(xe->xe_name_offset);
3061                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3062                 i = xbs->here - xbs->header->xh_entries;
3063                 old_in_xb = 1;
3064
3065                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3066                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3067                                                         bucket_xh(xbs->bucket),
3068                                                         i, &block_off,
3069                                                         &name_offset);
3070                         base = bucket_block(xbs->bucket, block_off);
3071                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3072                 } else {
3073                         base = xbs->base;
3074                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3075                 }
3076         }
3077
3078         /*
3079          * delete a xattr doesn't need metadata and cluster allocation.
3080          * so just calculate the credits and return.
3081          *
3082          * The credits for removing the value tree will be extended
3083          * by ocfs2_remove_extent itself.
3084          */
3085         if (!xi->xi_value) {
3086                 if (!ocfs2_xattr_is_local(xe))
3087                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3088
3089                 goto out;
3090         }
3091
3092         /* do cluster allocation guess first. */
3093         value_size = le64_to_cpu(xe->xe_value_size);
3094
3095         if (old_in_xb) {
3096                 /*
3097                  * In xattr set, we always try to set the xe in inode first,
3098                  * so if it can be inserted into inode successfully, the old
3099                  * one will be removed from the xattr block, and this xattr
3100                  * will be inserted into inode as a new xattr in inode.
3101                  */
3102                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3103                         clusters_add += new_clusters;
3104                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3105                                     OCFS2_INODE_UPDATE_CREDITS;
3106                         if (!ocfs2_xattr_is_local(xe))
3107                                 credits += ocfs2_calc_extend_credits(
3108                                                         inode->i_sb,
3109                                                         &def_xv.xv.xr_list,
3110                                                         new_clusters);
3111                         goto out;
3112                 }
3113         }
3114
3115         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3116                 /* the new values will be stored outside. */
3117                 u32 old_clusters = 0;
3118
3119                 if (!ocfs2_xattr_is_local(xe)) {
3120                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3121                                                                  value_size);
3122                         xv = (struct ocfs2_xattr_value_root *)
3123                              (base + name_offset + name_len);
3124                         value_size = OCFS2_XATTR_ROOT_SIZE;
3125                 } else
3126                         xv = &def_xv.xv;
3127
3128                 if (old_clusters >= new_clusters) {
3129                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3130                         goto out;
3131                 } else {
3132                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3133                         clusters_add += new_clusters - old_clusters;
3134                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3135                                                              &xv->xr_list,
3136                                                              new_clusters -
3137                                                              old_clusters);
3138                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3139                                 goto out;
3140                 }
3141         } else {
3142                 /*
3143                  * Now the new value will be stored inside. So if the new
3144                  * value is smaller than the size of value root or the old
3145                  * value, we don't need any allocation, otherwise we have
3146                  * to guess metadata allocation.
3147                  */
3148                 if ((ocfs2_xattr_is_local(xe) &&
3149                      (value_size >= xi->xi_value_len)) ||
3150                     (!ocfs2_xattr_is_local(xe) &&
3151                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3152                         goto out;
3153         }
3154
3155 meta_guess:
3156         /* calculate metadata allocation. */
3157         if (di->i_xattr_loc) {
3158                 if (!xbs->xattr_bh) {
3159                         ret = ocfs2_read_xattr_block(inode,
3160                                                      le64_to_cpu(di->i_xattr_loc),
3161                                                      &bh);
3162                         if (ret) {
3163                                 mlog_errno(ret);
3164                                 goto out;
3165                         }
3166
3167                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3168                 } else
3169                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3170
3171                 /*
3172                  * If there is already an xattr tree, good, we can calculate
3173                  * like other b-trees. Otherwise we may have the chance of
3174                  * create a tree, the credit calculation is borrowed from
3175                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3176                  * new tree will be cluster based, so no meta is needed.
3177                  */
3178                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3179                         struct ocfs2_extent_list *el =
3180                                  &xb->xb_attrs.xb_root.xt_list;
3181                         meta_add += ocfs2_extend_meta_needed(el);
3182                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3183                                                              el, 1);
3184                 } else
3185                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3186
3187                 /*
3188                  * This cluster will be used either for new bucket or for
3189                  * new xattr block.
3190                  * If the cluster size is the same as the bucket size, one
3191                  * more is needed since we may need to extend the bucket
3192                  * also.
3193                  */
3194                 clusters_add += 1;
3195                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3196                 if (OCFS2_XATTR_BUCKET_SIZE ==
3197                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3198                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3199                         clusters_add += 1;
3200                 }
3201         } else {
3202                 meta_add += 1;
3203                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3204         }
3205 out:
3206         if (clusters_need)
3207                 *clusters_need = clusters_add;
3208         if (meta_need)
3209                 *meta_need = meta_add;
3210         if (credits_need)
3211                 *credits_need = credits;
3212         brelse(bh);
3213         return ret;
3214 }
3215
3216 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3217                                      struct ocfs2_dinode *di,
3218                                      struct ocfs2_xattr_info *xi,
3219                                      struct ocfs2_xattr_search *xis,
3220                                      struct ocfs2_xattr_search *xbs,
3221                                      struct ocfs2_xattr_set_ctxt *ctxt,
3222                                      int extra_meta,
3223                                      int *credits)
3224 {
3225         int clusters_add, meta_add, ret;
3226         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3227
3228         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3229
3230         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3231
3232         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3233                                         &clusters_add, &meta_add, credits);
3234         if (ret) {
3235                 mlog_errno(ret);
3236                 return ret;
3237         }
3238
3239         meta_add += extra_meta;
3240         trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3241                                         clusters_add, *credits);
3242
3243         if (meta_add) {
3244                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3245                                                         &ctxt->meta_ac);
3246                 if (ret) {
3247                         mlog_errno(ret);
3248                         goto out;
3249                 }
3250         }
3251
3252         if (clusters_add) {
3253                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3254                 if (ret)
3255                         mlog_errno(ret);
3256         }
3257 out:
3258         if (ret) {
3259                 if (ctxt->meta_ac) {
3260                         ocfs2_free_alloc_context(ctxt->meta_ac);
3261                         ctxt->meta_ac = NULL;
3262                 }
3263
3264                 /*
3265                  * We cannot have an error and a non null ctxt->data_ac.
3266                  */
3267         }
3268
3269         return ret;
3270 }
3271
3272 static int __ocfs2_xattr_set_handle(struct inode *inode,
3273                                     struct ocfs2_dinode *di,
3274                                     struct ocfs2_xattr_info *xi,
3275                                     struct ocfs2_xattr_search *xis,
3276                                     struct ocfs2_xattr_search *xbs,
3277                                     struct ocfs2_xattr_set_ctxt *ctxt)
3278 {
3279         int ret = 0, credits, old_found;
3280
3281         if (!xi->xi_value) {
3282                 /* Remove existing extended attribute */
3283                 if (!xis->not_found)
3284                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3285                 else if (!xbs->not_found)
3286                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3287         } else {
3288                 /* We always try to set extended attribute into inode first*/
3289                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3290                 if (!ret && !xbs->not_found) {
3291                         /*
3292                          * If succeed and that extended attribute existing in
3293                          * external block, then we will remove it.
3294                          */
3295                         xi->xi_value = NULL;
3296                         xi->xi_value_len = 0;
3297
3298                         old_found = xis->not_found;
3299                         xis->not_found = -ENODATA;
3300                         ret = ocfs2_calc_xattr_set_need(inode,
3301                                                         di,
3302                                                         xi,
3303                                                         xis,
3304                                                         xbs,
3305                                                         NULL,
3306                                                         NULL,
3307                                                         &credits);
3308                         xis->not_found = old_found;
3309                         if (ret) {
3310                                 mlog_errno(ret);
3311                                 goto out;
3312                         }
3313
3314                         ret = ocfs2_extend_trans(ctxt->handle, credits);
3315                         if (ret) {
3316                                 mlog_errno(ret);
3317                                 goto out;
3318                         }
3319                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3320                 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3321                         if (di->i_xattr_loc && !xbs->xattr_bh) {
3322                                 ret = ocfs2_xattr_block_find(inode,
3323                                                              xi->xi_name_index,
3324                                                              xi->xi_name, xbs);
3325                                 if (ret)
3326                                         goto out;
3327
3328                                 old_found = xis->not_found;
3329                                 xis->not_found = -ENODATA;
3330                                 ret = ocfs2_calc_xattr_set_need(inode,
3331                                                                 di,
3332                                                                 xi,
3333                                                                 xis,
3334                                                                 xbs,
3335                                                                 NULL,
3336                                                                 NULL,
3337                                                                 &credits);
3338                                 xis->not_found = old_found;
3339                                 if (ret) {
3340                                         mlog_errno(ret);
3341                                         goto out;
3342                                 }
3343
3344                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3345                                 if (ret) {
3346                                         mlog_errno(ret);
3347                                         goto out;
3348                                 }
3349                         }
3350                         /*
3351                          * If no space in inode, we will set extended attribute
3352                          * into external block.
3353                          */
3354                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3355                         if (ret)
3356                                 goto out;
3357                         if (!xis->not_found) {
3358                                 /*
3359                                  * If succeed and that extended attribute
3360                                  * existing in inode, we will remove it.
3361                                  */
3362                                 xi->xi_value = NULL;
3363                                 xi->xi_value_len = 0;
3364                                 xbs->not_found = -ENODATA;
3365                                 ret = ocfs2_calc_xattr_set_need(inode,
3366                                                                 di,
3367                                                                 xi,
3368                                                                 xis,
3369                                                                 xbs,
3370                                                                 NULL,
3371                                                                 NULL,
3372                                                                 &credits);
3373                                 if (ret) {
3374                                         mlog_errno(ret);
3375                                         goto out;
3376                                 }
3377
3378                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3379                                 if (ret) {
3380                                         mlog_errno(ret);
3381                                         goto out;
3382                                 }
3383                                 ret = ocfs2_xattr_ibody_set(inode, xi,
3384                                                             xis, ctxt);
3385                         }
3386                 }
3387         }
3388
3389         if (!ret) {
3390                 /* Update inode ctime. */
3391                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3392                                               xis->inode_bh,
3393                                               OCFS2_JOURNAL_ACCESS_WRITE);
3394                 if (ret) {
3395                         mlog_errno(ret);
3396                         goto out;
3397                 }
3398
3399                 inode->i_ctime = CURRENT_TIME;
3400                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3401                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3402                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3403         }
3404 out:
3405         return ret;
3406 }
3407
3408 /*
3409  * This function only called duing creating inode
3410  * for init security/acl xattrs of the new inode.
3411  * All transanction credits have been reserved in mknod.
3412  */
3413 int ocfs2_xattr_set_handle(handle_t *handle,
3414                            struct inode *inode,
3415                            struct buffer_head *di_bh,
3416                            int name_index,
3417                            const char *name,
3418                            const void *value,
3419                            size_t value_len,
3420                            int flags,
3421                            struct ocfs2_alloc_context *meta_ac,
3422                            struct ocfs2_alloc_context *data_ac)
3423 {
3424         struct ocfs2_dinode *di;
3425         int ret;
3426
3427         struct ocfs2_xattr_info xi = {
3428                 .xi_name_index = name_index,
3429                 .xi_name = name,
3430                 .xi_name_len = strlen(name),
3431                 .xi_value = value,
3432                 .xi_value_len = value_len,
3433         };
3434
3435         struct ocfs2_xattr_search xis = {
3436                 .not_found = -ENODATA,
3437         };
3438
3439         struct ocfs2_xattr_search xbs = {
3440                 .not_found = -ENODATA,
3441         };
3442
3443         struct ocfs2_xattr_set_ctxt ctxt = {
3444                 .handle = handle,
3445                 .meta_ac = meta_ac,
3446                 .data_ac = data_ac,
3447         };
3448
3449         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3450                 return -EOPNOTSUPP;
3451
3452         /*
3453          * In extreme situation, may need xattr bucket when
3454          * block size is too small. And we have already reserved
3455          * the credits for bucket in mknod.
3456          */
3457         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3458                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3459                 if (!xbs.bucket) {
3460                         mlog_errno(-ENOMEM);
3461                         return -ENOMEM;
3462                 }
3463         }
3464
3465         xis.inode_bh = xbs.inode_bh = di_bh;
3466         di = (struct ocfs2_dinode *)di_bh->b_data;
3467
3468         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3469
3470         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3471         if (ret)
3472                 goto cleanup;
3473         if (xis.not_found) {
3474                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3475                 if (ret)
3476                         goto cleanup;
3477         }
3478
3479         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3480
3481 cleanup:
3482         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3483         brelse(xbs.xattr_bh);
3484         ocfs2_xattr_bucket_free(xbs.bucket);
3485
3486         return ret;
3487 }
3488
3489 /*
3490  * ocfs2_xattr_set()
3491  *
3492  * Set, replace or remove an extended attribute for this inode.
3493  * value is NULL to remove an existing extended attribute, else either
3494  * create or replace an extended attribute.
3495  */
3496 int ocfs2_xattr_set(struct inode *inode,
3497                     int name_index,
3498                     const char *name,
3499                     const void *value,
3500                     size_t value_len,
3501                     int flags)
3502 {
3503         struct buffer_head *di_bh = NULL;
3504         struct ocfs2_dinode *di;
3505         int ret, credits, ref_meta = 0, ref_credits = 0;
3506         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3507         struct inode *tl_inode = osb->osb_tl_inode;
3508         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
3509         struct ocfs2_refcount_tree *ref_tree = NULL;
3510
3511         struct ocfs2_xattr_info xi = {
3512                 .xi_name_index = name_index,
3513                 .xi_name = name,
3514                 .xi_name_len = strlen(name),
3515                 .xi_value = value,
3516                 .xi_value_len = value_len,
3517         };
3518
3519         struct ocfs2_xattr_search xis = {
3520                 .not_found = -ENODATA,
3521         };
3522
3523         struct ocfs2_xattr_search xbs = {
3524                 .not_found = -ENODATA,
3525         };
3526
3527         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3528                 return -EOPNOTSUPP;
3529
3530         /*
3531          * Only xbs will be used on indexed trees.  xis doesn't need a
3532          * bucket.
3533          */
3534         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3535         if (!xbs.bucket) {
3536                 mlog_errno(-ENOMEM);
3537                 return -ENOMEM;
3538         }
3539
3540         ret = ocfs2_inode_lock(inode, &di_bh, 1);
3541         if (ret < 0) {
3542                 mlog_errno(ret);
3543                 goto cleanup_nolock;
3544         }
3545         xis.inode_bh = xbs.inode_bh = di_bh;
3546         di = (struct ocfs2_dinode *)di_bh->b_data;
3547
3548         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3549         /*
3550          * Scan inode and external block to find the same name
3551          * extended attribute and collect search information.
3552          */
3553         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3554         if (ret)
3555                 goto cleanup;
3556         if (xis.not_found) {
3557                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3558                 if (ret)
3559                         goto cleanup;
3560         }
3561
3562         if (xis.not_found && xbs.not_found) {
3563                 ret = -ENODATA;
3564                 if (flags & XATTR_REPLACE)
3565                         goto cleanup;
3566                 ret = 0;
3567                 if (!value)
3568                         goto cleanup;
3569         } else {
3570                 ret = -EEXIST;
3571                 if (flags & XATTR_CREATE)
3572                         goto cleanup;
3573         }
3574
3575         /* Check whether the value is refcounted and do some preparation. */
3576         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3577             (!xis.not_found || !xbs.not_found)) {
3578                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3579                                                    &xis, &xbs, &ref_tree,
3580                                                    &ref_meta, &ref_credits);
3581                 if (ret) {
3582                         mlog_errno(ret);
3583                         goto cleanup;
3584                 }
3585         }
3586
3587         mutex_lock(&tl_inode->i_mutex);
3588
3589         if (ocfs2_truncate_log_needs_flush(osb)) {
3590                 ret = __ocfs2_flush_truncate_log(osb);
3591                 if (ret < 0) {
3592                         mutex_unlock(&tl_inode->i_mutex);
3593                         mlog_errno(ret);
3594                         goto cleanup;
3595                 }
3596         }
3597         mutex_unlock(&tl_inode->i_mutex);
3598
3599         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3600                                         &xbs, &ctxt, ref_meta, &credits);
3601         if (ret) {
3602                 mlog_errno(ret);
3603                 goto cleanup;
3604         }
3605
3606         /* we need to update inode's ctime field, so add credit for it. */
3607         credits += OCFS2_INODE_UPDATE_CREDITS;
3608         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3609         if (IS_ERR(ctxt.handle)) {
3610                 ret = PTR_ERR(ctxt.handle);
3611                 mlog_errno(ret);
3612                 goto cleanup;
3613         }
3614
3615         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3616
3617         ocfs2_commit_trans(osb, ctxt.handle);
3618
3619         if (ctxt.data_ac)
3620                 ocfs2_free_alloc_context(ctxt.data_ac);
3621         if (ctxt.meta_ac)
3622                 ocfs2_free_alloc_context(ctxt.meta_ac);
3623         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3624                 ocfs2_schedule_truncate_log_flush(osb, 1);
3625         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3626
3627 cleanup:
3628         if (ref_tree)
3629                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3630         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3631         if (!value && !ret) {
3632                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3633                 if (ret)
3634                         mlog_errno(ret);
3635         }
3636         ocfs2_inode_unlock(inode, 1);
3637 cleanup_nolock:
3638         brelse(di_bh);
3639         brelse(xbs.xattr_bh);
3640         ocfs2_xattr_bucket_free(xbs.bucket);
3641
3642         return ret;
3643 }
3644
3645 /*
3646  * Find the xattr extent rec which may contains name_hash.
3647  * e_cpos will be the first name hash of the xattr rec.
3648  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3649  */
3650 static int ocfs2_xattr_get_rec(struct inode *inode,
3651                                u32 name_hash,
3652                                u64 *p_blkno,
3653                                u32 *e_cpos,
3654                                u32 *num_clusters,
3655                                struct ocfs2_extent_list *el)
3656 {
3657         int ret = 0, i;
3658         struct buffer_head *eb_bh = NULL;
3659         struct ocfs2_extent_block *eb;
3660         struct ocfs2_extent_rec *rec = NULL;
3661         u64 e_blkno = 0;
3662
3663         if (el->l_tree_depth) {
3664                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3665                                       &eb_bh);
3666                 if (ret) {
3667                         mlog_errno(ret);
3668                         goto out;
3669                 }
3670
3671                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3672                 el = &eb->h_list;
3673
3674                 if (el->l_tree_depth) {
3675                         ocfs2_error(inode->i_sb,
3676                                     "Inode %lu has non zero tree depth in "
3677                                     "xattr tree block %llu\n", inode->i_ino,
3678                                     (unsigned long long)eb_bh->b_blocknr);
3679                         ret = -EROFS;
3680                         goto out;
3681                 }
3682         }
3683
3684         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3685                 rec = &el->l_recs[i];
3686
3687                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3688                         e_blkno = le64_to_cpu(rec->e_blkno);
3689                         break;
3690                 }
3691         }
3692
3693         if (!e_blkno) {
3694                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3695                             "record (%u, %u, 0) in xattr", inode->i_ino,
3696                             le32_to_cpu(rec->e_cpos),
3697                             ocfs2_rec_clusters(el, rec));
3698                 ret = -EROFS;
3699                 goto out;
3700         }
3701
3702         *p_blkno = le64_to_cpu(rec->e_blkno);
3703         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3704         if (e_cpos)
3705                 *e_cpos = le32_to_cpu(rec->e_cpos);
3706 out:
3707         brelse(eb_bh);
3708         return ret;
3709 }
3710
3711 typedef int (xattr_bucket_func)(struct inode *inode,
3712                                 struct ocfs2_xattr_bucket *bucket,
3713                                 void *para);
3714
3715 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3716                                    struct ocfs2_xattr_bucket *bucket,
3717                                    int name_index,
3718                                    const char *name,
3719                                    u32 name_hash,
3720                                    u16 *xe_index,
3721                                    int *found)
3722 {
3723         int i, ret = 0, cmp = 1, block_off, new_offset;
3724         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3725         size_t name_len = strlen(name);
3726         struct ocfs2_xattr_entry *xe = NULL;
3727         char *xe_name;
3728
3729         /*
3730          * We don't use binary search in the bucket because there
3731          * may be multiple entries with the same name hash.
3732          */
3733         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3734                 xe = &xh->xh_entries[i];
3735
3736                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3737                         continue;
3738                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3739                         break;
3740
3741                 cmp = name_index - ocfs2_xattr_get_type(xe);
3742                 if (!cmp)
3743                         cmp = name_len - xe->xe_name_len;
3744                 if (cmp)
3745                         continue;
3746
3747                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3748                                                         xh,
3749                                                         i,
3750                                                         &block_off,
3751                                                         &new_offset);
3752                 if (ret) {
3753                         mlog_errno(ret);
3754                         break;
3755                 }
3756
3757
3758                 xe_name = bucket_block(bucket, block_off) + new_offset;
3759                 if (!memcmp(name, xe_name, name_len)) {
3760                         *xe_index = i;
3761                         *found = 1;
3762                         ret = 0;
3763                         break;
3764                 }
3765         }
3766
3767         return ret;
3768 }
3769
3770 /*
3771  * Find the specified xattr entry in a series of buckets.
3772  * This series start from p_blkno and last for num_clusters.
3773  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3774  * the num of the valid buckets.
3775  *
3776  * Return the buffer_head this xattr should reside in. And if the xattr's
3777  * hash is in the gap of 2 buckets, return the lower bucket.
3778  */
3779 static int ocfs2_xattr_bucket_find(struct inode *inode,
3780                                    int name_index,
3781                                    const char *name,
3782                                    u32 name_hash,
3783                                    u64 p_blkno,
3784                                    u32 first_hash,
3785                                    u32 num_clusters,
3786                                    struct ocfs2_xattr_search *xs)
3787 {
3788         int ret, found = 0;
3789         struct ocfs2_xattr_header *xh = NULL;
3790         struct ocfs2_xattr_entry *xe = NULL;
3791         u16 index = 0;
3792         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3793         int low_bucket = 0, bucket, high_bucket;
3794         struct ocfs2_xattr_bucket *search;
3795         u32 last_hash;
3796         u64 blkno, lower_blkno = 0;
3797
3798         search = ocfs2_xattr_bucket_new(inode);
3799         if (!search) {
3800                 ret = -ENOMEM;
3801                 mlog_errno(ret);
3802                 goto out;
3803         }
3804
3805         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3806         if (ret) {
3807                 mlog_errno(ret);
3808                 goto out;
3809         }
3810
3811         xh = bucket_xh(search);
3812         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3813         while (low_bucket <= high_bucket) {
3814                 ocfs2_xattr_bucket_relse(search);
3815
3816                 bucket = (low_bucket + high_bucket) / 2;
3817                 blkno = p_blkno + bucket * blk_per_bucket;
3818                 ret = ocfs2_read_xattr_bucket(search, blkno);
3819                 if (ret) {
3820                         mlog_errno(ret);
3821                         goto out;
3822                 }
3823
3824                 xh = bucket_xh(search);
3825                 xe = &xh->xh_entries[0];
3826                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3827                         high_bucket = bucket - 1;
3828                         continue;
3829                 }
3830
3831                 /*
3832                  * Check whether the hash of the last entry in our
3833                  * bucket is larger than the search one. for an empty
3834                  * bucket, the last one is also the first one.
3835                  */
3836                 if (xh->xh_count)
3837                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3838
3839                 last_hash = le32_to_cpu(xe->xe_name_hash);
3840
3841                 /* record lower_blkno which may be the insert place. */
3842                 lower_blkno = blkno;
3843
3844                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3845                         low_bucket = bucket + 1;
3846                         continue;
3847                 }
3848
3849                 /* the searched xattr should reside in this bucket if exists. */
3850                 ret = ocfs2_find_xe_in_bucket(inode, search,
3851                                               name_index, name, name_hash,
3852                                               &index, &found);
3853                 if (ret) {
3854                         mlog_errno(ret);
3855                         goto out;
3856                 }
3857                 break;
3858         }
3859
3860         /*
3861          * Record the bucket we have found.
3862          * When the xattr's hash value is in the gap of 2 buckets, we will
3863          * always set it to the previous bucket.
3864          */
3865         if (!lower_blkno)
3866                 lower_blkno = p_blkno;
3867
3868         /* This should be in cache - we just read it during the search */
3869         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3870         if (ret) {
3871                 mlog_errno(ret);
3872                 goto out;
3873         }
3874
3875         xs->header = bucket_xh(xs->bucket);
3876         xs->base = bucket_block(xs->bucket, 0);
3877         xs->end = xs->base + inode->i_sb->s_blocksize;
3878
3879         if (found) {
3880                 xs->here = &xs->header->xh_entries[index];
3881                 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3882                         name, name_index, name_hash,
3883                         (unsigned long long)bucket_blkno(xs->bucket),
3884                         index);
3885         } else
3886                 ret = -ENODATA;
3887
3888 out:
3889         ocfs2_xattr_bucket_free(search);
3890         return ret;
3891 }
3892
3893 static int ocfs2_xattr_index_block_find(struct inode *inode,
3894                                         struct buffer_head *root_bh,
3895                                         int name_index,
3896                                         const char *name,
3897                                         struct ocfs2_xattr_search *xs)
3898 {
3899         int ret;
3900         struct ocfs2_xattr_block *xb =
3901                         (struct ocfs2_xattr_block *)root_bh->b_data;
3902         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3903         struct ocfs2_extent_list *el = &xb_root->xt_list;
3904         u64 p_blkno = 0;
3905         u32 first_hash, num_clusters = 0;
3906         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3907
3908         if (le16_to_cpu(el->l_next_free_rec) == 0)
3909                 return -ENODATA;
3910
3911         trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3912                                         name, name_index, name_hash,
3913                                         (unsigned long long)root_bh->b_blocknr,
3914                                         -1);
3915
3916         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3917                                   &num_clusters, el);
3918         if (ret) {
3919                 mlog_errno(ret);
3920                 goto out;
3921         }
3922
3923         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3924
3925         trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3926                                         name, name_index, first_hash,
3927                                         (unsigned long long)p_blkno,
3928                                         num_clusters);
3929
3930         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3931                                       p_blkno, first_hash, num_clusters, xs);
3932
3933 out:
3934         return ret;
3935 }
3936
3937 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3938                                        u64 blkno,
3939                                        u32 clusters,
3940                                        xattr_bucket_func *func,
3941                                        void *para)
3942 {
3943         int i, ret = 0;
3944         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3945         u32 num_buckets = clusters * bpc;
3946         struct ocfs2_xattr_bucket *bucket;
3947
3948         bucket = ocfs2_xattr_bucket_new(inode);
3949         if (!bucket) {
3950                 mlog_errno(-ENOMEM);
3951                 return -ENOMEM;
3952         }
3953
3954         trace_ocfs2_iterate_xattr_buckets(
3955                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3956                 (unsigned long long)blkno, clusters);
3957
3958         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3959                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3960                 if (ret) {
3961                         mlog_errno(ret);
3962                         break;
3963                 }
3964
3965                 /*
3966                  * The real bucket num in this series of blocks is stored
3967                  * in the 1st bucket.
3968                  */
3969                 if (i == 0)
3970                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3971
3972                 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
3973                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3974                 if (func) {
3975                         ret = func(inode, bucket, para);
3976                         if (ret && ret != -ERANGE)
3977                                 mlog_errno(ret);
3978                         /* Fall through to bucket_relse() */
3979                 }
3980
3981                 ocfs2_xattr_bucket_relse(bucket);
3982                 if (ret)
3983                         break;
3984         }
3985
3986         ocfs2_xattr_bucket_free(bucket);
3987         return ret;
3988 }
3989
3990 struct ocfs2_xattr_tree_list {
3991         char *buffer;
3992         size_t buffer_size;
3993         size_t result;
3994 };
3995
3996 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3997                                              struct ocfs2_xattr_header *xh,
3998                                              int index,
3999                                              int *block_off,
4000                                              int *new_offset)
4001 {
4002         u16 name_offset;
4003
4004         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4005                 return -EINVAL;
4006
4007         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4008
4009         *block_off = name_offset >> sb->s_blocksize_bits;
4010         *new_offset = name_offset % sb->s_blocksize;
4011
4012         return 0;
4013 }
4014
4015 static int ocfs2_list_xattr_bucket(struct inode *inode,
4016                                    struct ocfs2_xattr_bucket *bucket,
4017                                    void *para)
4018 {
4019         int ret = 0, type;
4020         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4021         int i, block_off, new_offset;
4022         const char *prefix, *name;
4023
4024         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4025                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4026                 type = ocfs2_xattr_get_type(entry);
4027                 prefix = ocfs2_xattr_prefix(type);
4028
4029                 if (prefix) {
4030                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4031                                                                 bucket_xh(bucket),
4032                                                                 i,
4033                                                                 &block_off,
4034                                                                 &new_offset);
4035                         if (ret)
4036                                 break;
4037
4038                         name = (const char *)bucket_block(bucket, block_off) +
4039                                 new_offset;
4040                         ret = ocfs2_xattr_list_entry(xl->buffer,
4041                                                      xl->buffer_size,
4042                                                      &xl->result,
4043                                                      prefix, name,
4044                                                      entry->xe_name_len);
4045                         if (ret)
4046                                 break;
4047                 }
4048         }
4049
4050         return ret;
4051 }
4052
4053 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4054                                            struct buffer_head *blk_bh,
4055                                            xattr_tree_rec_func *rec_func,
4056                                            void *para)
4057 {
4058         struct ocfs2_xattr_block *xb =
4059                         (struct ocfs2_xattr_block *)blk_bh->b_data;
4060         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4061         int ret = 0;
4062         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4063         u64 p_blkno = 0;
4064
4065         if (!el->l_next_free_rec || !rec_func)
4066                 return 0;
4067
4068         while (name_hash > 0) {
4069                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4070                                           &e_cpos, &num_clusters, el);
4071                 if (ret) {
4072                         mlog_errno(ret);
4073                         break;
4074                 }
4075
4076                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4077                                num_clusters, para);
4078                 if (ret) {
4079                         if (ret != -ERANGE)
4080                                 mlog_errno(ret);
4081                         break;
4082                 }
4083
4084                 if (e_cpos == 0)
4085                         break;
4086
4087                 name_hash = e_cpos - 1;
4088         }
4089
4090         return ret;
4091
4092 }
4093
4094 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4095                                      struct buffer_head *root_bh,
4096                                      u64 blkno, u32 cpos, u32 len, void *para)
4097 {
4098         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4099                                            ocfs2_list_xattr_bucket, para);
4100 }
4101
4102 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4103                                              struct buffer_head *blk_bh,
4104                                              char *buffer,
4105                                              size_t buffer_size)
4106 {
4107         int ret;
4108         struct ocfs2_xattr_tree_list xl = {
4109                 .buffer = buffer,
4110                 .buffer_size = buffer_size,
4111                 .result = 0,
4112         };
4113
4114         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4115                                               ocfs2_list_xattr_tree_rec, &xl);
4116         if (ret) {
4117                 mlog_errno(ret);
4118                 goto out;
4119         }
4120
4121         ret = xl.result;
4122 out:
4123         return ret;
4124 }
4125
4126 static int cmp_xe(const void *a, const void *b)
4127 {
4128         const struct ocfs2_xattr_entry *l = a, *r = b;
4129         u32 l_hash = le32_to_cpu(l->xe_name_hash);
4130         u32 r_hash = le32_to_cpu(r->xe_name_hash);
4131
4132         if (l_hash > r_hash)
4133                 return 1;
4134         if (l_hash < r_hash)
4135                 return -1;
4136         return 0;
4137 }
4138
4139 static void swap_xe(void *a, void *b, int size)
4140 {
4141         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4142
4143         tmp = *l;
4144         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4145         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4146 }
4147
4148 /*
4149  * When the ocfs2_xattr_block is filled up, new bucket will be created
4150  * and all the xattr entries will be moved to the new bucket.
4151  * The header goes at the start of the bucket, and the names+values are
4152  * filled from the end.  This is why *target starts as the last buffer.
4153  * Note: we need to sort the entries since they are not saved in order
4154  * in the ocfs2_xattr_block.
4155  */
4156 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4157                                            struct buffer_head *xb_bh,
4158                                            struct ocfs2_xattr_bucket *bucket)
4159 {
4160         int i, blocksize = inode->i_sb->s_blocksize;
4161         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4162         u16 offset, size, off_change;
4163         struct ocfs2_xattr_entry *xe;
4164         struct ocfs2_xattr_block *xb =
4165                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
4166         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4167         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4168         u16 count = le16_to_cpu(xb_xh->xh_count);
4169         char *src = xb_bh->b_data;
4170         char *target = bucket_block(bucket, blks - 1);
4171
4172         trace_ocfs2_cp_xattr_block_to_bucket_begin(
4173                                 (unsigned long long)xb_bh->b_blocknr,
4174                                 (unsigned long long)bucket_blkno(bucket));
4175
4176         for (i = 0; i < blks; i++)
4177                 memset(bucket_block(bucket, i), 0, blocksize);
4178
4179         /*
4180          * Since the xe_name_offset is based on ocfs2_xattr_header,
4181          * there is a offset change corresponding to the change of
4182          * ocfs2_xattr_header's position.
4183          */
4184         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4185         xe = &xb_xh->xh_entries[count - 1];
4186         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4187         size = blocksize - offset;
4188
4189         /* copy all the names and values. */
4190         memcpy(target + offset, src + offset, size);
4191
4192         /* Init new header now. */
4193         xh->xh_count = xb_xh->xh_count;
4194         xh->xh_num_buckets = cpu_to_le16(1);
4195         xh->xh_name_value_len = cpu_to_le16(size);
4196         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4197
4198         /* copy all the entries. */
4199         target = bucket_block(bucket, 0);
4200         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4201         size = count * sizeof(struct ocfs2_xattr_entry);
4202         memcpy(target + offset, (char *)xb_xh + offset, size);
4203
4204         /* Change the xe offset for all the xe because of the move. */
4205         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4206                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4207         for (i = 0; i < count; i++)
4208                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4209
4210         trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4211
4212         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4213              cmp_xe, swap_xe);
4214 }
4215
4216 /*
4217  * After we move xattr from block to index btree, we have to
4218  * update ocfs2_xattr_search to the new xe and base.
4219  *
4220  * When the entry is in xattr block, xattr_bh indicates the storage place.
4221  * While if the entry is in index b-tree, "bucket" indicates the
4222  * real place of the xattr.
4223  */
4224 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4225                                             struct ocfs2_xattr_search *xs,
4226                                             struct buffer_head *old_bh)
4227 {
4228         char *buf = old_bh->b_data;
4229         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4230         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4231         int i;
4232
4233         xs->header = bucket_xh(xs->bucket);
4234         xs->base = bucket_block(xs->bucket, 0);
4235         xs->end = xs->base + inode->i_sb->s_blocksize;
4236
4237         if (xs->not_found)
4238                 return;
4239
4240         i = xs->here - old_xh->xh_entries;
4241         xs->here = &xs->header->xh_entries[i];
4242 }
4243
4244 static int ocfs2_xattr_create_index_block(struct inode *inode,
4245                                           struct ocfs2_xattr_search *xs,
4246                                           struct ocfs2_xattr_set_ctxt *ctxt)
4247 {
4248         int ret;
4249         u32 bit_off, len;
4250         u64 blkno;
4251         handle_t *handle = ctxt->handle;
4252         struct ocfs2_inode_info *oi = OCFS2_I(inode);
4253         struct buffer_head *xb_bh = xs->xattr_bh;
4254         struct ocfs2_xattr_block *xb =
4255                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4256         struct ocfs2_xattr_tree_root *xr;
4257         u16 xb_flags = le16_to_cpu(xb->xb_flags);
4258
4259         trace_ocfs2_xattr_create_index_block_begin(
4260                                 (unsigned long long)xb_bh->b_blocknr);
4261
4262         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4263         BUG_ON(!xs->bucket);
4264
4265         /*
4266          * XXX:
4267          * We can use this lock for now, and maybe move to a dedicated mutex
4268          * if performance becomes a problem later.
4269          */
4270         down_write(&oi->ip_alloc_sem);
4271
4272         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4273                                       OCFS2_JOURNAL_ACCESS_WRITE);
4274         if (ret) {
4275                 mlog_errno(ret);
4276                 goto out;
4277         }
4278
4279         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4280                                      1, 1, &bit_off, &len);
4281         if (ret) {
4282                 mlog_errno(ret);
4283                 goto out;
4284         }
4285
4286         /*
4287          * The bucket may spread in many blocks, and
4288          * we will only touch the 1st block and the last block
4289          * in the whole bucket(one for entry and one for data).
4290          */
4291         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4292
4293         trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4294
4295         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
4296         if (ret) {
4297                 mlog_errno(ret);
4298                 goto out;
4299         }
4300
4301         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4302                                                 OCFS2_JOURNAL_ACCESS_CREATE);
4303         if (ret) {
4304                 mlog_errno(ret);
4305                 goto out;
4306         }
4307
4308         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4309         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4310
4311         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4312
4313         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4314         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4315                offsetof(struct ocfs2_xattr_block, xb_attrs));
4316
4317         xr = &xb->xb_attrs.xb_root;
4318         xr->xt_clusters = cpu_to_le32(1);
4319         xr->xt_last_eb_blk = 0;
4320         xr->xt_list.l_tree_depth = 0;
4321         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4322         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4323
4324         xr->xt_list.l_recs[0].e_cpos = 0;
4325         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4326         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4327
4328         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4329
4330         ocfs2_journal_dirty(handle, xb_bh);
4331
4332 out:
4333         up_write(&oi->ip_alloc_sem);
4334
4335         return ret;
4336 }
4337
4338 static int cmp_xe_offset(const void *a, const void *b)
4339 {
4340         const struct ocfs2_xattr_entry *l = a, *r = b;
4341         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4342         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4343
4344         if (l_name_offset < r_name_offset)
4345                 return 1;
4346         if (l_name_offset > r_name_offset)
4347                 return -1;
4348         return 0;
4349 }
4350
4351 /*
4352  * defrag a xattr bucket if we find that the bucket has some
4353  * holes beteen name/value pairs.
4354  * We will move all the name/value pairs to the end of the bucket
4355  * so that we can spare some space for insertion.
4356  */
4357 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4358                                      handle_t *handle,
4359                                      struct ocfs2_xattr_bucket *bucket)
4360 {
4361         int ret, i;
4362         size_t end, offset, len;
4363         struct ocfs2_xattr_header *xh;
4364         char *entries, *buf, *bucket_buf = NULL;
4365         u64 blkno = bucket_blkno(bucket);
4366         u16 xh_free_start;
4367         size_t blocksize = inode->i_sb->s_blocksize;
4368         struct ocfs2_xattr_entry *xe;
4369
4370         /*
4371          * In order to make the operation more efficient and generic,
4372          * we copy all the blocks into a contiguous memory and do the
4373          * defragment there, so if anything is error, we will not touch
4374          * the real block.
4375          */
4376         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4377         if (!bucket_buf) {
4378                 ret = -EIO;
4379                 goto out;
4380         }
4381
4382         buf = bucket_buf;
4383         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4384                 memcpy(buf, bucket_block(bucket, i), blocksize);
4385
4386         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4387                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4388         if (ret < 0) {
4389                 mlog_errno(ret);
4390                 goto out;
4391         }
4392
4393         xh = (struct ocfs2_xattr_header *)bucket_buf;
4394         entries = (char *)xh->xh_entries;
4395         xh_free_start = le16_to_cpu(xh->xh_free_start);
4396
4397         trace_ocfs2_defrag_xattr_bucket(
4398              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4399              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4400
4401         /*
4402          * sort all the entries by their offset.
4403          * the largest will be the first, so that we can
4404          * move them to the end one by one.
4405          */
4406         sort(entries, le16_to_cpu(xh->xh_count),
4407              sizeof(struct ocfs2_xattr_entry),
4408              cmp_xe_offset, swap_xe);
4409
4410         /* Move all name/values to the end of the bucket. */
4411         xe = xh->xh_entries;
4412         end = OCFS2_XATTR_BUCKET_SIZE;
4413         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4414                 offset = le16_to_cpu(xe->xe_name_offset);
4415                 len = namevalue_size_xe(xe);
4416
4417                 /*
4418                  * We must make sure that the name/value pair
4419                  * exist in the same block. So adjust end to
4420                  * the previous block end if needed.
4421                  */
4422                 if (((end - len) / blocksize !=
4423                         (end - 1) / blocksize))
4424                         end = end - end % blocksize;
4425
4426                 if (end > offset + len) {
4427                         memmove(bucket_buf + end - len,
4428                                 bucket_buf + offset, len);
4429                         xe->xe_name_offset = cpu_to_le16(end - len);
4430                 }
4431
4432                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4433                                 "bucket %llu\n", (unsigned long long)blkno);
4434
4435                 end -= len;
4436         }
4437
4438         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4439                         "bucket %llu\n", (unsigned long long)blkno);
4440
4441         if (xh_free_start == end)
4442                 goto out;
4443
4444         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4445         xh->xh_free_start = cpu_to_le16(end);
4446
4447         /* sort the entries by their name_hash. */
4448         sort(entries, le16_to_cpu(xh->xh_count),
4449              sizeof(struct ocfs2_xattr_entry),
4450              cmp_xe, swap_xe);
4451
4452         buf = bucket_buf;
4453         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4454                 memcpy(bucket_block(bucket, i), buf, blocksize);
4455         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4456
4457 out:
4458         kfree(bucket_buf);
4459         return ret;
4460 }
4461
4462 /*
4463  * prev_blkno points to the start of an existing extent.  new_blkno
4464  * points to a newly allocated extent.  Because we know each of our
4465  * clusters contains more than bucket, we can easily split one cluster
4466  * at a bucket boundary.  So we take the last cluster of the existing
4467  * extent and split it down the middle.  We move the last half of the
4468  * buckets in the last cluster of the existing extent over to the new
4469  * extent.
4470  *
4471  * first_bh is the buffer at prev_blkno so we can update the existing
4472  * extent's bucket count.  header_bh is the bucket were we were hoping
4473  * to insert our xattr.  If the bucket move places the target in the new
4474  * extent, we'll update first_bh and header_bh after modifying the old
4475  * extent.
4476  *
4477  * first_hash will be set as the 1st xe's name_hash in the new extent.
4478  */
4479 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4480                                                handle_t *handle,
4481                                                struct ocfs2_xattr_bucket *first,
4482                                                struct ocfs2_xattr_bucket *target,
4483                                                u64 new_blkno,
4484                                                u32 num_clusters,
4485                                                u32 *first_hash)
4486 {
4487         int ret;
4488         struct super_block *sb = inode->i_sb;
4489         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4490         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4491         int to_move = num_buckets / 2;
4492         u64 src_blkno;
4493         u64 last_cluster_blkno = bucket_blkno(first) +
4494                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4495
4496         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4497         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4498
4499         trace_ocfs2_mv_xattr_bucket_cross_cluster(
4500                                 (unsigned long long)last_cluster_blkno,
4501                                 (unsigned long long)new_blkno);
4502
4503         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4504                                      last_cluster_blkno, new_blkno,
4505                                      to_move, first_hash);
4506         if (ret) {
4507                 mlog_errno(ret);
4508                 goto out;
4509         }
4510
4511         /* This is the first bucket that got moved */
4512         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4513
4514         /*
4515          * If the target bucket was part of the moved buckets, we need to
4516          * update first and target.
4517          */
4518         if (bucket_blkno(target) >= src_blkno) {
4519                 /* Find the block for the new target bucket */
4520                 src_blkno = new_blkno +
4521                         (bucket_blkno(target) - src_blkno);
4522
4523                 ocfs2_xattr_bucket_relse(first);
4524                 ocfs2_xattr_bucket_relse(target);
4525
4526                 /*
4527                  * These shouldn't fail - the buffers are in the
4528                  * journal from ocfs2_cp_xattr_bucket().
4529                  */
4530                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4531                 if (ret) {
4532                         mlog_errno(ret);
4533                         goto out;
4534                 }
4535                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4536                 if (ret)
4537                         mlog_errno(ret);
4538
4539         }
4540
4541 out:
4542         return ret;
4543 }
4544
4545 /*
4546  * Find the suitable pos when we divide a bucket into 2.
4547  * We have to make sure the xattrs with the same hash value exist
4548  * in the same bucket.
4549  *
4550  * If this ocfs2_xattr_header covers more than one hash value, find a
4551  * place where the hash value changes.  Try to find the most even split.
4552  * The most common case is that all entries have different hash values,
4553  * and the first check we make will find a place to split.
4554  */
4555 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4556 {
4557         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4558         int count = le16_to_cpu(xh->xh_count);
4559         int delta, middle = count / 2;
4560
4561         /*
4562          * We start at the middle.  Each step gets farther away in both
4563          * directions.  We therefore hit the change in hash value
4564          * nearest to the middle.  Note that this loop does not execute for
4565          * count < 2.
4566          */
4567         for (delta = 0; delta < middle; delta++) {
4568                 /* Let's check delta earlier than middle */
4569                 if (cmp_xe(&entries[middle - delta - 1],
4570                            &entries[middle - delta]))
4571                         return middle - delta;
4572
4573                 /* For even counts, don't walk off the end */
4574                 if ((middle + delta + 1) == count)
4575                         continue;
4576
4577                 /* Now try delta past middle */
4578                 if (cmp_xe(&entries[middle + delta],
4579                            &entries[middle + delta + 1]))
4580                         return middle + delta + 1;
4581         }
4582
4583         /* Every entry had the same hash */
4584         return count;
4585 }
4586
4587 /*
4588  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4589  * first_hash will record the 1st hash of the new bucket.
4590  *
4591  * Normally half of the xattrs will be moved.  But we have to make
4592  * sure that the xattrs with the same hash value are stored in the
4593  * same bucket. If all the xattrs in this bucket have the same hash
4594  * value, the new bucket will be initialized as an empty one and the
4595  * first_hash will be initialized as (hash_value+1).
4596  */
4597 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4598                                     handle_t *handle,
4599                                     u64 blk,
4600                                     u64 new_blk,
4601                                     u32 *first_hash,
4602                                     int new_bucket_head)
4603 {
4604         int ret, i;
4605         int count, start, len, name_value_len = 0, name_offset = 0;
4606         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4607         struct ocfs2_xattr_header *xh;
4608         struct ocfs2_xattr_entry *xe;
4609         int blocksize = inode->i_sb->s_blocksize;
4610
4611         trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4612                                               (unsigned long long)new_blk);
4613
4614         s_bucket = ocfs2_xattr_bucket_new(inode);
4615         t_bucket = ocfs2_xattr_bucket_new(inode);
4616         if (!s_bucket || !t_bucket) {
4617                 ret = -ENOMEM;
4618                 mlog_errno(ret);
4619                 goto out;
4620         }
4621
4622         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4623         if (ret) {
4624                 mlog_errno(ret);
4625                 goto out;
4626         }
4627
4628         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4629                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4630         if (ret) {
4631                 mlog_errno(ret);
4632                 goto out;
4633         }
4634
4635         /*
4636          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4637          * there's no need to read it.
4638          */
4639         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4640         if (ret) {
4641                 mlog_errno(ret);
4642                 goto out;
4643         }
4644
4645         /*
4646          * Hey, if we're overwriting t_bucket, what difference does
4647          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4648          * same part of ocfs2_cp_xattr_bucket().
4649          */
4650         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4651                                                 new_bucket_head ?
4652                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4653                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4654         if (ret) {
4655                 mlog_errno(ret);
4656                 goto out;
4657         }
4658
4659         xh = bucket_xh(s_bucket);
4660         count = le16_to_cpu(xh->xh_count);
4661         start = ocfs2_xattr_find_divide_pos(xh);
4662
4663         if (start == count) {
4664                 xe = &xh->xh_entries[start-1];
4665
4666                 /*
4667                  * initialized a new empty bucket here.
4668                  * The hash value is set as one larger than
4669                  * that of the last entry in the previous bucket.
4670                  */
4671                 for (i = 0; i < t_bucket->bu_blocks; i++)
4672                         memset(bucket_block(t_bucket, i), 0, blocksize);
4673
4674                 xh = bucket_xh(t_bucket);
4675                 xh->xh_free_start = cpu_to_le16(blocksize);
4676                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4677                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4678
4679                 goto set_num_buckets;
4680         }
4681
4682         /* copy the whole bucket to the new first. */
4683         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4684
4685         /* update the new bucket. */
4686         xh = bucket_xh(t_bucket);
4687
4688         /*
4689          * Calculate the total name/value len and xh_free_start for
4690          * the old bucket first.
4691          */
4692         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4693         name_value_len = 0;
4694         for (i = 0; i < start; i++) {
4695                 xe = &xh->xh_entries[i];
4696                 name_value_len += namevalue_size_xe(xe);
4697                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4698                         name_offset = le16_to_cpu(xe->xe_name_offset);
4699         }
4700
4701         /*
4702          * Now begin the modification to the new bucket.
4703          *
4704          * In the new bucket, We just move the xattr entry to the beginning
4705          * and don't touch the name/value. So there will be some holes in the
4706          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4707          * called.
4708          */
4709         xe = &xh->xh_entries[start];
4710         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4711         trace_ocfs2_divide_xattr_bucket_move(len,
4712                         (int)((char *)xe - (char *)xh),
4713                         (int)((char *)xh->xh_entries - (char *)xh));
4714         memmove((char *)xh->xh_entries, (char *)xe, len);
4715         xe = &xh->xh_entries[count - start];
4716         len = sizeof(struct ocfs2_xattr_entry) * start;
4717         memset((char *)xe, 0, len);
4718
4719         le16_add_cpu(&xh->xh_count, -start);
4720         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4721
4722         /* Calculate xh_free_start for the new bucket. */
4723         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4724         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4725                 xe = &xh->xh_entries[i];
4726                 if (le16_to_cpu(xe->xe_name_offset) <
4727                     le16_to_cpu(xh->xh_free_start))
4728                         xh->xh_free_start = xe->xe_name_offset;
4729         }
4730
4731 set_num_buckets:
4732         /* set xh->xh_num_buckets for the new xh. */
4733         if (new_bucket_head)
4734                 xh->xh_num_buckets = cpu_to_le16(1);
4735         else
4736                 xh->xh_num_buckets = 0;
4737
4738         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4739
4740         /* store the first_hash of the new bucket. */
4741         if (first_hash)
4742                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4743
4744         /*
4745          * Now only update the 1st block of the old bucket.  If we
4746          * just added a new empty bucket, there is no need to modify
4747          * it.
4748          */
4749         if (start == count)
4750                 goto out;
4751
4752         xh = bucket_xh(s_bucket);
4753         memset(&xh->xh_entries[start], 0,
4754                sizeof(struct ocfs2_xattr_entry) * (count - start));
4755         xh->xh_count = cpu_to_le16(start);
4756         xh->xh_free_start = cpu_to_le16(name_offset);
4757         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4758
4759         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4760
4761 out:
4762         ocfs2_xattr_bucket_free(s_bucket);
4763         ocfs2_xattr_bucket_free(t_bucket);
4764
4765         return ret;
4766 }
4767
4768 /*
4769  * Copy xattr from one bucket to another bucket.
4770  *
4771  * The caller must make sure that the journal transaction
4772  * has enough space for journaling.
4773  */
4774 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4775                                  handle_t *handle,
4776                                  u64 s_blkno,
4777                                  u64 t_blkno,
4778                                  int t_is_new)
4779 {
4780         int ret;
4781         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4782
4783         BUG_ON(s_blkno == t_blkno);
4784
4785         trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4786                                     (unsigned long long)t_blkno,
4787                                     t_is_new);
4788
4789         s_bucket = ocfs2_xattr_bucket_new(inode);
4790         t_bucket = ocfs2_xattr_bucket_new(inode);
4791         if (!s_bucket || !t_bucket) {
4792                 ret = -ENOMEM;
4793                 mlog_errno(ret);
4794                 goto out;
4795         }
4796
4797         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4798         if (ret)
4799                 goto out;
4800
4801         /*
4802          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4803          * there's no need to read it.
4804          */
4805         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4806         if (ret)
4807                 goto out;
4808
4809         /*
4810          * Hey, if we're overwriting t_bucket, what difference does
4811          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4812          * cluster to fill, we came here from
4813          * ocfs2_mv_xattr_buckets(), and it is really new -
4814          * ACCESS_CREATE is required.  But we also might have moved data
4815          * out of t_bucket before extending back into it.
4816          * ocfs2_add_new_xattr_bucket() can do this - its call to
4817          * ocfs2_add_new_xattr_cluster() may have created a new extent
4818          * and copied out the end of the old extent.  Then it re-extends
4819          * the old extent back to create space for new xattrs.  That's
4820          * how we get here, and the bucket isn't really new.
4821          */
4822         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4823                                                 t_is_new ?
4824                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4825                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4826         if (ret)
4827                 goto out;
4828
4829         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4830         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4831
4832 out:
4833         ocfs2_xattr_bucket_free(t_bucket);
4834         ocfs2_xattr_bucket_free(s_bucket);
4835
4836         return ret;
4837 }
4838
4839 /*
4840  * src_blk points to the start of an existing extent.  last_blk points to
4841  * last cluster in that extent.  to_blk points to a newly allocated
4842  * extent.  We copy the buckets from the cluster at last_blk to the new
4843  * extent.  If start_bucket is non-zero, we skip that many buckets before
4844  * we start copying.  The new extent's xh_num_buckets gets set to the
4845  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4846  * by the same amount.
4847  */
4848 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4849                                   u64 src_blk, u64 last_blk, u64 to_blk,
4850                                   unsigned int start_bucket,
4851                                   u32 *first_hash)
4852 {
4853         int i, ret, credits;
4854         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4855         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4856         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4857         struct ocfs2_xattr_bucket *old_first, *new_first;
4858
4859         trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4860                                      (unsigned long long)to_blk);
4861
4862         BUG_ON(start_bucket >= num_buckets);
4863         if (start_bucket) {
4864                 num_buckets -= start_bucket;
4865                 last_blk += (start_bucket * blks_per_bucket);
4866         }
4867
4868         /* The first bucket of the original extent */
4869         old_first = ocfs2_xattr_bucket_new(inode);
4870         /* The first bucket of the new extent */
4871         new_first = ocfs2_xattr_bucket_new(inode);
4872         if (!old_first || !new_first) {
4873                 ret = -ENOMEM;
4874                 mlog_errno(ret);
4875                 goto out;
4876         }
4877
4878         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4879         if (ret) {
4880                 mlog_errno(ret);
4881                 goto out;
4882         }
4883
4884         /*
4885          * We need to update the first bucket of the old extent and all
4886          * the buckets going to the new extent.
4887          */
4888         credits = ((num_buckets + 1) * blks_per_bucket);
4889         ret = ocfs2_extend_trans(handle, credits);
4890         if (ret) {
4891                 mlog_errno(ret);
4892                 goto out;
4893         }
4894
4895         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4896                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4897         if (ret) {
4898                 mlog_errno(ret);
4899                 goto out;
4900         }
4901
4902         for (i = 0; i < num_buckets; i++) {
4903                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4904                                             last_blk + (i * blks_per_bucket),
4905                                             to_blk + (i * blks_per_bucket),
4906                                             1);
4907                 if (ret) {
4908                         mlog_errno(ret);
4909                         goto out;
4910                 }
4911         }
4912
4913         /*
4914          * Get the new bucket ready before we dirty anything
4915          * (This actually shouldn't fail, because we already dirtied
4916          * it once in ocfs2_cp_xattr_bucket()).
4917          */
4918         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4919         if (ret) {
4920                 mlog_errno(ret);
4921                 goto out;
4922         }
4923         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4924                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4925         if (ret) {
4926                 mlog_errno(ret);
4927                 goto out;
4928         }
4929
4930         /* Now update the headers */
4931         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4932         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4933
4934         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4935         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4936
4937         if (first_hash)
4938                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4939
4940 out:
4941         ocfs2_xattr_bucket_free(new_first);
4942         ocfs2_xattr_bucket_free(old_first);
4943         return ret;
4944 }
4945
4946 /*
4947  * Move some xattrs in this cluster to the new cluster.
4948  * This function should only be called when bucket size == cluster size.
4949  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4950  */
4951 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4952                                       handle_t *handle,
4953                                       u64 prev_blk,
4954                                       u64 new_blk,
4955                                       u32 *first_hash)
4956 {
4957         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4958         int ret, credits = 2 * blk_per_bucket;
4959
4960         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4961
4962         ret = ocfs2_extend_trans(handle, credits);
4963         if (ret) {
4964                 mlog_errno(ret);
4965                 return ret;
4966         }
4967
4968         /* Move half of the xattr in start_blk to the next bucket. */
4969         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4970                                           new_blk, first_hash, 1);
4971 }
4972
4973 /*
4974  * Move some xattrs from the old cluster to the new one since they are not
4975  * contiguous in ocfs2 xattr tree.
4976  *
4977  * new_blk starts a new separate cluster, and we will move some xattrs from
4978  * prev_blk to it. v_start will be set as the first name hash value in this
4979  * new cluster so that it can be used as e_cpos during tree insertion and
4980  * don't collide with our original b-tree operations. first_bh and header_bh
4981  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4982  * to extend the insert bucket.
4983  *
4984  * The problem is how much xattr should we move to the new one and when should
4985  * we update first_bh and header_bh?
4986  * 1. If cluster size > bucket size, that means the previous cluster has more
4987  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4988  *    update the first_bh and header_bh if the insert bucket has been moved
4989  *    to the new cluster.
4990  * 2. If cluster_size == bucket_size:
4991  *    a) If the previous extent rec has more than one cluster and the insert
4992  *       place isn't in the last cluster, copy the entire last cluster to the
4993  *       new one. This time, we don't need to upate the first_bh and header_bh
4994  *       since they will not be moved into the new cluster.
4995  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4996  *       the new one. And we set the extend flag to zero if the insert place is
4997  *       moved into the new allocated cluster since no extend is needed.
4998  */
4999 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5000                                             handle_t *handle,
5001                                             struct ocfs2_xattr_bucket *first,
5002                                             struct ocfs2_xattr_bucket *target,
5003                                             u64 new_blk,
5004                                             u32 prev_clusters,
5005                                             u32 *v_start,
5006                                             int *extend)
5007 {
5008         int ret;
5009
5010         trace_ocfs2_adjust_xattr_cross_cluster(
5011                         (unsigned long long)bucket_blkno(first),
5012                         (unsigned long long)new_blk, prev_clusters);
5013
5014         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5015                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5016                                                           handle,
5017                                                           first, target,
5018                                                           new_blk,
5019                                                           prev_clusters,
5020                                                           v_start);
5021                 if (ret)
5022                         mlog_errno(ret);
5023         } else {
5024                 /* The start of the last cluster in the first extent */
5025                 u64 last_blk = bucket_blkno(first) +
5026                         ((prev_clusters - 1) *
5027                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
5028
5029                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5030                         ret = ocfs2_mv_xattr_buckets(inode, handle,
5031                                                      bucket_blkno(first),
5032                                                      last_blk, new_blk, 0,
5033                                                      v_start);
5034                         if (ret)
5035                                 mlog_errno(ret);
5036                 } else {
5037                         ret = ocfs2_divide_xattr_cluster(inode, handle,
5038                                                          last_blk, new_blk,
5039                                                          v_start);
5040                         if (ret)
5041                                 mlog_errno(ret);
5042
5043                         if ((bucket_blkno(target) == last_blk) && extend)
5044                                 *extend = 0;
5045                 }
5046         }
5047
5048         return ret;
5049 }
5050
5051 /*
5052  * Add a new cluster for xattr storage.
5053  *
5054  * If the new cluster is contiguous with the previous one, it will be
5055  * appended to the same extent record, and num_clusters will be updated.
5056  * If not, we will insert a new extent for it and move some xattrs in
5057  * the last cluster into the new allocated one.
5058  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5059  * lose the benefits of hashing because we'll have to search large leaves.
5060  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5061  * if it's bigger).
5062  *
5063  * first_bh is the first block of the previous extent rec and header_bh
5064  * indicates the bucket we will insert the new xattrs. They will be updated
5065  * when the header_bh is moved into the new cluster.
5066  */
5067 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5068                                        struct buffer_head *root_bh,
5069                                        struct ocfs2_xattr_bucket *first,
5070                                        struct ocfs2_xattr_bucket *target,
5071                                        u32 *num_clusters,
5072                                        u32 prev_cpos,
5073                                        int *extend,
5074                                        struct ocfs2_xattr_set_ctxt *ctxt)
5075 {
5076         int ret;
5077         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5078         u32 prev_clusters = *num_clusters;
5079         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5080         u64 block;
5081         handle_t *handle = ctxt->handle;
5082         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5083         struct ocfs2_extent_tree et;
5084
5085         trace_ocfs2_add_new_xattr_cluster_begin(
5086                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5087                 (unsigned long long)bucket_blkno(first),
5088                 prev_cpos, prev_clusters);
5089
5090         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5091
5092         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5093                                       OCFS2_JOURNAL_ACCESS_WRITE);
5094         if (ret < 0) {
5095                 mlog_errno(ret);
5096                 goto leave;
5097         }
5098
5099         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5100                                      clusters_to_add, &bit_off, &num_bits);
5101         if (ret < 0) {
5102                 if (ret != -ENOSPC)
5103                         mlog_errno(ret);
5104                 goto leave;
5105         }
5106
5107         BUG_ON(num_bits > clusters_to_add);
5108
5109         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5110         trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5111
5112         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5113             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5114              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5115                 /*
5116                  * If this cluster is contiguous with the old one and
5117                  * adding this new cluster, we don't surpass the limit of
5118                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5119                  * initialized and used like other buckets in the previous
5120                  * cluster.
5121                  * So add it as a contiguous one. The caller will handle
5122                  * its init process.
5123                  */
5124                 v_start = prev_cpos + prev_clusters;
5125                 *num_clusters = prev_clusters + num_bits;
5126         } else {
5127                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5128                                                        handle,
5129                                                        first,
5130                                                        target,
5131                                                        block,
5132                                                        prev_clusters,
5133                                                        &v_start,
5134                                                        extend);
5135                 if (ret) {
5136                         mlog_errno(ret);
5137                         goto leave;
5138                 }
5139         }
5140
5141         trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5142                                                  v_start, num_bits);
5143         ret = ocfs2_insert_extent(handle, &et, v_start, block,
5144                                   num_bits, 0, ctxt->meta_ac);
5145         if (ret < 0) {
5146                 mlog_errno(ret);
5147                 goto leave;
5148         }
5149
5150         ocfs2_journal_dirty(handle, root_bh);
5151
5152 leave:
5153         return ret;
5154 }
5155
5156 /*
5157  * We are given an extent.  'first' is the bucket at the very front of
5158  * the extent.  The extent has space for an additional bucket past
5159  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5160  * of the target bucket.  We wish to shift every bucket past the target
5161  * down one, filling in that additional space.  When we get back to the
5162  * target, we split the target between itself and the now-empty bucket
5163  * at target+1 (aka, target_blkno + blks_per_bucket).
5164  */
5165 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5166                                      handle_t *handle,
5167                                      struct ocfs2_xattr_bucket *first,
5168                                      u64 target_blk,
5169                                      u32 num_clusters)
5170 {
5171         int ret, credits;
5172         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5173         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5174         u64 end_blk;
5175         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5176
5177         trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5178                                         (unsigned long long)bucket_blkno(first),
5179                                         num_clusters, new_bucket);
5180
5181         /* The extent must have room for an additional bucket */
5182         BUG_ON(new_bucket >=
5183                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5184
5185         /* end_blk points to the last existing bucket */
5186         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5187
5188         /*
5189          * end_blk is the start of the last existing bucket.
5190          * Thus, (end_blk - target_blk) covers the target bucket and
5191          * every bucket after it up to, but not including, the last
5192          * existing bucket.  Then we add the last existing bucket, the
5193          * new bucket, and the first bucket (3 * blk_per_bucket).
5194          */
5195         credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5196         ret = ocfs2_extend_trans(handle, credits);
5197         if (ret) {
5198                 mlog_errno(ret);
5199                 goto out;
5200         }
5201
5202         ret = ocfs2_xattr_bucket_journal_access(handle, first,
5203                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5204         if (ret) {
5205                 mlog_errno(ret);
5206                 goto out;
5207         }
5208
5209         while (end_blk != target_blk) {
5210                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5211                                             end_blk + blk_per_bucket, 0);
5212                 if (ret)
5213                         goto out;
5214                 end_blk -= blk_per_bucket;
5215         }
5216
5217         /* Move half of the xattr in target_blkno to the next bucket. */
5218         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5219                                         target_blk + blk_per_bucket, NULL, 0);
5220
5221         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5222         ocfs2_xattr_bucket_journal_dirty(handle, first);
5223
5224 out:
5225         return ret;
5226 }
5227
5228 /*
5229  * Add new xattr bucket in an extent record and adjust the buckets
5230  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5231  * bucket we want to insert into.
5232  *
5233  * In the easy case, we will move all the buckets after target down by
5234  * one. Half of target's xattrs will be moved to the next bucket.
5235  *
5236  * If current cluster is full, we'll allocate a new one.  This may not
5237  * be contiguous.  The underlying calls will make sure that there is
5238  * space for the insert, shifting buckets around if necessary.
5239  * 'target' may be moved by those calls.
5240  */
5241 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5242                                       struct buffer_head *xb_bh,
5243                                       struct ocfs2_xattr_bucket *target,
5244                                       struct ocfs2_xattr_set_ctxt *ctxt)
5245 {
5246         struct ocfs2_xattr_block *xb =
5247                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5248         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5249         struct ocfs2_extent_list *el = &xb_root->xt_list;
5250         u32 name_hash =
5251                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5252         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5253         int ret, num_buckets, extend = 1;
5254         u64 p_blkno;
5255         u32 e_cpos, num_clusters;
5256         /* The bucket at the front of the extent */
5257         struct ocfs2_xattr_bucket *first;
5258
5259         trace_ocfs2_add_new_xattr_bucket(
5260                                 (unsigned long long)bucket_blkno(target));
5261
5262         /* The first bucket of the original extent */
5263         first = ocfs2_xattr_bucket_new(inode);
5264         if (!first) {
5265                 ret = -ENOMEM;
5266                 mlog_errno(ret);
5267                 goto out;
5268         }
5269
5270         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5271                                   &num_clusters, el);
5272         if (ret) {
5273                 mlog_errno(ret);
5274                 goto out;
5275         }
5276
5277         ret = ocfs2_read_xattr_bucket(first, p_blkno);
5278         if (ret) {
5279                 mlog_errno(ret);
5280                 goto out;
5281         }
5282
5283         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5284         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5285                 /*
5286                  * This can move first+target if the target bucket moves
5287                  * to the new extent.
5288                  */
5289                 ret = ocfs2_add_new_xattr_cluster(inode,
5290                                                   xb_bh,
5291                                                   first,
5292                                                   target,
5293                                                   &num_clusters,
5294                                                   e_cpos,
5295                                                   &extend,
5296                                                   ctxt);
5297                 if (ret) {
5298                         mlog_errno(ret);
5299                         goto out;
5300                 }
5301         }
5302
5303         if (extend) {
5304                 ret = ocfs2_extend_xattr_bucket(inode,
5305                                                 ctxt->handle,
5306                                                 first,
5307                                                 bucket_blkno(target),
5308                                                 num_clusters);
5309                 if (ret)
5310                         mlog_errno(ret);
5311         }
5312
5313 out:
5314         ocfs2_xattr_bucket_free(first);
5315
5316         return ret;
5317 }
5318
5319 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
5320                                         struct ocfs2_xattr_bucket *bucket,
5321                                         int offs)
5322 {
5323         int block_off = offs >> inode->i_sb->s_blocksize_bits;
5324
5325         offs = offs % inode->i_sb->s_blocksize;
5326         return bucket_block(bucket, block_off) + offs;
5327 }
5328
5329 /*
5330  * Truncate the specified xe_off entry in xattr bucket.
5331  * bucket is indicated by header_bh and len is the new length.
5332  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5333  *
5334  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5335  */
5336 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5337                                              struct ocfs2_xattr_bucket *bucket,
5338                                              int xe_off,
5339                                              int len,
5340                                              struct ocfs2_xattr_set_ctxt *ctxt)
5341 {
5342         int ret, offset;
5343         u64 value_blk;
5344         struct ocfs2_xattr_entry *xe;
5345         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5346         size_t blocksize = inode->i_sb->s_blocksize;
5347         struct ocfs2_xattr_value_buf vb = {
5348                 .vb_access = ocfs2_journal_access,
5349         };
5350
5351         xe = &xh->xh_entries[xe_off];
5352
5353         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5354
5355         offset = le16_to_cpu(xe->xe_name_offset) +
5356                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5357
5358         value_blk = offset / blocksize;
5359
5360         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5361         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5362
5363         vb.vb_bh = bucket->bu_bhs[value_blk];
5364         BUG_ON(!vb.vb_bh);
5365
5366         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5367                 (vb.vb_bh->b_data + offset % blocksize);
5368
5369         /*
5370          * From here on out we have to dirty the bucket.  The generic
5371          * value calls only modify one of the bucket's bhs, but we need
5372          * to send the bucket at once.  So if they error, they *could* have
5373          * modified something.  We have to assume they did, and dirty
5374          * the whole bucket.  This leaves us in a consistent state.
5375          */
5376         trace_ocfs2_xattr_bucket_value_truncate(
5377                         (unsigned long long)bucket_blkno(bucket), xe_off, len);
5378         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5379         if (ret) {
5380                 mlog_errno(ret);
5381                 goto out;
5382         }
5383
5384         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5385                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5386         if (ret) {
5387                 mlog_errno(ret);
5388                 goto out;
5389         }
5390
5391         xe->xe_value_size = cpu_to_le64(len);
5392
5393         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5394
5395 out:
5396         return ret;
5397 }
5398
5399 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5400                                   struct buffer_head *root_bh,
5401                                   u64 blkno,
5402                                   u32 cpos,
5403                                   u32 len,
5404                                   void *para)
5405 {
5406         int ret;
5407         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5408         struct inode *tl_inode = osb->osb_tl_inode;
5409         handle_t *handle;
5410         struct ocfs2_xattr_block *xb =
5411                         (struct ocfs2_xattr_block *)root_bh->b_data;
5412         struct ocfs2_alloc_context *meta_ac = NULL;
5413         struct ocfs2_cached_dealloc_ctxt dealloc;
5414         struct ocfs2_extent_tree et;
5415
5416         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5417                                           ocfs2_delete_xattr_in_bucket, para);
5418         if (ret) {
5419                 mlog_errno(ret);
5420                 return ret;
5421         }
5422
5423         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5424
5425         ocfs2_init_dealloc_ctxt(&dealloc);
5426
5427         trace_ocfs2_rm_xattr_cluster(
5428                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
5429                         (unsigned long long)blkno, cpos, len);
5430
5431         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5432                                                len);
5433
5434         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5435         if (ret) {
5436                 mlog_errno(ret);
5437                 return ret;
5438         }
5439
5440         mutex_lock(&tl_inode->i_mutex);
5441
5442         if (ocfs2_truncate_log_needs_flush(osb)) {
5443                 ret = __ocfs2_flush_truncate_log(osb);
5444                 if (ret < 0) {
5445                         mlog_errno(ret);
5446                         goto out;
5447                 }
5448         }
5449
5450         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5451         if (IS_ERR(handle)) {
5452                 ret = -ENOMEM;
5453                 mlog_errno(ret);
5454                 goto out;
5455         }
5456
5457         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5458                                       OCFS2_JOURNAL_ACCESS_WRITE);
5459         if (ret) {
5460                 mlog_errno(ret);
5461                 goto out_commit;
5462         }
5463
5464         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5465                                   &dealloc);
5466         if (ret) {
5467                 mlog_errno(ret);
5468                 goto out_commit;
5469         }
5470
5471         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5472         ocfs2_journal_dirty(handle, root_bh);
5473
5474         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5475         if (ret)
5476                 mlog_errno(ret);
5477
5478 out_commit:
5479         ocfs2_commit_trans(osb, handle);
5480 out:
5481         ocfs2_schedule_truncate_log_flush(osb, 1);
5482
5483         mutex_unlock(&tl_inode->i_mutex);
5484
5485         if (meta_ac)
5486                 ocfs2_free_alloc_context(meta_ac);
5487
5488         ocfs2_run_deallocs(osb, &dealloc);
5489
5490         return ret;
5491 }
5492
5493 /*
5494  * check whether the xattr bucket is filled up with the same hash value.
5495  * If we want to insert the xattr with the same hash, return -ENOSPC.
5496  * If we want to insert a xattr with different hash value, go ahead
5497  * and ocfs2_divide_xattr_bucket will handle this.
5498  */
5499 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5500                                               struct ocfs2_xattr_bucket *bucket,
5501                                               const char *name)
5502 {
5503         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5504         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5505
5506         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5507                 return 0;
5508
5509         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5510             xh->xh_entries[0].xe_name_hash) {
5511                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5512                      "hash = %u\n",
5513                      (unsigned long long)bucket_blkno(bucket),
5514                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5515                 return -ENOSPC;
5516         }
5517
5518         return 0;
5519 }
5520
5521 /*
5522  * Try to set the entry in the current bucket.  If we fail, the caller
5523  * will handle getting us another bucket.
5524  */
5525 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5526                                         struct ocfs2_xattr_info *xi,
5527                                         struct ocfs2_xattr_search *xs,
5528                                         struct ocfs2_xattr_set_ctxt *ctxt)
5529 {
5530         int ret;
5531         struct ocfs2_xa_loc loc;
5532
5533         trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5534
5535         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5536                                        xs->not_found ? NULL : xs->here);
5537         ret = ocfs2_xa_set(&loc, xi, ctxt);
5538         if (!ret) {
5539                 xs->here = loc.xl_entry;
5540                 goto out;
5541         }
5542         if (ret != -ENOSPC) {
5543                 mlog_errno(ret);
5544                 goto out;
5545         }
5546
5547         /* Ok, we need space.  Let's try defragmenting the bucket. */
5548         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5549                                         xs->bucket);
5550         if (ret) {
5551                 mlog_errno(ret);
5552                 goto out;
5553         }
5554
5555         ret = ocfs2_xa_set(&loc, xi, ctxt);
5556         if (!ret) {
5557                 xs->here = loc.xl_entry;
5558                 goto out;
5559         }
5560         if (ret != -ENOSPC)
5561                 mlog_errno(ret);
5562
5563
5564 out:
5565         return ret;
5566 }
5567
5568 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5569                                              struct ocfs2_xattr_info *xi,
5570                                              struct ocfs2_xattr_search *xs,
5571                                              struct ocfs2_xattr_set_ctxt *ctxt)
5572 {
5573         int ret;
5574
5575         trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5576
5577         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5578         if (!ret)
5579                 goto out;
5580         if (ret != -ENOSPC) {
5581                 mlog_errno(ret);
5582                 goto out;
5583         }
5584
5585         /* Ack, need more space.  Let's try to get another bucket! */
5586
5587         /*
5588          * We do not allow for overlapping ranges between buckets. And
5589          * the maximum number of collisions we will allow for then is
5590          * one bucket's worth, so check it here whether we need to
5591          * add a new bucket for the insert.
5592          */
5593         ret = ocfs2_check_xattr_bucket_collision(inode,
5594                                                  xs->bucket,
5595                                                  xi->xi_name);
5596         if (ret) {
5597                 mlog_errno(ret);
5598                 goto out;
5599         }
5600
5601         ret = ocfs2_add_new_xattr_bucket(inode,
5602                                          xs->xattr_bh,
5603                                          xs->bucket,
5604                                          ctxt);
5605         if (ret) {
5606                 mlog_errno(ret);
5607                 goto out;
5608         }
5609
5610         /*
5611          * ocfs2_add_new_xattr_bucket() will have updated
5612          * xs->bucket if it moved, but it will not have updated
5613          * any of the other search fields.  Thus, we drop it and
5614          * re-search.  Everything should be cached, so it'll be
5615          * quick.
5616          */
5617         ocfs2_xattr_bucket_relse(xs->bucket);
5618         ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5619                                            xi->xi_name_index,
5620                                            xi->xi_name, xs);
5621         if (ret && ret != -ENODATA)
5622                 goto out;
5623         xs->not_found = ret;
5624
5625         /* Ok, we have a new bucket, let's try again */
5626         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5627         if (ret && (ret != -ENOSPC))
5628                 mlog_errno(ret);
5629
5630 out:
5631         return ret;
5632 }
5633
5634 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5635                                         struct ocfs2_xattr_bucket *bucket,
5636                                         void *para)
5637 {
5638         int ret = 0, ref_credits;
5639         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5640         u16 i;
5641         struct ocfs2_xattr_entry *xe;
5642         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5643         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5644         int credits = ocfs2_remove_extent_credits(osb->sb) +
5645                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5646         struct ocfs2_xattr_value_root *xv;
5647         struct ocfs2_rm_xattr_bucket_para *args =
5648                         (struct ocfs2_rm_xattr_bucket_para *)para;
5649
5650         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5651
5652         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5653                 xe = &xh->xh_entries[i];
5654                 if (ocfs2_xattr_is_local(xe))
5655                         continue;
5656
5657                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5658                                                       i, &xv, NULL);
5659
5660                 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5661                                                          args->ref_ci,
5662                                                          args->ref_root_bh,
5663                                                          &ctxt.meta_ac,
5664                                                          &ref_credits);
5665
5666                 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5667                 if (IS_ERR(ctxt.handle)) {
5668                         ret = PTR_ERR(ctxt.handle);
5669                         mlog_errno(ret);
5670                         break;
5671                 }
5672
5673                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5674                                                         i, 0, &ctxt);
5675
5676                 ocfs2_commit_trans(osb, ctxt.handle);
5677                 if (ctxt.meta_ac) {
5678                         ocfs2_free_alloc_context(ctxt.meta_ac);
5679                         ctxt.meta_ac = NULL;
5680                 }
5681                 if (ret) {
5682                         mlog_errno(ret);
5683                         break;
5684                 }
5685         }
5686
5687         if (ctxt.meta_ac)
5688                 ocfs2_free_alloc_context(ctxt.meta_ac);
5689         ocfs2_schedule_truncate_log_flush(osb, 1);
5690         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5691         return ret;
5692 }
5693
5694 /*
5695  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5696  * or change the extent record flag), we need to recalculate
5697  * the metaecc for the whole bucket. So it is done here.
5698  *
5699  * Note:
5700  * We have to give the extra credits for the caller.
5701  */
5702 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5703                                             handle_t *handle,
5704                                             void *para)
5705 {
5706         int ret;
5707         struct ocfs2_xattr_bucket *bucket =
5708                         (struct ocfs2_xattr_bucket *)para;
5709
5710         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5711                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5712         if (ret) {
5713                 mlog_errno(ret);
5714                 return ret;
5715         }
5716
5717         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5718
5719         return 0;
5720 }
5721
5722 /*
5723  * Special action we need if the xattr value is refcounted.
5724  *
5725  * 1. If the xattr is refcounted, lock the tree.
5726  * 2. CoW the xattr if we are setting the new value and the value
5727  *    will be stored outside.
5728  * 3. In other case, decrease_refcount will work for us, so just
5729  *    lock the refcount tree, calculate the meta and credits is OK.
5730  *
5731  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5732  * currently CoW is a completed transaction, while this function
5733  * will also lock the allocators and let us deadlock. So we will
5734  * CoW the whole xattr value.
5735  */
5736 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5737                                         struct ocfs2_dinode *di,
5738                                         struct ocfs2_xattr_info *xi,
5739                                         struct ocfs2_xattr_search *xis,
5740                                         struct ocfs2_xattr_search *xbs,
5741                                         struct ocfs2_refcount_tree **ref_tree,
5742                                         int *meta_add,
5743                                         int *credits)
5744 {
5745         int ret = 0;
5746         struct ocfs2_xattr_block *xb;
5747         struct ocfs2_xattr_entry *xe;
5748         char *base;
5749         u32 p_cluster, num_clusters;
5750         unsigned int ext_flags;
5751         int name_offset, name_len;
5752         struct ocfs2_xattr_value_buf vb;
5753         struct ocfs2_xattr_bucket *bucket = NULL;
5754         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5755         struct ocfs2_post_refcount refcount;
5756         struct ocfs2_post_refcount *p = NULL;
5757         struct buffer_head *ref_root_bh = NULL;
5758
5759         if (!xis->not_found) {
5760                 xe = xis->here;
5761                 name_offset = le16_to_cpu(xe->xe_name_offset);
5762                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5763                 base = xis->base;
5764                 vb.vb_bh = xis->inode_bh;
5765                 vb.vb_access = ocfs2_journal_access_di;
5766         } else {
5767                 int i, block_off = 0;
5768                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5769                 xe = xbs->here;
5770                 name_offset = le16_to_cpu(xe->xe_name_offset);
5771                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5772                 i = xbs->here - xbs->header->xh_entries;
5773
5774                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5775                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5776                                                         bucket_xh(xbs->bucket),
5777                                                         i, &block_off,
5778                                                         &name_offset);
5779                         if (ret) {
5780                                 mlog_errno(ret);
5781                                 goto out;
5782                         }
5783                         base = bucket_block(xbs->bucket, block_off);
5784                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5785                         vb.vb_access = ocfs2_journal_access;
5786
5787                         if (ocfs2_meta_ecc(osb)) {
5788                                 /*create parameters for ocfs2_post_refcount. */
5789                                 bucket = xbs->bucket;
5790                                 refcount.credits = bucket->bu_blocks;
5791                                 refcount.para = bucket;
5792                                 refcount.func =
5793                                         ocfs2_xattr_bucket_post_refcount;
5794                                 p = &refcount;
5795                         }
5796                 } else {
5797                         base = xbs->base;
5798                         vb.vb_bh = xbs->xattr_bh;
5799                         vb.vb_access = ocfs2_journal_access_xb;
5800                 }
5801         }
5802
5803         if (ocfs2_xattr_is_local(xe))
5804                 goto out;
5805
5806         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5807                                 (base + name_offset + name_len);
5808
5809         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5810                                        &num_clusters, &vb.vb_xv->xr_list,
5811                                        &ext_flags);
5812         if (ret) {
5813                 mlog_errno(ret);
5814                 goto out;
5815         }
5816
5817         /*
5818          * We just need to check the 1st extent record, since we always
5819          * CoW the whole xattr. So there shouldn't be a xattr with
5820          * some REFCOUNT extent recs after the 1st one.
5821          */
5822         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5823                 goto out;
5824
5825         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5826                                        1, ref_tree, &ref_root_bh);
5827         if (ret) {
5828                 mlog_errno(ret);
5829                 goto out;
5830         }
5831
5832         /*
5833          * If we are deleting the xattr or the new size will be stored inside,
5834          * cool, leave it there, the xattr truncate process will remove them
5835          * for us(it still needs the refcount tree lock and the meta, credits).
5836          * And the worse case is that every cluster truncate will split the
5837          * refcount tree, and make the original extent become 3. So we will need
5838          * 2 * cluster more extent recs at most.
5839          */
5840         if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5841
5842                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5843                                                          &(*ref_tree)->rf_ci,
5844                                                          ref_root_bh, vb.vb_xv,
5845                                                          meta_add, credits);
5846                 if (ret)
5847                         mlog_errno(ret);
5848                 goto out;
5849         }
5850
5851         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5852                                        *ref_tree, ref_root_bh, 0,
5853                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5854         if (ret)
5855                 mlog_errno(ret);
5856
5857 out:
5858         brelse(ref_root_bh);
5859         return ret;
5860 }
5861
5862 /*
5863  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5864  * The physical clusters will be added to refcount tree.
5865  */
5866 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5867                                 struct ocfs2_xattr_value_root *xv,
5868                                 struct ocfs2_extent_tree *value_et,
5869                                 struct ocfs2_caching_info *ref_ci,
5870                                 struct buffer_head *ref_root_bh,
5871                                 struct ocfs2_cached_dealloc_ctxt *dealloc,
5872                                 struct ocfs2_post_refcount *refcount)
5873 {
5874         int ret = 0;
5875         u32 clusters = le32_to_cpu(xv->xr_clusters);
5876         u32 cpos, p_cluster, num_clusters;
5877         struct ocfs2_extent_list *el = &xv->xr_list;
5878         unsigned int ext_flags;
5879
5880         cpos = 0;
5881         while (cpos < clusters) {
5882                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5883                                                &num_clusters, el, &ext_flags);
5884
5885                 cpos += num_clusters;
5886                 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5887                         continue;
5888
5889                 BUG_ON(!p_cluster);
5890
5891                 ret = ocfs2_add_refcount_flag(inode, value_et,
5892                                               ref_ci, ref_root_bh,
5893                                               cpos - num_clusters,
5894                                               p_cluster, num_clusters,
5895                                               dealloc, refcount);
5896                 if (ret) {
5897                         mlog_errno(ret);
5898                         break;
5899                 }
5900         }
5901
5902         return ret;
5903 }
5904
5905 /*
5906  * Given a normal ocfs2_xattr_header, refcount all the entries which
5907  * have value stored outside.
5908  * Used for xattrs stored in inode and ocfs2_xattr_block.
5909  */
5910 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5911                                 struct ocfs2_xattr_value_buf *vb,
5912                                 struct ocfs2_xattr_header *header,
5913                                 struct ocfs2_caching_info *ref_ci,
5914                                 struct buffer_head *ref_root_bh,
5915                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5916 {
5917
5918         struct ocfs2_xattr_entry *xe;
5919         struct ocfs2_xattr_value_root *xv;
5920         struct ocfs2_extent_tree et;
5921         int i, ret = 0;
5922
5923         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5924                 xe = &header->xh_entries[i];
5925
5926                 if (ocfs2_xattr_is_local(xe))
5927                         continue;
5928
5929                 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5930                         le16_to_cpu(xe->xe_name_offset) +
5931                         OCFS2_XATTR_SIZE(xe->xe_name_len));
5932
5933                 vb->vb_xv = xv;
5934                 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5935
5936                 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5937                                                         ref_ci, ref_root_bh,
5938                                                         dealloc, NULL);
5939                 if (ret) {
5940                         mlog_errno(ret);
5941                         break;
5942                 }
5943         }
5944
5945         return ret;
5946 }
5947
5948 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5949                                 struct buffer_head *fe_bh,
5950                                 struct ocfs2_caching_info *ref_ci,
5951                                 struct buffer_head *ref_root_bh,
5952                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5953 {
5954         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5955         struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5956                                 (fe_bh->b_data + inode->i_sb->s_blocksize -
5957                                 le16_to_cpu(di->i_xattr_inline_size));
5958         struct ocfs2_xattr_value_buf vb = {
5959                 .vb_bh = fe_bh,
5960                 .vb_access = ocfs2_journal_access_di,
5961         };
5962
5963         return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5964                                                   ref_ci, ref_root_bh, dealloc);
5965 }
5966
5967 struct ocfs2_xattr_tree_value_refcount_para {
5968         struct ocfs2_caching_info *ref_ci;
5969         struct buffer_head *ref_root_bh;
5970         struct ocfs2_cached_dealloc_ctxt *dealloc;
5971 };
5972
5973 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5974                                            struct ocfs2_xattr_bucket *bucket,
5975                                            int offset,
5976                                            struct ocfs2_xattr_value_root **xv,
5977                                            struct buffer_head **bh)
5978 {
5979         int ret, block_off, name_offset;
5980         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5981         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
5982         void *base;
5983
5984         ret = ocfs2_xattr_bucket_get_name_value(sb,
5985                                                 bucket_xh(bucket),
5986                                                 offset,
5987                                                 &block_off,
5988                                                 &name_offset);
5989         if (ret) {
5990                 mlog_errno(ret);
5991                 goto out;
5992         }
5993
5994         base = bucket_block(bucket, block_off);
5995
5996         *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
5997                          OCFS2_XATTR_SIZE(xe->xe_name_len));
5998
5999         if (bh)
6000                 *bh = bucket->bu_bhs[block_off];
6001 out:
6002         return ret;
6003 }
6004
6005 /*
6006  * For a given xattr bucket, refcount all the entries which
6007  * have value stored outside.
6008  */
6009 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6010                                              struct ocfs2_xattr_bucket *bucket,
6011                                              void *para)
6012 {
6013         int i, ret = 0;
6014         struct ocfs2_extent_tree et;
6015         struct ocfs2_xattr_tree_value_refcount_para *ref =
6016                         (struct ocfs2_xattr_tree_value_refcount_para *)para;
6017         struct ocfs2_xattr_header *xh =
6018                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6019         struct ocfs2_xattr_entry *xe;
6020         struct ocfs2_xattr_value_buf vb = {
6021                 .vb_access = ocfs2_journal_access,
6022         };
6023         struct ocfs2_post_refcount refcount = {
6024                 .credits = bucket->bu_blocks,
6025                 .para = bucket,
6026                 .func = ocfs2_xattr_bucket_post_refcount,
6027         };
6028         struct ocfs2_post_refcount *p = NULL;
6029
6030         /* We only need post_refcount if we support metaecc. */
6031         if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6032                 p = &refcount;
6033
6034         trace_ocfs2_xattr_bucket_value_refcount(
6035                                 (unsigned long long)bucket_blkno(bucket),
6036                                 le16_to_cpu(xh->xh_count));
6037         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6038                 xe = &xh->xh_entries[i];
6039
6040                 if (ocfs2_xattr_is_local(xe))
6041                         continue;
6042
6043                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6044                                                       &vb.vb_xv, &vb.vb_bh);
6045                 if (ret) {
6046                         mlog_errno(ret);
6047                         break;
6048                 }
6049
6050                 ocfs2_init_xattr_value_extent_tree(&et,
6051                                                    INODE_CACHE(inode), &vb);
6052
6053                 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6054                                                         &et, ref->ref_ci,
6055                                                         ref->ref_root_bh,
6056                                                         ref->dealloc, p);
6057                 if (ret) {
6058                         mlog_errno(ret);
6059                         break;
6060                 }
6061         }
6062
6063         return ret;
6064
6065 }
6066
6067 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6068                                      struct buffer_head *root_bh,
6069                                      u64 blkno, u32 cpos, u32 len, void *para)
6070 {
6071         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6072                                            ocfs2_xattr_bucket_value_refcount,
6073                                            para);
6074 }
6075
6076 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6077                                 struct buffer_head *blk_bh,
6078                                 struct ocfs2_caching_info *ref_ci,
6079                                 struct buffer_head *ref_root_bh,
6080                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
6081 {
6082         int ret = 0;
6083         struct ocfs2_xattr_block *xb =
6084                                 (struct ocfs2_xattr_block *)blk_bh->b_data;
6085
6086         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6087                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6088                 struct ocfs2_xattr_value_buf vb = {
6089                         .vb_bh = blk_bh,
6090                         .vb_access = ocfs2_journal_access_xb,
6091                 };
6092
6093                 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6094                                                          ref_ci, ref_root_bh,
6095                                                          dealloc);
6096         } else {
6097                 struct ocfs2_xattr_tree_value_refcount_para para = {
6098                         .ref_ci = ref_ci,
6099                         .ref_root_bh = ref_root_bh,
6100                         .dealloc = dealloc,
6101                 };
6102
6103                 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6104                                                 ocfs2_refcount_xattr_tree_rec,
6105                                                 &para);
6106         }
6107
6108         return ret;
6109 }
6110
6111 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6112                                      struct buffer_head *fe_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_inode_info *oi = OCFS2_I(inode);
6119         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6120         struct buffer_head *blk_bh = NULL;
6121
6122         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6123                 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6124                                                          ref_ci, ref_root_bh,
6125                                                          dealloc);
6126                 if (ret) {
6127                         mlog_errno(ret);
6128                         goto out;
6129                 }
6130         }
6131
6132         if (!di->i_xattr_loc)
6133                 goto out;
6134
6135         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6136                                      &blk_bh);
6137         if (ret < 0) {
6138                 mlog_errno(ret);
6139                 goto out;
6140         }
6141
6142         ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6143                                                 ref_root_bh, dealloc);
6144         if (ret)
6145                 mlog_errno(ret);
6146
6147         brelse(blk_bh);
6148 out:
6149
6150         return ret;
6151 }
6152
6153 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6154 /*
6155  * Store the information we need in xattr reflink.
6156  * old_bh and new_bh are inode bh for the old and new inode.
6157  */
6158 struct ocfs2_xattr_reflink {
6159         struct inode *old_inode;
6160         struct inode *new_inode;
6161         struct buffer_head *old_bh;
6162         struct buffer_head *new_bh;
6163         struct ocfs2_caching_info *ref_ci;
6164         struct buffer_head *ref_root_bh;
6165         struct ocfs2_cached_dealloc_ctxt *dealloc;
6166         should_xattr_reflinked *xattr_reflinked;
6167 };
6168
6169 /*
6170  * Given a xattr header and xe offset,
6171  * return the proper xv and the corresponding bh.
6172  * xattr in inode, block and xattr tree have different implementaions.
6173  */
6174 typedef int (get_xattr_value_root)(struct super_block *sb,
6175                                    struct buffer_head *bh,
6176                                    struct ocfs2_xattr_header *xh,
6177                                    int offset,
6178                                    struct ocfs2_xattr_value_root **xv,
6179                                    struct buffer_head **ret_bh,
6180                                    void *para);
6181
6182 /*
6183  * Calculate all the xattr value root metadata stored in this xattr header and
6184  * credits we need if we create them from the scratch.
6185  * We use get_xattr_value_root so that all types of xattr container can use it.
6186  */
6187 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6188                                              struct buffer_head *bh,
6189                                              struct ocfs2_xattr_header *xh,
6190                                              int *metas, int *credits,
6191                                              int *num_recs,
6192                                              get_xattr_value_root *func,
6193                                              void *para)
6194 {
6195         int i, ret = 0;
6196         struct ocfs2_xattr_value_root *xv;
6197         struct ocfs2_xattr_entry *xe;
6198
6199         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6200                 xe = &xh->xh_entries[i];
6201                 if (ocfs2_xattr_is_local(xe))
6202                         continue;
6203
6204                 ret = func(sb, bh, xh, i, &xv, NULL, para);
6205                 if (ret) {
6206                         mlog_errno(ret);
6207                         break;
6208                 }
6209
6210                 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6211                           le16_to_cpu(xv->xr_list.l_next_free_rec);
6212
6213                 *credits += ocfs2_calc_extend_credits(sb,
6214                                                 &def_xv.xv.xr_list,
6215                                                 le32_to_cpu(xv->xr_clusters));
6216
6217                 /*
6218                  * If the value is a tree with depth > 1, We don't go deep
6219                  * to the extent block, so just calculate a maximum record num.
6220                  */
6221                 if (!xv->xr_list.l_tree_depth)
6222                         *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6223                 else
6224                         *num_recs += ocfs2_clusters_for_bytes(sb,
6225                                                               XATTR_SIZE_MAX);
6226         }
6227
6228         return ret;
6229 }
6230
6231 /* Used by xattr inode and block to return the right xv and buffer_head. */
6232 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6233                                       struct buffer_head *bh,
6234                                       struct ocfs2_xattr_header *xh,
6235                                       int offset,
6236                                       struct ocfs2_xattr_value_root **xv,
6237                                       struct buffer_head **ret_bh,
6238                                       void *para)
6239 {
6240         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6241
6242         *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6243                 le16_to_cpu(xe->xe_name_offset) +
6244                 OCFS2_XATTR_SIZE(xe->xe_name_len));
6245
6246         if (ret_bh)
6247                 *ret_bh = bh;
6248
6249         return 0;
6250 }
6251
6252 /*
6253  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6254  * It is only used for inline xattr and xattr block.
6255  */
6256 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6257                                         struct ocfs2_xattr_header *xh,
6258                                         struct buffer_head *ref_root_bh,
6259                                         int *credits,
6260                                         struct ocfs2_alloc_context **meta_ac)
6261 {
6262         int ret, meta_add = 0, num_recs = 0;
6263         struct ocfs2_refcount_block *rb =
6264                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6265
6266         *credits = 0;
6267
6268         ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6269                                                 &meta_add, credits, &num_recs,
6270                                                 ocfs2_get_xattr_value_root,
6271                                                 NULL);
6272         if (ret) {
6273                 mlog_errno(ret);
6274                 goto out;
6275         }
6276
6277         /*
6278          * We need to add/modify num_recs in refcount tree, so just calculate
6279          * an approximate number we need for refcount tree change.
6280          * Sometimes we need to split the tree, and after split,  half recs
6281          * will be moved to the new block, and a new block can only provide
6282          * half number of recs. So we multiple new blocks by 2.
6283          */
6284         num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6285         meta_add += num_recs;
6286         *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6287         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6288                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6289                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6290         else
6291                 *credits += 1;
6292
6293         ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6294         if (ret)
6295                 mlog_errno(ret);
6296
6297 out:
6298         return ret;
6299 }
6300
6301 /*
6302  * Given a xattr header, reflink all the xattrs in this container.
6303  * It can be used for inode, block and bucket.
6304  *
6305  * NOTE:
6306  * Before we call this function, the caller has memcpy the xattr in
6307  * old_xh to the new_xh.
6308  *
6309  * If args.xattr_reflinked is set, call it to decide whether the xe should
6310  * be reflinked or not. If not, remove it from the new xattr header.
6311  */
6312 static int ocfs2_reflink_xattr_header(handle_t *handle,
6313                                       struct ocfs2_xattr_reflink *args,
6314                                       struct buffer_head *old_bh,
6315                                       struct ocfs2_xattr_header *xh,
6316                                       struct buffer_head *new_bh,
6317                                       struct ocfs2_xattr_header *new_xh,
6318                                       struct ocfs2_xattr_value_buf *vb,
6319                                       struct ocfs2_alloc_context *meta_ac,
6320                                       get_xattr_value_root *func,
6321                                       void *para)
6322 {
6323         int ret = 0, i, j;
6324         struct super_block *sb = args->old_inode->i_sb;
6325         struct buffer_head *value_bh;
6326         struct ocfs2_xattr_entry *xe, *last;
6327         struct ocfs2_xattr_value_root *xv, *new_xv;
6328         struct ocfs2_extent_tree data_et;
6329         u32 clusters, cpos, p_cluster, num_clusters;
6330         unsigned int ext_flags = 0;
6331
6332         trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6333                                          le16_to_cpu(xh->xh_count));
6334
6335         last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6336         for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6337                 xe = &xh->xh_entries[i];
6338
6339                 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6340                         xe = &new_xh->xh_entries[j];
6341
6342                         le16_add_cpu(&new_xh->xh_count, -1);
6343                         if (new_xh->xh_count) {
6344                                 memmove(xe, xe + 1,
6345                                         (void *)last - (void *)xe);
6346                                 memset(last, 0,
6347                                        sizeof(struct ocfs2_xattr_entry));
6348                         }
6349
6350                         /*
6351                          * We don't want j to increase in the next round since
6352                          * it is already moved ahead.
6353                          */
6354                         j--;
6355                         continue;
6356                 }
6357
6358                 if (ocfs2_xattr_is_local(xe))
6359                         continue;
6360
6361                 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6362                 if (ret) {
6363                         mlog_errno(ret);
6364                         break;
6365                 }
6366
6367                 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6368                 if (ret) {
6369                         mlog_errno(ret);
6370                         break;
6371                 }
6372
6373                 /*
6374                  * For the xattr which has l_tree_depth = 0, all the extent
6375                  * recs have already be copied to the new xh with the
6376                  * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6377                  * increase the refount count int the refcount tree.
6378                  *
6379                  * For the xattr which has l_tree_depth > 0, we need
6380                  * to initialize it to the empty default value root,
6381                  * and then insert the extents one by one.
6382                  */
6383                 if (xv->xr_list.l_tree_depth) {
6384                         memcpy(new_xv, &def_xv, sizeof(def_xv));
6385                         vb->vb_xv = new_xv;
6386                         vb->vb_bh = value_bh;
6387                         ocfs2_init_xattr_value_extent_tree(&data_et,
6388                                         INODE_CACHE(args->new_inode), vb);
6389                 }
6390
6391                 clusters = le32_to_cpu(xv->xr_clusters);
6392                 cpos = 0;
6393                 while (cpos < clusters) {
6394                         ret = ocfs2_xattr_get_clusters(args->old_inode,
6395                                                        cpos,
6396                                                        &p_cluster,
6397                                                        &num_clusters,
6398                                                        &xv->xr_list,
6399                                                        &ext_flags);
6400                         if (ret) {
6401                                 mlog_errno(ret);
6402                                 goto out;
6403                         }
6404
6405                         BUG_ON(!p_cluster);
6406
6407                         if (xv->xr_list.l_tree_depth) {
6408                                 ret = ocfs2_insert_extent(handle,
6409                                                 &data_et, cpos,
6410                                                 ocfs2_clusters_to_blocks(
6411                                                         args->old_inode->i_sb,
6412                                                         p_cluster),
6413                                                 num_clusters, ext_flags,
6414                                                 meta_ac);
6415                                 if (ret) {
6416                                         mlog_errno(ret);
6417                                         goto out;
6418                                 }
6419                         }
6420
6421                         ret = ocfs2_increase_refcount(handle, args->ref_ci,
6422                                                       args->ref_root_bh,
6423                                                       p_cluster, num_clusters,
6424                                                       meta_ac, args->dealloc);
6425                         if (ret) {
6426                                 mlog_errno(ret);
6427                                 goto out;
6428                         }
6429
6430                         cpos += num_clusters;
6431                 }
6432         }
6433
6434 out:
6435         return ret;
6436 }
6437
6438 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6439 {
6440         int ret = 0, credits = 0;
6441         handle_t *handle;
6442         struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6443         struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6444         int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6445         int header_off = osb->sb->s_blocksize - inline_size;
6446         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6447                                         (args->old_bh->b_data + header_off);
6448         struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6449                                         (args->new_bh->b_data + header_off);
6450         struct ocfs2_alloc_context *meta_ac = NULL;
6451         struct ocfs2_inode_info *new_oi;
6452         struct ocfs2_dinode *new_di;
6453         struct ocfs2_xattr_value_buf vb = {
6454                 .vb_bh = args->new_bh,
6455                 .vb_access = ocfs2_journal_access_di,
6456         };
6457
6458         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6459                                                   &credits, &meta_ac);
6460         if (ret) {
6461                 mlog_errno(ret);
6462                 goto out;
6463         }
6464
6465         handle = ocfs2_start_trans(osb, credits);
6466         if (IS_ERR(handle)) {
6467                 ret = PTR_ERR(handle);
6468                 mlog_errno(ret);
6469                 goto out;
6470         }
6471
6472         ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6473                                       args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6474         if (ret) {
6475                 mlog_errno(ret);
6476                 goto out_commit;
6477         }
6478
6479         memcpy(args->new_bh->b_data + header_off,
6480                args->old_bh->b_data + header_off, inline_size);
6481
6482         new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6483         new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6484
6485         ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6486                                          args->new_bh, new_xh, &vb, meta_ac,
6487                                          ocfs2_get_xattr_value_root, NULL);
6488         if (ret) {
6489                 mlog_errno(ret);
6490                 goto out_commit;
6491         }
6492
6493         new_oi = OCFS2_I(args->new_inode);
6494         /*
6495          * Adjust extent record count to reserve space for extended attribute.
6496          * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6497          */
6498         if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6499             !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6500                 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6501                 le16_add_cpu(&el->l_count, -(inline_size /
6502                                         sizeof(struct ocfs2_extent_rec)));
6503         }
6504         spin_lock(&new_oi->ip_lock);
6505         new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6506         new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6507         spin_unlock(&new_oi->ip_lock);
6508
6509         ocfs2_journal_dirty(handle, args->new_bh);
6510
6511 out_commit:
6512         ocfs2_commit_trans(osb, handle);
6513
6514 out:
6515         if (meta_ac)
6516                 ocfs2_free_alloc_context(meta_ac);
6517         return ret;
6518 }
6519
6520 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6521                                           struct buffer_head *fe_bh,
6522                                           struct buffer_head **ret_bh,
6523                                           int indexed)
6524 {
6525         int ret;
6526         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6527         struct ocfs2_xattr_set_ctxt ctxt;
6528
6529         memset(&ctxt, 0, sizeof(ctxt));
6530         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6531         if (ret < 0) {
6532                 mlog_errno(ret);
6533                 return ret;
6534         }
6535
6536         ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6537         if (IS_ERR(ctxt.handle)) {
6538                 ret = PTR_ERR(ctxt.handle);
6539                 mlog_errno(ret);
6540                 goto out;
6541         }
6542
6543         trace_ocfs2_create_empty_xattr_block(
6544                                 (unsigned long long)fe_bh->b_blocknr, indexed);
6545         ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6546                                        ret_bh);
6547         if (ret)
6548                 mlog_errno(ret);
6549
6550         ocfs2_commit_trans(osb, ctxt.handle);
6551 out:
6552         ocfs2_free_alloc_context(ctxt.meta_ac);
6553         return ret;
6554 }
6555
6556 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6557                                      struct buffer_head *blk_bh,
6558                                      struct buffer_head *new_blk_bh)
6559 {
6560         int ret = 0, credits = 0;
6561         handle_t *handle;
6562         struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6563         struct ocfs2_dinode *new_di;
6564         struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6565         int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6566         struct ocfs2_xattr_block *xb =
6567                         (struct ocfs2_xattr_block *)blk_bh->b_data;
6568         struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6569         struct ocfs2_xattr_block *new_xb =
6570                         (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6571         struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6572         struct ocfs2_alloc_context *meta_ac;
6573         struct ocfs2_xattr_value_buf vb = {
6574                 .vb_bh = new_blk_bh,
6575                 .vb_access = ocfs2_journal_access_xb,
6576         };
6577
6578         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6579                                                   &credits, &meta_ac);
6580         if (ret) {
6581                 mlog_errno(ret);
6582                 return ret;
6583         }
6584
6585         /* One more credits in case we need to add xattr flags in new inode. */
6586         handle = ocfs2_start_trans(osb, credits + 1);
6587         if (IS_ERR(handle)) {
6588                 ret = PTR_ERR(handle);
6589                 mlog_errno(ret);
6590                 goto out;
6591         }
6592
6593         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6594                 ret = ocfs2_journal_access_di(handle,
6595                                               INODE_CACHE(args->new_inode),
6596                                               args->new_bh,
6597                                               OCFS2_JOURNAL_ACCESS_WRITE);
6598                 if (ret) {
6599                         mlog_errno(ret);
6600                         goto out_commit;
6601                 }
6602         }
6603
6604         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6605                                       new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6606         if (ret) {
6607                 mlog_errno(ret);
6608                 goto out_commit;
6609         }
6610
6611         memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6612                osb->sb->s_blocksize - header_off);
6613
6614         ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6615                                          new_blk_bh, new_xh, &vb, meta_ac,
6616                                          ocfs2_get_xattr_value_root, NULL);
6617         if (ret) {
6618                 mlog_errno(ret);
6619                 goto out_commit;
6620         }
6621
6622         ocfs2_journal_dirty(handle, new_blk_bh);
6623
6624         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6625                 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6626                 spin_lock(&new_oi->ip_lock);
6627                 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6628                 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6629                 spin_unlock(&new_oi->ip_lock);
6630
6631                 ocfs2_journal_dirty(handle, args->new_bh);
6632         }
6633
6634 out_commit:
6635         ocfs2_commit_trans(osb, handle);
6636
6637 out:
6638         ocfs2_free_alloc_context(meta_ac);
6639         return ret;
6640 }
6641
6642 struct ocfs2_reflink_xattr_tree_args {
6643         struct ocfs2_xattr_reflink *reflink;
6644         struct buffer_head *old_blk_bh;
6645         struct buffer_head *new_blk_bh;
6646         struct ocfs2_xattr_bucket *old_bucket;
6647         struct ocfs2_xattr_bucket *new_bucket;
6648 };
6649
6650 /*
6651  * NOTE:
6652  * We have to handle the case that both old bucket and new bucket
6653  * will call this function to get the right ret_bh.
6654  * So The caller must give us the right bh.
6655  */
6656 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6657                                         struct buffer_head *bh,
6658                                         struct ocfs2_xattr_header *xh,
6659                                         int offset,
6660                                         struct ocfs2_xattr_value_root **xv,
6661                                         struct buffer_head **ret_bh,
6662                                         void *para)
6663 {
6664         struct ocfs2_reflink_xattr_tree_args *args =
6665                         (struct ocfs2_reflink_xattr_tree_args *)para;
6666         struct ocfs2_xattr_bucket *bucket;
6667
6668         if (bh == args->old_bucket->bu_bhs[0])
6669                 bucket = args->old_bucket;
6670         else
6671                 bucket = args->new_bucket;
6672
6673         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6674                                                xv, ret_bh);
6675 }
6676
6677 struct ocfs2_value_tree_metas {
6678         int num_metas;
6679         int credits;
6680         int num_recs;
6681 };
6682
6683 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6684                                         struct buffer_head *bh,
6685                                         struct ocfs2_xattr_header *xh,
6686                                         int offset,
6687                                         struct ocfs2_xattr_value_root **xv,
6688                                         struct buffer_head **ret_bh,
6689                                         void *para)
6690 {
6691         struct ocfs2_xattr_bucket *bucket =
6692                                 (struct ocfs2_xattr_bucket *)para;
6693
6694         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6695                                                xv, ret_bh);
6696 }
6697
6698 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6699                                       struct ocfs2_xattr_bucket *bucket,
6700                                       void *para)
6701 {
6702         struct ocfs2_value_tree_metas *metas =
6703                         (struct ocfs2_value_tree_metas *)para;
6704         struct ocfs2_xattr_header *xh =
6705                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6706
6707         /* Add the credits for this bucket first. */
6708         metas->credits += bucket->bu_blocks;
6709         return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6710                                         xh, &metas->num_metas,
6711                                         &metas->credits, &metas->num_recs,
6712                                         ocfs2_value_tree_metas_in_bucket,
6713                                         bucket);
6714 }
6715
6716 /*
6717  * Given a xattr extent rec starting from blkno and having len clusters,
6718  * iterate all the buckets calculate how much metadata we need for reflinking
6719  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6720  */
6721 static int ocfs2_lock_reflink_xattr_rec_allocators(
6722                                 struct ocfs2_reflink_xattr_tree_args *args,
6723                                 struct ocfs2_extent_tree *xt_et,
6724                                 u64 blkno, u32 len, int *credits,
6725                                 struct ocfs2_alloc_context **meta_ac,
6726                                 struct ocfs2_alloc_context **data_ac)
6727 {
6728         int ret, num_free_extents;
6729         struct ocfs2_value_tree_metas metas;
6730         struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6731         struct ocfs2_refcount_block *rb;
6732
6733         memset(&metas, 0, sizeof(metas));
6734
6735         ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6736                                           ocfs2_calc_value_tree_metas, &metas);
6737         if (ret) {
6738                 mlog_errno(ret);
6739                 goto out;
6740         }
6741
6742         *credits = metas.credits;
6743
6744         /*
6745          * Calculate we need for refcount tree change.
6746          *
6747          * We need to add/modify num_recs in refcount tree, so just calculate
6748          * an approximate number we need for refcount tree change.
6749          * Sometimes we need to split the tree, and after split,  half recs
6750          * will be moved to the new block, and a new block can only provide
6751          * half number of recs. So we multiple new blocks by 2.
6752          * In the end, we have to add credits for modifying the already
6753          * existed refcount block.
6754          */
6755         rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6756         metas.num_recs =
6757                 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6758                  ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6759         metas.num_metas += metas.num_recs;
6760         *credits += metas.num_recs +
6761                     metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6762         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6763                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6764                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6765         else
6766                 *credits += 1;
6767
6768         /* count in the xattr tree change. */
6769         num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6770         if (num_free_extents < 0) {
6771                 ret = num_free_extents;
6772                 mlog_errno(ret);
6773                 goto out;
6774         }
6775
6776         if (num_free_extents < len)
6777                 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6778
6779         *credits += ocfs2_calc_extend_credits(osb->sb,
6780                                               xt_et->et_root_el, len);
6781
6782         if (metas.num_metas) {
6783                 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6784                                                         meta_ac);
6785                 if (ret) {
6786                         mlog_errno(ret);
6787                         goto out;
6788                 }
6789         }
6790
6791         if (len) {
6792                 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6793                 if (ret)
6794                         mlog_errno(ret);
6795         }
6796 out:
6797         if (ret) {
6798                 if (*meta_ac) {
6799                         ocfs2_free_alloc_context(*meta_ac);
6800                         meta_ac = NULL;
6801                 }
6802         }
6803
6804         return ret;
6805 }
6806
6807 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6808                                 u64 blkno, u64 new_blkno, u32 clusters,
6809                                 u32 *cpos, int num_buckets,
6810                                 struct ocfs2_alloc_context *meta_ac,
6811                                 struct ocfs2_alloc_context *data_ac,
6812                                 struct ocfs2_reflink_xattr_tree_args *args)
6813 {
6814         int i, j, ret = 0;
6815         struct super_block *sb = args->reflink->old_inode->i_sb;
6816         int bpb = args->old_bucket->bu_blocks;
6817         struct ocfs2_xattr_value_buf vb = {
6818                 .vb_access = ocfs2_journal_access,
6819         };
6820
6821         for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6822                 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6823                 if (ret) {
6824                         mlog_errno(ret);
6825                         break;
6826                 }
6827
6828                 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno);
6829                 if (ret) {
6830                         mlog_errno(ret);
6831                         break;
6832                 }
6833
6834                 ret = ocfs2_xattr_bucket_journal_access(handle,
6835                                                 args->new_bucket,
6836                                                 OCFS2_JOURNAL_ACCESS_CREATE);
6837                 if (ret) {
6838                         mlog_errno(ret);
6839                         break;
6840                 }
6841
6842                 for (j = 0; j < bpb; j++)
6843                         memcpy(bucket_block(args->new_bucket, j),
6844                                bucket_block(args->old_bucket, j),
6845                                sb->s_blocksize);
6846
6847                 /*
6848                  * Record the start cpos so that we can use it to initialize
6849                  * our xattr tree we also set the xh_num_bucket for the new
6850                  * bucket.
6851                  */
6852                 if (i == 0) {
6853                         *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6854                                             xh_entries[0].xe_name_hash);
6855                         bucket_xh(args->new_bucket)->xh_num_buckets =
6856                                 cpu_to_le16(num_buckets);
6857                 }
6858
6859                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6860
6861                 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6862                                         args->old_bucket->bu_bhs[0],
6863                                         bucket_xh(args->old_bucket),
6864                                         args->new_bucket->bu_bhs[0],
6865                                         bucket_xh(args->new_bucket),
6866                                         &vb, meta_ac,
6867                                         ocfs2_get_reflink_xattr_value_root,
6868                                         args);
6869                 if (ret) {
6870                         mlog_errno(ret);
6871                         break;
6872                 }
6873
6874                 /*
6875                  * Re-access and dirty the bucket to calculate metaecc.
6876                  * Because we may extend the transaction in reflink_xattr_header
6877                  * which will let the already accessed block gone.
6878                  */
6879                 ret = ocfs2_xattr_bucket_journal_access(handle,
6880                                                 args->new_bucket,
6881                                                 OCFS2_JOURNAL_ACCESS_WRITE);
6882                 if (ret) {
6883                         mlog_errno(ret);
6884                         break;
6885                 }
6886
6887                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6888
6889                 ocfs2_xattr_bucket_relse(args->old_bucket);
6890                 ocfs2_xattr_bucket_relse(args->new_bucket);
6891         }
6892
6893         ocfs2_xattr_bucket_relse(args->old_bucket);
6894         ocfs2_xattr_bucket_relse(args->new_bucket);
6895         return ret;
6896 }
6897
6898 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6899                                 struct inode *inode,
6900                                 struct ocfs2_reflink_xattr_tree_args *args,
6901                                 struct ocfs2_extent_tree *et,
6902                                 struct ocfs2_alloc_context *meta_ac,
6903                                 struct ocfs2_alloc_context *data_ac,
6904                                 u64 blkno, u32 cpos, u32 len)
6905 {
6906         int ret, first_inserted = 0;
6907         u32 p_cluster, num_clusters, reflink_cpos = 0;
6908         u64 new_blkno;
6909         unsigned int num_buckets, reflink_buckets;
6910         unsigned int bpc =
6911                 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6912
6913         ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6914         if (ret) {
6915                 mlog_errno(ret);
6916                 goto out;
6917         }
6918         num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6919         ocfs2_xattr_bucket_relse(args->old_bucket);
6920
6921         while (len && num_buckets) {
6922                 ret = ocfs2_claim_clusters(handle, data_ac,
6923                                            1, &p_cluster, &num_clusters);
6924                 if (ret) {
6925                         mlog_errno(ret);
6926                         goto out;
6927                 }
6928
6929                 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6930                 reflink_buckets = min(num_buckets, bpc * num_clusters);
6931
6932                 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6933                                                  new_blkno, num_clusters,
6934                                                  &reflink_cpos, reflink_buckets,
6935                                                  meta_ac, data_ac, args);
6936                 if (ret) {
6937                         mlog_errno(ret);
6938                         goto out;
6939                 }
6940
6941                 /*
6942                  * For the 1st allocated cluster, we make it use the same cpos
6943                  * so that the xattr tree looks the same as the original one
6944                  * in the most case.
6945                  */
6946                 if (!first_inserted) {
6947                         reflink_cpos = cpos;
6948                         first_inserted = 1;
6949                 }
6950                 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6951                                           num_clusters, 0, meta_ac);
6952                 if (ret)
6953                         mlog_errno(ret);
6954
6955                 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6956                                                   num_clusters, reflink_cpos);
6957
6958                 len -= num_clusters;
6959                 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6960                 num_buckets -= reflink_buckets;
6961         }
6962 out:
6963         return ret;
6964 }
6965
6966 /*
6967  * Create the same xattr extent record in the new inode's xattr tree.
6968  */
6969 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6970                                    struct buffer_head *root_bh,
6971                                    u64 blkno,
6972                                    u32 cpos,
6973                                    u32 len,
6974                                    void *para)
6975 {
6976         int ret, credits = 0;
6977         handle_t *handle;
6978         struct ocfs2_reflink_xattr_tree_args *args =
6979                         (struct ocfs2_reflink_xattr_tree_args *)para;
6980         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6981         struct ocfs2_alloc_context *meta_ac = NULL;
6982         struct ocfs2_alloc_context *data_ac = NULL;
6983         struct ocfs2_extent_tree et;
6984
6985         trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
6986
6987         ocfs2_init_xattr_tree_extent_tree(&et,
6988                                           INODE_CACHE(args->reflink->new_inode),
6989                                           args->new_blk_bh);
6990
6991         ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
6992                                                       len, &credits,
6993                                                       &meta_ac, &data_ac);
6994         if (ret) {
6995                 mlog_errno(ret);
6996                 goto out;
6997         }
6998
6999         handle = ocfs2_start_trans(osb, credits);
7000         if (IS_ERR(handle)) {
7001                 ret = PTR_ERR(handle);
7002                 mlog_errno(ret);
7003                 goto out;
7004         }
7005
7006         ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7007                                           meta_ac, data_ac,
7008                                           blkno, cpos, len);
7009         if (ret)
7010                 mlog_errno(ret);
7011
7012         ocfs2_commit_trans(osb, handle);
7013
7014 out:
7015         if (meta_ac)
7016                 ocfs2_free_alloc_context(meta_ac);
7017         if (data_ac)
7018                 ocfs2_free_alloc_context(data_ac);
7019         return ret;
7020 }
7021
7022 /*
7023  * Create reflinked xattr buckets.
7024  * We will add bucket one by one, and refcount all the xattrs in the bucket
7025  * if they are stored outside.
7026  */
7027 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7028                                     struct buffer_head *blk_bh,
7029                                     struct buffer_head *new_blk_bh)
7030 {
7031         int ret;
7032         struct ocfs2_reflink_xattr_tree_args para;
7033
7034         memset(&para, 0, sizeof(para));
7035         para.reflink = args;
7036         para.old_blk_bh = blk_bh;
7037         para.new_blk_bh = new_blk_bh;
7038
7039         para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7040         if (!para.old_bucket) {
7041                 mlog_errno(-ENOMEM);
7042                 return -ENOMEM;
7043         }
7044
7045         para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7046         if (!para.new_bucket) {
7047                 ret = -ENOMEM;
7048                 mlog_errno(ret);
7049                 goto out;
7050         }
7051
7052         ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7053                                               ocfs2_reflink_xattr_rec,
7054                                               &para);
7055         if (ret)
7056                 mlog_errno(ret);
7057
7058 out:
7059         ocfs2_xattr_bucket_free(para.old_bucket);
7060         ocfs2_xattr_bucket_free(para.new_bucket);
7061         return ret;
7062 }
7063
7064 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7065                                         struct buffer_head *blk_bh)
7066 {
7067         int ret, indexed = 0;
7068         struct buffer_head *new_blk_bh = NULL;
7069         struct ocfs2_xattr_block *xb =
7070                         (struct ocfs2_xattr_block *)blk_bh->b_data;
7071
7072
7073         if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7074                 indexed = 1;
7075
7076         ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7077                                              &new_blk_bh, indexed);
7078         if (ret) {
7079                 mlog_errno(ret);
7080                 goto out;
7081         }
7082
7083         if (!indexed)
7084                 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7085         else
7086                 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7087         if (ret)
7088                 mlog_errno(ret);
7089
7090 out:
7091         brelse(new_blk_bh);
7092         return ret;
7093 }
7094
7095 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7096 {
7097         int type = ocfs2_xattr_get_type(xe);
7098
7099         return type != OCFS2_XATTR_INDEX_SECURITY &&
7100                type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7101                type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7102 }
7103
7104 int ocfs2_reflink_xattrs(struct inode *old_inode,
7105                          struct buffer_head *old_bh,
7106                          struct inode *new_inode,
7107                          struct buffer_head *new_bh,
7108                          bool preserve_security)
7109 {
7110         int ret;
7111         struct ocfs2_xattr_reflink args;
7112         struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7113         struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7114         struct buffer_head *blk_bh = NULL;
7115         struct ocfs2_cached_dealloc_ctxt dealloc;
7116         struct ocfs2_refcount_tree *ref_tree;
7117         struct buffer_head *ref_root_bh = NULL;
7118
7119         ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7120                                        le64_to_cpu(di->i_refcount_loc),
7121                                        1, &ref_tree, &ref_root_bh);
7122         if (ret) {
7123                 mlog_errno(ret);
7124                 goto out;
7125         }
7126
7127         ocfs2_init_dealloc_ctxt(&dealloc);
7128
7129         args.old_inode = old_inode;
7130         args.new_inode = new_inode;
7131         args.old_bh = old_bh;
7132         args.new_bh = new_bh;
7133         args.ref_ci = &ref_tree->rf_ci;
7134         args.ref_root_bh = ref_root_bh;
7135         args.dealloc = &dealloc;
7136         if (preserve_security)
7137                 args.xattr_reflinked = NULL;
7138         else
7139                 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7140
7141         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7142                 ret = ocfs2_reflink_xattr_inline(&args);
7143                 if (ret) {
7144                         mlog_errno(ret);
7145                         goto out_unlock;
7146                 }
7147         }
7148
7149         if (!di->i_xattr_loc)
7150                 goto out_unlock;
7151
7152         ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7153                                      &blk_bh);
7154         if (ret < 0) {
7155                 mlog_errno(ret);
7156                 goto out_unlock;
7157         }
7158
7159         ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7160         if (ret)
7161                 mlog_errno(ret);
7162
7163         brelse(blk_bh);
7164
7165 out_unlock:
7166         ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7167                                    ref_tree, 1);
7168         brelse(ref_root_bh);
7169
7170         if (ocfs2_dealloc_has_cluster(&dealloc)) {
7171                 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7172                 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7173         }
7174
7175 out:
7176         return ret;
7177 }
7178
7179 /*
7180  * Initialize security and acl for a already created inode.
7181  * Used for reflink a non-preserve-security file.
7182  *
7183  * It uses common api like ocfs2_xattr_set, so the caller
7184  * must not hold any lock expect i_mutex.
7185  */
7186 int ocfs2_init_security_and_acl(struct inode *dir,
7187                                 struct inode *inode,
7188                                 const struct qstr *qstr)
7189 {
7190         int ret = 0;
7191         struct buffer_head *dir_bh = NULL;
7192
7193         ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7194         if (ret) {
7195                 mlog_errno(ret);
7196                 goto leave;
7197         }
7198
7199         ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7200         if (ret) {
7201                 mlog_errno(ret);
7202                 goto leave;
7203         }
7204
7205         ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7206         if (ret)
7207                 mlog_errno(ret);
7208
7209         ocfs2_inode_unlock(dir, 0);
7210         brelse(dir_bh);
7211 leave:
7212         return ret;
7213 }
7214 /*
7215  * 'security' attributes support
7216  */
7217 static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
7218                                         size_t list_size, const char *name,
7219                                         size_t name_len, int type)
7220 {
7221         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7222         const size_t total_len = prefix_len + name_len + 1;
7223
7224         if (list && total_len <= list_size) {
7225                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7226                 memcpy(list + prefix_len, name, name_len);
7227                 list[prefix_len + name_len] = '\0';
7228         }
7229         return total_len;
7230 }
7231
7232 static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7233                                     void *buffer, size_t size, int type)
7234 {
7235         if (strcmp(name, "") == 0)
7236                 return -EINVAL;
7237         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7238                                name, buffer, size);
7239 }
7240
7241 static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7242                 const void *value, size_t size, int flags, int type)
7243 {
7244         if (strcmp(name, "") == 0)
7245                 return -EINVAL;
7246
7247         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7248                                name, value, size, flags);
7249 }
7250
7251 int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7252                      void *fs_info)
7253 {
7254         const struct xattr *xattr;
7255         int err = 0;
7256
7257         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7258                 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7259                                       xattr->name, xattr->value,
7260                                       xattr->value_len, XATTR_CREATE);
7261                 if (err)
7262                         break;
7263         }
7264         return err;
7265 }
7266
7267 int ocfs2_init_security_get(struct inode *inode,
7268                             struct inode *dir,
7269                             const struct qstr *qstr,
7270                             struct ocfs2_security_xattr_info *si)
7271 {
7272         /* check whether ocfs2 support feature xattr */
7273         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7274                 return -EOPNOTSUPP;
7275         if (si)
7276                 return security_old_inode_init_security(inode, dir, qstr,
7277                                                         &si->name, &si->value,
7278                                                         &si->value_len);
7279
7280         return security_inode_init_security(inode, dir, qstr,
7281                                             &ocfs2_initxattrs, NULL);
7282 }
7283
7284 int ocfs2_init_security_set(handle_t *handle,
7285                             struct inode *inode,
7286                             struct buffer_head *di_bh,
7287                             struct ocfs2_security_xattr_info *si,
7288                             struct ocfs2_alloc_context *xattr_ac,
7289                             struct ocfs2_alloc_context *data_ac)
7290 {
7291         return ocfs2_xattr_set_handle(handle, inode, di_bh,
7292                                      OCFS2_XATTR_INDEX_SECURITY,
7293                                      si->name, si->value, si->value_len, 0,
7294                                      xattr_ac, data_ac);
7295 }
7296
7297 const struct xattr_handler ocfs2_xattr_security_handler = {
7298         .prefix = XATTR_SECURITY_PREFIX,
7299         .list   = ocfs2_xattr_security_list,
7300         .get    = ocfs2_xattr_security_get,
7301         .set    = ocfs2_xattr_security_set,
7302 };
7303
7304 /*
7305  * 'trusted' attributes support
7306  */
7307 static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
7308                                        size_t list_size, const char *name,
7309                                        size_t name_len, int type)
7310 {
7311         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7312         const size_t total_len = prefix_len + name_len + 1;
7313
7314         if (list && total_len <= list_size) {
7315                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7316                 memcpy(list + prefix_len, name, name_len);
7317                 list[prefix_len + name_len] = '\0';
7318         }
7319         return total_len;
7320 }
7321
7322 static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7323                 void *buffer, size_t size, int type)
7324 {
7325         if (strcmp(name, "") == 0)
7326                 return -EINVAL;
7327         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7328                                name, buffer, size);
7329 }
7330
7331 static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7332                 const void *value, size_t size, int flags, int type)
7333 {
7334         if (strcmp(name, "") == 0)
7335                 return -EINVAL;
7336
7337         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7338                                name, value, size, flags);
7339 }
7340
7341 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7342         .prefix = XATTR_TRUSTED_PREFIX,
7343         .list   = ocfs2_xattr_trusted_list,
7344         .get    = ocfs2_xattr_trusted_get,
7345         .set    = ocfs2_xattr_trusted_set,
7346 };
7347
7348 /*
7349  * 'user' attributes support
7350  */
7351 static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
7352                                     size_t list_size, const char *name,
7353                                     size_t name_len, int type)
7354 {
7355         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7356         const size_t total_len = prefix_len + name_len + 1;
7357         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7358
7359         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7360                 return 0;
7361
7362         if (list && total_len <= list_size) {
7363                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7364                 memcpy(list + prefix_len, name, name_len);
7365                 list[prefix_len + name_len] = '\0';
7366         }
7367         return total_len;
7368 }
7369
7370 static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7371                 void *buffer, size_t size, int type)
7372 {
7373         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7374
7375         if (strcmp(name, "") == 0)
7376                 return -EINVAL;
7377         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7378                 return -EOPNOTSUPP;
7379         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
7380                                buffer, size);
7381 }
7382
7383 static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7384                 const void *value, size_t size, int flags, int type)
7385 {
7386         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7387
7388         if (strcmp(name, "") == 0)
7389                 return -EINVAL;
7390         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7391                 return -EOPNOTSUPP;
7392
7393         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7394                                name, value, size, flags);
7395 }
7396
7397 const struct xattr_handler ocfs2_xattr_user_handler = {
7398         .prefix = XATTR_USER_PREFIX,
7399         .list   = ocfs2_xattr_user_list,
7400         .get    = ocfs2_xattr_user_get,
7401         .set    = ocfs2_xattr_user_set,
7402 };