fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  11  *
  12  * This program is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU General Public
  14  * License version 2 as published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  */
  21
  22 #include <linux/capability.h>
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/uio.h>
  29 #include <linux/sched.h>
  30 #include <linux/splice.h>
  31 #include <linux/mount.h>
  32 #include <linux/writeback.h>
  33 #include <linux/falloc.h>
  34 #include <linux/sort.h>
  35 #include <linux/init.h>
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/security.h>
  39
  40 #include <cluster/masklog.h>
  41
  42 #include "ocfs2.h"
  43 #include "alloc.h"
  44 #include "blockcheck.h"
  45 #include "dlmglue.h"
  46 #include "file.h"
  47 #include "symlink.h"
  48 #include "sysfile.h"
  49 #include "inode.h"
  50 #include "journal.h"
  51 #include "ocfs2_fs.h"
  52 #include "suballoc.h"
  53 #include "uptodate.h"
  54 #include "buffer_head_io.h"
  55 #include "super.h"
  56 #include "xattr.h"
  57 #include "refcounttree.h"
  58 #include "acl.h"
  59 #include "ocfs2_trace.h"
  60
  61 struct ocfs2_xattr_def_value_root {
  62         struct ocfs2_xattr_value_root   xv;
  63         struct ocfs2_extent_rec         er;
  64 };
  65
  66 struct ocfs2_xattr_bucket {
  67         /* The inode these xattrs are associated with */
  68         struct inode *bu_inode;
  69
  70         /* The actual buffers that make up the bucket */
  71         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  72
  73         /* How many blocks make up one bucket for this filesystem */
  74         int bu_blocks;
  75 };
  76
  77 struct ocfs2_xattr_set_ctxt {
  78         handle_t *handle;
  79         struct ocfs2_alloc_context *meta_ac;
  80         struct ocfs2_alloc_context *data_ac;
  81         struct ocfs2_cached_dealloc_ctxt dealloc;
  82         int set_abort;
  83 };
  84
  85 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  86 #define OCFS2_XATTR_INLINE_SIZE 80
  87 #define OCFS2_XATTR_HEADER_GAP  4
  88 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  89                                          - sizeof(struct ocfs2_xattr_header) \
  90                                          - OCFS2_XATTR_HEADER_GAP)
  91 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  92                                          - sizeof(struct ocfs2_xattr_block) \
  93                                          - sizeof(struct ocfs2_xattr_header) \
  94                                          - OCFS2_XATTR_HEADER_GAP)
  95
  96 static struct ocfs2_xattr_def_value_root def_xv = {
  97         .xv.xr_list.l_count = cpu_to_le16(1),
  98 };
  99
 100 const struct xattr_handler *ocfs2_xattr_handlers[] = {
 101         &ocfs2_xattr_user_handler,
 102         &posix_acl_access_xattr_handler,
 103         &posix_acl_default_xattr_handler,
 104         &ocfs2_xattr_trusted_handler,
 105         &ocfs2_xattr_security_handler,
 106         NULL
 107 };
 108
 109 static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 110         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 111         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 112                                         = &posix_acl_access_xattr_handler,
 113         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 114                                         = &posix_acl_default_xattr_handler,
 115         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 116         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 117 };
 118
 119 struct ocfs2_xattr_info {
 120         int             xi_name_index;
 121         const char      *xi_name;
 122         int             xi_name_len;
 123         const void      *xi_value;
 124         size_t          xi_value_len;
 125 };
 126
 127 struct ocfs2_xattr_search {
 128         struct buffer_head *inode_bh;
 129         /*
 130          * xattr_bh point to the block buffer head which has extended attribute
 131          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 132          */
 133         struct buffer_head *xattr_bh;
 134         struct ocfs2_xattr_header *header;
 135         struct ocfs2_xattr_bucket *bucket;
 136         void *base;
 137         void *end;
 138         struct ocfs2_xattr_entry *here;
 139         int not_found;
 140 };
 141
 142 /* Operations on struct ocfs2_xa_entry */
 143 struct ocfs2_xa_loc;
 144 struct ocfs2_xa_loc_operations {
 145         /*
 146          * Journal functions
 147          */
 148         int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
 149                                   int type);
 150         void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
 151
 152         /*
 153          * Return a pointer to the appropriate buffer in loc->xl_storage
 154          * at the given offset from loc->xl_header.
 155          */
 156         void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
 157
 158         /* Can we reuse the existing entry for the new value? */
 159         int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
 160                              struct ocfs2_xattr_info *xi);
 161
 162         /* How much space is needed for the new value? */
 163         int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
 164                                struct ocfs2_xattr_info *xi);
 165
 166         /*
 167          * Return the offset of the first name+value pair.  This is
 168          * the start of our downward-filling free space.
 169          */
 170         int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
 171
 172         /*
 173          * Remove the name+value at this location.  Do whatever is
 174          * appropriate with the remaining name+value pairs.
 175          */
 176         void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
 177
 178         /* Fill xl_entry with a new entry */
 179         void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
 180
 181         /* Add name+value storage to an entry */
 182         void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
 183
 184         /*
 185          * Initialize the value buf's access and bh fields for this entry.
 186          * ocfs2_xa_fill_value_buf() will handle the xv pointer.
 187          */
 188         void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
 189                                    struct ocfs2_xattr_value_buf *vb);
 190 };
 191
 192 /*
 193  * Describes an xattr entry location.  This is a memory structure
 194  * tracking the on-disk structure.
 195  */
 196 struct ocfs2_xa_loc {
 197         /* This xattr belongs to this inode */
 198         struct inode *xl_inode;
 199
 200         /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
 201         struct ocfs2_xattr_header *xl_header;
 202
 203         /* Bytes from xl_header to the end of the storage */
 204         int xl_size;
 205
 206         /*
 207          * The ocfs2_xattr_entry this location describes.  If this is
 208          * NULL, this location describes the on-disk structure where it
 209          * would have been.
 210          */
 211         struct ocfs2_xattr_entry *xl_entry;
 212
 213         /*
 214          * Internal housekeeping
 215          */
 216
 217         /* Buffer(s) containing this entry */
 218         void *xl_storage;
 219
 220         /* Operations on the storage backing this location */
 221         const struct ocfs2_xa_loc_operations *xl_ops;
 222 };
 223
 224 /*
 225  * Convenience functions to calculate how much space is needed for a
 226  * given name+value pair
 227  */
 228 static int namevalue_size(int name_len, uint64_t value_len)
 229 {
 230         if (value_len > OCFS2_XATTR_INLINE_SIZE)
 231                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 232         else
 233                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 234 }
 235
 236 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
 237 {
 238         return namevalue_size(xi->xi_name_len, xi->xi_value_len);
 239 }
 240
 241 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
 242 {
 243         u64 value_len = le64_to_cpu(xe->xe_value_size);
 244
 245         BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
 246                ocfs2_xattr_is_local(xe));
 247         return namevalue_size(xe->xe_name_len, value_len);
 248 }
 249
 250
 251 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 252                                              struct ocfs2_xattr_header *xh,
 253                                              int index,
 254                                              int *block_off,
 255                                              int *new_offset);
 256
 257 static int ocfs2_xattr_block_find(struct inode *inode,
 258                                   int name_index,
 259                                   const char *name,
 260                                   struct ocfs2_xattr_search *xs);
 261 static int ocfs2_xattr_index_block_find(struct inode *inode,
 262                                         struct buffer_head *root_bh,
 263                                         int name_index,
 264                                         const char *name,
 265                                         struct ocfs2_xattr_search *xs);
 266
 267 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 268                                         struct buffer_head *blk_bh,
 269                                         char *buffer,
 270                                         size_t buffer_size);
 271
 272 static int ocfs2_xattr_create_index_block(struct inode *inode,
 273                                           struct ocfs2_xattr_search *xs,
 274                                           struct ocfs2_xattr_set_ctxt *ctxt);
 275
 276 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 277                                              struct ocfs2_xattr_info *xi,
 278                                              struct ocfs2_xattr_search *xs,
 279                                              struct ocfs2_xattr_set_ctxt *ctxt);
 280
 281 typedef int (xattr_tree_rec_func)(struct inode *inode,
 282                                   struct buffer_head *root_bh,
 283                                   u64 blkno, u32 cpos, u32 len, void *para);
 284 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
 285                                            struct buffer_head *root_bh,
 286                                            xattr_tree_rec_func *rec_func,
 287                                            void *para);
 288 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
 289                                         struct ocfs2_xattr_bucket *bucket,
 290                                         void *para);
 291 static int ocfs2_rm_xattr_cluster(struct inode *inode,
 292                                   struct buffer_head *root_bh,
 293                                   u64 blkno,
 294                                   u32 cpos,
 295                                   u32 len,
 296                                   void *para);
 297
 298 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 299                                   u64 src_blk, u64 last_blk, u64 to_blk,
 300                                   unsigned int start_bucket,
 301                                   u32 *first_hash);
 302 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
 303                                         struct ocfs2_dinode *di,
 304                                         struct ocfs2_xattr_info *xi,
 305                                         struct ocfs2_xattr_search *xis,
 306                                         struct ocfs2_xattr_search *xbs,
 307                                         struct ocfs2_refcount_tree **ref_tree,
 308                                         int *meta_need,
 309                                         int *credits);
 310 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
 311                                            struct ocfs2_xattr_bucket *bucket,
 312                                            int offset,
 313                                            struct ocfs2_xattr_value_root **xv,
 314                                            struct buffer_head **bh);
 315
 316 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 317 {
 318         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 319 }
 320
 321 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 322 {
 323         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 324 }
 325
 326 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 327 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 328 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 329
 330 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 331 {
 332         struct ocfs2_xattr_bucket *bucket;
 333         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 334
 335         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 336
 337         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 338         if (bucket) {
 339                 bucket->bu_inode = inode;
 340                 bucket->bu_blocks = blks;
 341         }
 342
 343         return bucket;
 344 }
 345
 346 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 347 {
 348         int i;
 349
 350         for (i = 0; i < bucket->bu_blocks; i++) {
 351                 brelse(bucket->bu_bhs[i]);
 352                 bucket->bu_bhs[i] = NULL;
 353         }
 354 }
 355
 356 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 357 {
 358         if (bucket) {
 359                 ocfs2_xattr_bucket_relse(bucket);
 360                 bucket->bu_inode = NULL;
 361                 kfree(bucket);
 362         }
 363 }
 364
 365 /*
 366  * A bucket that has never been written to disk doesn't need to be
 367  * read.  We just need the buffer_heads.  Don't call this for
 368  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 369  * them fully.
 370  */
 371 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 372                                    u64 xb_blkno, int new)
 373 {
 374         int i, rc = 0;
 375
 376         for (i = 0; i < bucket->bu_blocks; i++) {
 377                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 378                                               xb_blkno + i);
 379                 if (!bucket->bu_bhs[i]) {
 380                         rc = -ENOMEM;
 381                         mlog_errno(rc);
 382                         break;
 383                 }
 384
 385                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 386                                            bucket->bu_bhs[i])) {
 387                         if (new)
 388                                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 389                                                               bucket->bu_bhs[i]);
 390                         else {
 391                                 set_buffer_uptodate(bucket->bu_bhs[i]);
 392                                 ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 393                                                           bucket->bu_bhs[i]);
 394                         }
 395                 }
 396         }
 397
 398         if (rc)
 399                 ocfs2_xattr_bucket_relse(bucket);
 400         return rc;
 401 }
 402
 403 /* Read the xattr bucket at xb_blkno */
 404 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 405                                    u64 xb_blkno)
 406 {
 407         int rc;
 408
 409         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 410                                bucket->bu_blocks, bucket->bu_bhs, 0,
 411                                NULL);
 412         if (!rc) {
 413                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 414                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 415                                                  bucket->bu_bhs,
 416                                                  bucket->bu_blocks,
 417                                                  &bucket_xh(bucket)->xh_check);
 418                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 419                 if (rc)
 420                         mlog_errno(rc);
 421         }
 422
 423         if (rc)
 424                 ocfs2_xattr_bucket_relse(bucket);
 425         return rc;
 426 }
 427
 428 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 429                                              struct ocfs2_xattr_bucket *bucket,
 430                                              int type)
 431 {
 432         int i, rc = 0;
 433
 434         for (i = 0; i < bucket->bu_blocks; i++) {
 435                 rc = ocfs2_journal_access(handle,
 436                                           INODE_CACHE(bucket->bu_inode),
 437                                           bucket->bu_bhs[i], type);
 438                 if (rc) {
 439                         mlog_errno(rc);
 440                         break;
 441                 }
 442         }
 443
 444         return rc;
 445 }
 446
 447 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 448                                              struct ocfs2_xattr_bucket *bucket)
 449 {
 450         int i;
 451
 452         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 453         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 454                                    bucket->bu_bhs, bucket->bu_blocks,
 455                                    &bucket_xh(bucket)->xh_check);
 456         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 457
 458         for (i = 0; i < bucket->bu_blocks; i++)
 459                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 460 }
 461
 462 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 463                                          struct ocfs2_xattr_bucket *src)
 464 {
 465         int i;
 466         int blocksize = src->bu_inode->i_sb->s_blocksize;
 467
 468         BUG_ON(dest->bu_blocks != src->bu_blocks);
 469         BUG_ON(dest->bu_inode != src->bu_inode);
 470
 471         for (i = 0; i < src->bu_blocks; i++) {
 472                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 473                        blocksize);
 474         }
 475 }
 476
 477 static int ocfs2_validate_xattr_block(struct super_block *sb,
 478                                       struct buffer_head *bh)
 479 {
 480         int rc;
 481         struct ocfs2_xattr_block *xb =
 482                 (struct ocfs2_xattr_block *)bh->b_data;
 483
 484         trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
 485
 486         BUG_ON(!buffer_uptodate(bh));
 487
 488         /*
 489          * If the ecc fails, we return the error but otherwise
 490          * leave the filesystem running.  We know any error is
 491          * local to this block.
 492          */
 493         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 494         if (rc)
 495                 return rc;
 496
 497         /*
 498          * Errors after here are fatal
 499          */
 500
 501         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 502                 return ocfs2_error(sb,
 503                                    "Extended attribute block #%llu has bad signature %.*s\n",
 504                                    (unsigned long long)bh->b_blocknr, 7,
 505                                    xb->xb_signature);
 506         }
 507
 508         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 509                 return ocfs2_error(sb,
 510                                    "Extended attribute block #%llu has an invalid xb_blkno of %llu\n",
 511                                    (unsigned long long)bh->b_blocknr,
 512                                    (unsigned long long)le64_to_cpu(xb->xb_blkno));
 513         }
 514
 515         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 516                 return ocfs2_error(sb,
 517                                    "Extended attribute block #%llu has an invalid xb_fs_generation of #%u\n",
 518                                    (unsigned long long)bh->b_blocknr,
 519                                    le32_to_cpu(xb->xb_fs_generation));
 520         }
 521
 522         return 0;
 523 }
 524
 525 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 526                                   struct buffer_head **bh)
 527 {
 528         int rc;
 529         struct buffer_head *tmp = *bh;
 530
 531         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 532                               ocfs2_validate_xattr_block);
 533
 534         /* If ocfs2_read_block() got us a new bh, pass it up. */
 535         if (!rc && !*bh)
 536                 *bh = tmp;
 537
 538         return rc;
 539 }
 540
 541 static inline const char *ocfs2_xattr_prefix(int name_index)
 542 {
 543         const struct xattr_handler *handler = NULL;
 544
 545         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 546                 handler = ocfs2_xattr_handler_map[name_index];
 547         return handler ? xattr_prefix(handler) : NULL;
 548 }
 549
 550 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 551                                  const char *name,
 552                                  int name_len)
 553 {
 554         /* Get hash value of uuid from super block */
 555         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 556         int i;
 557
 558         /* hash extended attribute name */
 559         for (i = 0; i < name_len; i++) {
 560                 hash = (hash << OCFS2_HASH_SHIFT) ^
 561                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 562                        *name++;
 563         }
 564
 565         return hash;
 566 }
 567
 568 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 569 {
 570         return namevalue_size(name_len, value_len) +
 571                 sizeof(struct ocfs2_xattr_entry);
 572 }
 573
 574 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 575 {
 576         return namevalue_size_xi(xi) +
 577                 sizeof(struct ocfs2_xattr_entry);
 578 }
 579
 580 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 581 {
 582         return namevalue_size_xe(xe) +
 583                 sizeof(struct ocfs2_xattr_entry);
 584 }
 585
 586 int ocfs2_calc_security_init(struct inode *dir,
 587                              struct ocfs2_security_xattr_info *si,
 588                              int *want_clusters,
 589                              int *xattr_credits,
 590                              struct ocfs2_alloc_context **xattr_ac)
 591 {
 592         int ret = 0;
 593         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 594         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 595                                                  si->value_len);
 596
 597         /*
 598          * The max space of security xattr taken inline is
 599          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 600          * So reserve one metadata block for it is ok.
 601          */
 602         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 603             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 604                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 605                 if (ret) {
 606                         mlog_errno(ret);
 607                         return ret;
 608                 }
 609                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 610         }
 611
 612         /* reserve clusters for xattr value which will be set in B tree*/
 613         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 614                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 615                                                             si->value_len);
 616
 617                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 618                                                            new_clusters);
 619                 *want_clusters += new_clusters;
 620         }
 621         return ret;
 622 }
 623
 624 int ocfs2_calc_xattr_init(struct inode *dir,
 625                           struct buffer_head *dir_bh,
 626                           umode_t mode,
 627                           struct ocfs2_security_xattr_info *si,
 628                           int *want_clusters,
 629                           int *xattr_credits,
 630                           int *want_meta)
 631 {
 632         int ret = 0;
 633         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 634         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 635
 636         if (si->enable)
 637                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 638                                                      si->value_len);
 639
 640         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 641                 down_read(&OCFS2_I(dir)->ip_xattr_sem);
 642                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 643                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 644                                         "", NULL, 0);
 645                 up_read(&OCFS2_I(dir)->ip_xattr_sem);
 646                 if (acl_len > 0) {
 647                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 648                         if (S_ISDIR(mode))
 649                                 a_size <<= 1;
 650                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 651                         ret = acl_len;
 652                         mlog_errno(ret);
 653                         return ret;
 654                 }
 655         }
 656
 657         if (!(s_size + a_size))
 658                 return ret;
 659
 660         /*
 661          * The max space of security xattr taken inline is
 662          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 663          * The max space of acl xattr taken inline is
 664          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 665          * when blocksize = 512, may reserve one more cluser for
 666          * xattr bucket, otherwise reserve one metadata block
 667          * for them is ok.
 668          * If this is a new directory with inline data,
 669          * we choose to reserve the entire inline area for
 670          * directory contents and force an external xattr block.
 671          */
 672         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 673             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 674             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 675                 *want_meta = *want_meta + 1;
 676                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 677         }
 678
 679         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 680             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 681                 *want_clusters += 1;
 682                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 683         }
 684
 685         /*
 686          * reserve credits and clusters for xattrs which has large value
 687          * and have to be set outside
 688          */
 689         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 690                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 691                                                         si->value_len);
 692                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 693                                                            new_clusters);
 694                 *want_clusters += new_clusters;
 695         }
 696         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 697             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 698                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 699                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 700                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 701                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 702                                                            new_clusters);
 703                 *want_clusters += new_clusters;
 704         }
 705
 706         return ret;
 707 }
 708
 709 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 710                                          u32 clusters_to_add,
 711                                          struct ocfs2_xattr_value_buf *vb,
 712                                          struct ocfs2_xattr_set_ctxt *ctxt)
 713 {
 714         int status = 0, credits;
 715         handle_t *handle = ctxt->handle;
 716         enum ocfs2_alloc_restarted why;
 717         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 718         struct ocfs2_extent_tree et;
 719
 720         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 721
 722         while (clusters_to_add) {
 723                 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 724
 725                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 726                                        OCFS2_JOURNAL_ACCESS_WRITE);
 727                 if (status < 0) {
 728                         mlog_errno(status);
 729                         break;
 730                 }
 731
 732                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 733                 status = ocfs2_add_clusters_in_btree(handle,
 734                                                      &et,
 735                                                      &logical_start,
 736                                                      clusters_to_add,
 737                                                      0,
 738                                                      ctxt->data_ac,
 739                                                      ctxt->meta_ac,
 740                                                      &why);
 741                 if ((status < 0) && (status != -EAGAIN)) {
 742                         if (status != -ENOSPC)
 743                                 mlog_errno(status);
 744                         break;
 745                 }
 746
 747                 ocfs2_journal_dirty(handle, vb->vb_bh);
 748
 749                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 750                                          prev_clusters;
 751
 752                 if (why != RESTART_NONE && clusters_to_add) {
 753                         /*
 754                          * We can only fail in case the alloc file doesn't give
 755                          * up enough clusters.
 756                          */
 757                         BUG_ON(why == RESTART_META);
 758
 759                         credits = ocfs2_calc_extend_credits(inode->i_sb,
 760                                                             &vb->vb_xv->xr_list);
 761                         status = ocfs2_extend_trans(handle, credits);
 762                         if (status < 0) {
 763                                 status = -ENOMEM;
 764                                 mlog_errno(status);
 765                                 break;
 766                         }
 767                 }
 768         }
 769
 770         return status;
 771 }
 772
 773 static int __ocfs2_remove_xattr_range(struct inode *inode,
 774                                       struct ocfs2_xattr_value_buf *vb,
 775                                       u32 cpos, u32 phys_cpos, u32 len,
 776                                       unsigned int ext_flags,
 777                                       struct ocfs2_xattr_set_ctxt *ctxt)
 778 {
 779         int ret;
 780         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 781         handle_t *handle = ctxt->handle;
 782         struct ocfs2_extent_tree et;
 783
 784         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 785
 786         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 787                             OCFS2_JOURNAL_ACCESS_WRITE);
 788         if (ret) {
 789                 mlog_errno(ret);
 790                 goto out;
 791         }
 792
 793         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 794                                   &ctxt->dealloc);
 795         if (ret) {
 796                 mlog_errno(ret);
 797                 goto out;
 798         }
 799
 800         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 801         ocfs2_journal_dirty(handle, vb->vb_bh);
 802
 803         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 804                 ret = ocfs2_decrease_refcount(inode, handle,
 805                                         ocfs2_blocks_to_clusters(inode->i_sb,
 806                                                                  phys_blkno),
 807                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 808         else
 809                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 810                                                   phys_blkno, len);
 811         if (ret)
 812                 mlog_errno(ret);
 813
 814 out:
 815         return ret;
 816 }
 817
 818 static int ocfs2_xattr_shrink_size(struct inode *inode,
 819                                    u32 old_clusters,
 820                                    u32 new_clusters,
 821                                    struct ocfs2_xattr_value_buf *vb,
 822                                    struct ocfs2_xattr_set_ctxt *ctxt)
 823 {
 824         int ret = 0;
 825         unsigned int ext_flags;
 826         u32 trunc_len, cpos, phys_cpos, alloc_size;
 827         u64 block;
 828
 829         if (old_clusters <= new_clusters)
 830                 return 0;
 831
 832         cpos = new_clusters;
 833         trunc_len = old_clusters - new_clusters;
 834         while (trunc_len) {
 835                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 836                                                &alloc_size,
 837                                                &vb->vb_xv->xr_list, &ext_flags);
 838                 if (ret) {
 839                         mlog_errno(ret);
 840                         goto out;
 841                 }
 842
 843                 if (alloc_size > trunc_len)
 844                         alloc_size = trunc_len;
 845
 846                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 847                                                  phys_cpos, alloc_size,
 848                                                  ext_flags, ctxt);
 849                 if (ret) {
 850                         mlog_errno(ret);
 851                         goto out;
 852                 }
 853
 854                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 855                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 856                                                        block, alloc_size);
 857                 cpos += alloc_size;
 858                 trunc_len -= alloc_size;
 859         }
 860
 861 out:
 862         return ret;
 863 }
 864
 865 static int ocfs2_xattr_value_truncate(struct inode *inode,
 866                                       struct ocfs2_xattr_value_buf *vb,
 867                                       int len,
 868                                       struct ocfs2_xattr_set_ctxt *ctxt)
 869 {
 870         int ret;
 871         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 872         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 873
 874         if (new_clusters == old_clusters)
 875                 return 0;
 876
 877         if (new_clusters > old_clusters)
 878                 ret = ocfs2_xattr_extend_allocation(inode,
 879                                                     new_clusters - old_clusters,
 880                                                     vb, ctxt);
 881         else
 882                 ret = ocfs2_xattr_shrink_size(inode,
 883                                               old_clusters, new_clusters,
 884                                               vb, ctxt);
 885
 886         return ret;
 887 }
 888
 889 static int ocfs2_xattr_list_entry(struct super_block *sb,
 890                                   char *buffer, size_t size,
 891                                   size_t *result, int type,
 892                                   const char *name, int name_len)
 893 {
 894         char *p = buffer + *result;
 895         const char *prefix;
 896         int prefix_len;
 897         int total_len;
 898
 899         switch(type) {
 900         case OCFS2_XATTR_INDEX_USER:
 901                 if (OCFS2_SB(sb)->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
 902                         return 0;
 903                 break;
 904
 905         case OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS:
 906         case OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT:
 907                 if (!(sb->s_flags & SB_POSIXACL))
 908                         return 0;
 909                 break;
 910
 911         case OCFS2_XATTR_INDEX_TRUSTED:
 912                 if (!capable(CAP_SYS_ADMIN))
 913                         return 0;
 914                 break;
 915         }
 916
 917         prefix = ocfs2_xattr_prefix(type);
 918         if (!prefix)
 919                 return 0;
 920         prefix_len = strlen(prefix);
 921         total_len = prefix_len + name_len + 1;
 922         *result += total_len;
 923
 924         /* we are just looking for how big our buffer needs to be */
 925         if (!size)
 926                 return 0;
 927
 928         if (*result > size)
 929                 return -ERANGE;
 930
 931         memcpy(p, prefix, prefix_len);
 932         memcpy(p + prefix_len, name, name_len);
 933         p[prefix_len + name_len] = '\0';
 934
 935         return 0;
 936 }
 937
 938 static int ocfs2_xattr_list_entries(struct inode *inode,
 939                                     struct ocfs2_xattr_header *header,
 940                                     char *buffer, size_t buffer_size)
 941 {
 942         size_t result = 0;
 943         int i, type, ret;
 944         const char *name;
 945
 946         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 947                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 948                 type = ocfs2_xattr_get_type(entry);
 949                 name = (const char *)header +
 950                         le16_to_cpu(entry->xe_name_offset);
 951
 952                 ret = ocfs2_xattr_list_entry(inode->i_sb,
 953                                              buffer, buffer_size,
 954                                              &result, type, name,
 955                                              entry->xe_name_len);
 956                 if (ret)
 957                         return ret;
 958         }
 959
 960         return result;
 961 }
 962
 963 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 964                                          struct ocfs2_dinode *di)
 965 {
 966         struct ocfs2_xattr_header *xh;
 967         int i;
 968
 969         xh = (struct ocfs2_xattr_header *)
 970                  ((void *)di + inode->i_sb->s_blocksize -
 971                  le16_to_cpu(di->i_xattr_inline_size));
 972
 973         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 974                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 975                         return 1;
 976
 977         return 0;
 978 }
 979
 980 static int ocfs2_xattr_ibody_list(struct inode *inode,
 981                                   struct ocfs2_dinode *di,
 982                                   char *buffer,
 983                                   size_t buffer_size)
 984 {
 985         struct ocfs2_xattr_header *header = NULL;
 986         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 987         int ret = 0;
 988
 989         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 990                 return ret;
 991
 992         header = (struct ocfs2_xattr_header *)
 993                  ((void *)di + inode->i_sb->s_blocksize -
 994                  le16_to_cpu(di->i_xattr_inline_size));
 995
 996         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 997
 998         return ret;
 999 }
1000
1001 static int ocfs2_xattr_block_list(struct inode *inode,
1002                                   struct ocfs2_dinode *di,
1003                                   char *buffer,
1004                                   size_t buffer_size)
1005 {
1006         struct buffer_head *blk_bh = NULL;
1007         struct ocfs2_xattr_block *xb;
1008         int ret = 0;
1009
1010         if (!di->i_xattr_loc)
1011                 return ret;
1012
1013         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
1014                                      &blk_bh);
1015         if (ret < 0) {
1016                 mlog_errno(ret);
1017                 return ret;
1018         }
1019
1020         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1021         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1022                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1023                 ret = ocfs2_xattr_list_entries(inode, header,
1024                                                buffer, buffer_size);
1025         } else
1026                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1027                                                    buffer, buffer_size);
1028
1029         brelse(blk_bh);
1030
1031         return ret;
1032 }
1033
1034 ssize_t ocfs2_listxattr(struct dentry *dentry,
1035                         char *buffer,
1036                         size_t size)
1037 {
1038         int ret = 0, i_ret = 0, b_ret = 0;
1039         struct buffer_head *di_bh = NULL;
1040         struct ocfs2_dinode *di = NULL;
1041         struct ocfs2_inode_info *oi = OCFS2_I(d_inode(dentry));
1042
1043         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1044                 return -EOPNOTSUPP;
1045
1046         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1047                 return ret;
1048
1049         ret = ocfs2_inode_lock(d_inode(dentry), &di_bh, 0);
1050         if (ret < 0) {
1051                 mlog_errno(ret);
1052                 return ret;
1053         }
1054
1055         di = (struct ocfs2_dinode *)di_bh->b_data;
1056
1057         down_read(&oi->ip_xattr_sem);
1058         i_ret = ocfs2_xattr_ibody_list(d_inode(dentry), di, buffer, size);
1059         if (i_ret < 0)
1060                 b_ret = 0;
1061         else {
1062                 if (buffer) {
1063                         buffer += i_ret;
1064                         size -= i_ret;
1065                 }
1066                 b_ret = ocfs2_xattr_block_list(d_inode(dentry), di,
1067                                                buffer, size);
1068                 if (b_ret < 0)
1069                         i_ret = 0;
1070         }
1071         up_read(&oi->ip_xattr_sem);
1072         ocfs2_inode_unlock(d_inode(dentry), 0);
1073
1074         brelse(di_bh);
1075
1076         return i_ret + b_ret;
1077 }
1078
1079 static int ocfs2_xattr_find_entry(int name_index,
1080                                   const char *name,
1081                                   struct ocfs2_xattr_search *xs)
1082 {
1083         struct ocfs2_xattr_entry *entry;
1084         size_t name_len;
1085         int i, cmp = 1;
1086
1087         if (name == NULL)
1088                 return -EINVAL;
1089
1090         name_len = strlen(name);
1091         entry = xs->here;
1092         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1093                 cmp = name_index - ocfs2_xattr_get_type(entry);
1094                 if (!cmp)
1095                         cmp = name_len - entry->xe_name_len;
1096                 if (!cmp)
1097                         cmp = memcmp(name, (xs->base +
1098                                      le16_to_cpu(entry->xe_name_offset)),
1099                                      name_len);
1100                 if (cmp == 0)
1101                         break;
1102                 entry += 1;
1103         }
1104         xs->here = entry;
1105
1106         return cmp ? -ENODATA : 0;
1107 }
1108
1109 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1110                                          struct ocfs2_xattr_value_root *xv,
1111                                          void *buffer,
1112                                          size_t len)
1113 {
1114         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1115         u64 blkno;
1116         int i, ret = 0;
1117         size_t cplen, blocksize;
1118         struct buffer_head *bh = NULL;
1119         struct ocfs2_extent_list *el;
1120
1121         el = &xv->xr_list;
1122         clusters = le32_to_cpu(xv->xr_clusters);
1123         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1124         blocksize = inode->i_sb->s_blocksize;
1125
1126         cpos = 0;
1127         while (cpos < clusters) {
1128                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1129                                                &num_clusters, el, NULL);
1130                 if (ret) {
1131                         mlog_errno(ret);
1132                         goto out;
1133                 }
1134
1135                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1136                 /* Copy ocfs2_xattr_value */
1137                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1138                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1139                                                &bh, NULL);
1140                         if (ret) {
1141                                 mlog_errno(ret);
1142                                 goto out;
1143                         }
1144
1145                         cplen = len >= blocksize ? blocksize : len;
1146                         memcpy(buffer, bh->b_data, cplen);
1147                         len -= cplen;
1148                         buffer += cplen;
1149
1150                         brelse(bh);
1151                         bh = NULL;
1152                         if (len == 0)
1153                                 break;
1154                 }
1155                 cpos += num_clusters;
1156         }
1157 out:
1158         return ret;
1159 }
1160
1161 static int ocfs2_xattr_ibody_get(struct inode *inode,
1162                                  int name_index,
1163                                  const char *name,
1164                                  void *buffer,
1165                                  size_t buffer_size,
1166                                  struct ocfs2_xattr_search *xs)
1167 {
1168         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1169         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1170         struct ocfs2_xattr_value_root *xv;
1171         size_t size;
1172         int ret = 0;
1173
1174         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1175                 return -ENODATA;
1176
1177         xs->end = (void *)di + inode->i_sb->s_blocksize;
1178         xs->header = (struct ocfs2_xattr_header *)
1179                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1180         xs->base = (void *)xs->header;
1181         xs->here = xs->header->xh_entries;
1182
1183         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1184         if (ret)
1185                 return ret;
1186         size = le64_to_cpu(xs->here->xe_value_size);
1187         if (buffer) {
1188                 if (size > buffer_size)
1189                         return -ERANGE;
1190                 if (ocfs2_xattr_is_local(xs->here)) {
1191                         memcpy(buffer, (void *)xs->base +
1192                                le16_to_cpu(xs->here->xe_name_offset) +
1193                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1194                 } else {
1195                         xv = (struct ocfs2_xattr_value_root *)
1196                                 (xs->base + le16_to_cpu(
1197                                  xs->here->xe_name_offset) +
1198                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1199                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1200                                                             buffer, size);
1201                         if (ret < 0) {
1202                                 mlog_errno(ret);
1203                                 return ret;
1204                         }
1205                 }
1206         }
1207
1208         return size;
1209 }
1210
1211 static int ocfs2_xattr_block_get(struct inode *inode,
1212                                  int name_index,
1213                                  const char *name,
1214                                  void *buffer,
1215                                  size_t buffer_size,
1216                                  struct ocfs2_xattr_search *xs)
1217 {
1218         struct ocfs2_xattr_block *xb;
1219         struct ocfs2_xattr_value_root *xv;
1220         size_t size;
1221         int ret = -ENODATA, name_offset, name_len, i;
1222         int uninitialized_var(block_off);
1223
1224         xs->bucket = ocfs2_xattr_bucket_new(inode);
1225         if (!xs->bucket) {
1226                 ret = -ENOMEM;
1227                 mlog_errno(ret);
1228                 goto cleanup;
1229         }
1230
1231         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1232         if (ret) {
1233                 mlog_errno(ret);
1234                 goto cleanup;
1235         }
1236
1237         if (xs->not_found) {
1238                 ret = -ENODATA;
1239                 goto cleanup;
1240         }
1241
1242         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1243         size = le64_to_cpu(xs->here->xe_value_size);
1244         if (buffer) {
1245                 ret = -ERANGE;
1246                 if (size > buffer_size)
1247                         goto cleanup;
1248
1249                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1250                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1251                 i = xs->here - xs->header->xh_entries;
1252
1253                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1254                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1255                                                                 bucket_xh(xs->bucket),
1256                                                                 i,
1257                                                                 &block_off,
1258                                                                 &name_offset);
1259                         if (ret) {
1260                                 mlog_errno(ret);
1261                                 goto cleanup;
1262                         }
1263                         xs->base = bucket_block(xs->bucket, block_off);
1264                 }
1265                 if (ocfs2_xattr_is_local(xs->here)) {
1266                         memcpy(buffer, (void *)xs->base +
1267                                name_offset + name_len, size);
1268                 } else {
1269                         xv = (struct ocfs2_xattr_value_root *)
1270                                 (xs->base + name_offset + name_len);
1271                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1272                                                             buffer, size);
1273                         if (ret < 0) {
1274                                 mlog_errno(ret);
1275                                 goto cleanup;
1276                         }
1277                 }
1278         }
1279         ret = size;
1280 cleanup:
1281         ocfs2_xattr_bucket_free(xs->bucket);
1282
1283         brelse(xs->xattr_bh);
1284         xs->xattr_bh = NULL;
1285         return ret;
1286 }
1287
1288 int ocfs2_xattr_get_nolock(struct inode *inode,
1289                            struct buffer_head *di_bh,
1290                            int name_index,
1291                            const char *name,
1292                            void *buffer,
1293                            size_t buffer_size)
1294 {
1295         int ret;
1296         struct ocfs2_dinode *di = NULL;
1297         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1298         struct ocfs2_xattr_search xis = {
1299                 .not_found = -ENODATA,
1300         };
1301         struct ocfs2_xattr_search xbs = {
1302                 .not_found = -ENODATA,
1303         };
1304
1305         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1306                 return -EOPNOTSUPP;
1307
1308         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1309                 return -ENODATA;
1310
1311         xis.inode_bh = xbs.inode_bh = di_bh;
1312         di = (struct ocfs2_dinode *)di_bh->b_data;
1313
1314         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1315                                     buffer_size, &xis);
1316         if (ret == -ENODATA && di->i_xattr_loc)
1317                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1318                                             buffer_size, &xbs);
1319
1320         return ret;
1321 }
1322
1323 /* ocfs2_xattr_get()
1324  *
1325  * Copy an extended attribute into the buffer provided.
1326  * Buffer is NULL to compute the size of buffer required.
1327  */
1328 static int ocfs2_xattr_get(struct inode *inode,
1329                            int name_index,
1330                            const char *name,
1331                            void *buffer,
1332                            size_t buffer_size)
1333 {
1334         int ret, had_lock;
1335         struct buffer_head *di_bh = NULL;
1336         struct ocfs2_lock_holder oh;
1337
1338         had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
1339         if (had_lock < 0) {
1340                 mlog_errno(had_lock);
1341                 return had_lock;
1342         }
1343         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1344         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1345                                      name, buffer, buffer_size);
1346         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1347
1348         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1349
1350         brelse(di_bh);
1351
1352         return ret;
1353 }
1354
1355 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1356                                            handle_t *handle,
1357                                            struct ocfs2_xattr_value_buf *vb,
1358                                            const void *value,
1359                                            int value_len)
1360 {
1361         int ret = 0, i, cp_len;
1362         u16 blocksize = inode->i_sb->s_blocksize;
1363         u32 p_cluster, num_clusters;
1364         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1365         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1366         u64 blkno;
1367         struct buffer_head *bh = NULL;
1368         unsigned int ext_flags;
1369         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1370
1371         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1372
1373         while (cpos < clusters) {
1374                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1375                                                &num_clusters, &xv->xr_list,
1376                                                &ext_flags);
1377                 if (ret) {
1378                         mlog_errno(ret);
1379                         goto out;
1380                 }
1381
1382                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1383
1384                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1385
1386                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1387                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1388                                                &bh, NULL);
1389                         if (ret) {
1390                                 mlog_errno(ret);
1391                                 goto out;
1392                         }
1393
1394                         ret = ocfs2_journal_access(handle,
1395                                                    INODE_CACHE(inode),
1396                                                    bh,
1397                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1398                         if (ret < 0) {
1399                                 mlog_errno(ret);
1400                                 goto out;
1401                         }
1402
1403                         cp_len = value_len > blocksize ? blocksize : value_len;
1404                         memcpy(bh->b_data, value, cp_len);
1405                         value_len -= cp_len;
1406                         value += cp_len;
1407                         if (cp_len < blocksize)
1408                                 memset(bh->b_data + cp_len, 0,
1409                                        blocksize - cp_len);
1410
1411                         ocfs2_journal_dirty(handle, bh);
1412                         brelse(bh);
1413                         bh = NULL;
1414
1415                         /*
1416                          * XXX: do we need to empty all the following
1417                          * blocks in this cluster?
1418                          */
1419                         if (!value_len)
1420                                 break;
1421                 }
1422                 cpos += num_clusters;
1423         }
1424 out:
1425         brelse(bh);
1426
1427         return ret;
1428 }
1429
1430 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1431                                        int num_entries)
1432 {
1433         int free_space;
1434
1435         if (!needed_space)
1436                 return 0;
1437
1438         free_space = free_start -
1439                 sizeof(struct ocfs2_xattr_header) -
1440                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1441                 OCFS2_XATTR_HEADER_GAP;
1442         if (free_space < 0)
1443                 return -EIO;
1444         if (free_space < needed_space)
1445                 return -ENOSPC;
1446
1447         return 0;
1448 }
1449
1450 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1451                                    int type)
1452 {
1453         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1454 }
1455
1456 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1457 {
1458         loc->xl_ops->xlo_journal_dirty(handle, loc);
1459 }
1460
1461 /* Give a pointer into the storage for the given offset */
1462 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1463 {
1464         BUG_ON(offset >= loc->xl_size);
1465         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1466 }
1467
1468 /*
1469  * Wipe the name+value pair and allow the storage to reclaim it.  This
1470  * must be followed by either removal of the entry or a call to
1471  * ocfs2_xa_add_namevalue().
1472  */
1473 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1474 {
1475         loc->xl_ops->xlo_wipe_namevalue(loc);
1476 }
1477
1478 /*
1479  * Find lowest offset to a name+value pair.  This is the start of our
1480  * downward-growing free space.
1481  */
1482 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1483 {
1484         return loc->xl_ops->xlo_get_free_start(loc);
1485 }
1486
1487 /* Can we reuse loc->xl_entry for xi? */
1488 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1489                                     struct ocfs2_xattr_info *xi)
1490 {
1491         return loc->xl_ops->xlo_can_reuse(loc, xi);
1492 }
1493
1494 /* How much free space is needed to set the new value */
1495 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1496                                 struct ocfs2_xattr_info *xi)
1497 {
1498         return loc->xl_ops->xlo_check_space(loc, xi);
1499 }
1500
1501 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1502 {
1503         loc->xl_ops->xlo_add_entry(loc, name_hash);
1504         loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1505         /*
1506          * We can't leave the new entry's xe_name_offset at zero or
1507          * add_namevalue() will go nuts.  We set it to the size of our
1508          * storage so that it can never be less than any other entry.
1509          */
1510         loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1511 }
1512
1513 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1514                                    struct ocfs2_xattr_info *xi)
1515 {
1516         int size = namevalue_size_xi(xi);
1517         int nameval_offset;
1518         char *nameval_buf;
1519
1520         loc->xl_ops->xlo_add_namevalue(loc, size);
1521         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1522         loc->xl_entry->xe_name_len = xi->xi_name_len;
1523         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1524         ocfs2_xattr_set_local(loc->xl_entry,
1525                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1526
1527         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1528         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1529         memset(nameval_buf, 0, size);
1530         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1531 }
1532
1533 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1534                                     struct ocfs2_xattr_value_buf *vb)
1535 {
1536         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1537         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1538
1539         /* Value bufs are for value trees */
1540         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1541         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1542                (name_size + OCFS2_XATTR_ROOT_SIZE));
1543
1544         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1545         vb->vb_xv =
1546                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1547                                                         nameval_offset +
1548                                                         name_size);
1549 }
1550
1551 static int ocfs2_xa_block_journal_access(handle_t *handle,
1552                                          struct ocfs2_xa_loc *loc, int type)
1553 {
1554         struct buffer_head *bh = loc->xl_storage;
1555         ocfs2_journal_access_func access;
1556
1557         if (loc->xl_size == (bh->b_size -
1558                              offsetof(struct ocfs2_xattr_block,
1559                                       xb_attrs.xb_header)))
1560                 access = ocfs2_journal_access_xb;
1561         else
1562                 access = ocfs2_journal_access_di;
1563         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1564 }
1565
1566 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1567                                          struct ocfs2_xa_loc *loc)
1568 {
1569         struct buffer_head *bh = loc->xl_storage;
1570
1571         ocfs2_journal_dirty(handle, bh);
1572 }
1573
1574 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1575                                            int offset)
1576 {
1577         return (char *)loc->xl_header + offset;
1578 }
1579
1580 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1581                                     struct ocfs2_xattr_info *xi)
1582 {
1583         /*
1584          * Block storage is strict.  If the sizes aren't exact, we will
1585          * remove the old one and reinsert the new.
1586          */
1587         return namevalue_size_xe(loc->xl_entry) ==
1588                 namevalue_size_xi(xi);
1589 }
1590
1591 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1592 {
1593         struct ocfs2_xattr_header *xh = loc->xl_header;
1594         int i, count = le16_to_cpu(xh->xh_count);
1595         int offset, free_start = loc->xl_size;
1596
1597         for (i = 0; i < count; i++) {
1598                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1599                 if (offset < free_start)
1600                         free_start = offset;
1601         }
1602
1603         return free_start;
1604 }
1605
1606 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1607                                       struct ocfs2_xattr_info *xi)
1608 {
1609         int count = le16_to_cpu(loc->xl_header->xh_count);
1610         int free_start = ocfs2_xa_get_free_start(loc);
1611         int needed_space = ocfs2_xi_entry_usage(xi);
1612
1613         /*
1614          * Block storage will reclaim the original entry before inserting
1615          * the new value, so we only need the difference.  If the new
1616          * entry is smaller than the old one, we don't need anything.
1617          */
1618         if (loc->xl_entry) {
1619                 /* Don't need space if we're reusing! */
1620                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1621                         needed_space = 0;
1622                 else
1623                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1624         }
1625         if (needed_space < 0)
1626                 needed_space = 0;
1627         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1628 }
1629
1630 /*
1631  * Block storage for xattrs keeps the name+value pairs compacted.  When
1632  * we remove one, we have to shift any that preceded it towards the end.
1633  */
1634 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1635 {
1636         int i, offset;
1637         int namevalue_offset, first_namevalue_offset, namevalue_size;
1638         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1639         struct ocfs2_xattr_header *xh = loc->xl_header;
1640         int count = le16_to_cpu(xh->xh_count);
1641
1642         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1643         namevalue_size = namevalue_size_xe(entry);
1644         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1645
1646         /* Shift the name+value pairs */
1647         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1648                 (char *)xh + first_namevalue_offset,
1649                 namevalue_offset - first_namevalue_offset);
1650         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1651
1652         /* Now tell xh->xh_entries about it */
1653         for (i = 0; i < count; i++) {
1654                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1655                 if (offset <= namevalue_offset)
1656                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1657                                      namevalue_size);
1658         }
1659
1660         /*
1661          * Note that we don't update xh_free_start or xh_name_value_len
1662          * because they're not used in block-stored xattrs.
1663          */
1664 }
1665
1666 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1667 {
1668         int count = le16_to_cpu(loc->xl_header->xh_count);
1669         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1670         le16_add_cpu(&loc->xl_header->xh_count, 1);
1671         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1672 }
1673
1674 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1675 {
1676         int free_start = ocfs2_xa_get_free_start(loc);
1677
1678         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1679 }
1680
1681 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1682                                           struct ocfs2_xattr_value_buf *vb)
1683 {
1684         struct buffer_head *bh = loc->xl_storage;
1685
1686         if (loc->xl_size == (bh->b_size -
1687                              offsetof(struct ocfs2_xattr_block,
1688                                       xb_attrs.xb_header)))
1689                 vb->vb_access = ocfs2_journal_access_xb;
1690         else
1691                 vb->vb_access = ocfs2_journal_access_di;
1692         vb->vb_bh = bh;
1693 }
1694
1695 /*
1696  * Operations for xattrs stored in blocks.  This includes inline inode
1697  * storage and unindexed ocfs2_xattr_blocks.
1698  */
1699 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1700         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1701         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1702         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1703         .xlo_check_space        = ocfs2_xa_block_check_space,
1704         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1705         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1706         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1707         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1708         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1709         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1710 };
1711
1712 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1713                                           struct ocfs2_xa_loc *loc, int type)
1714 {
1715         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1716
1717         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1718 }
1719
1720 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1721                                           struct ocfs2_xa_loc *loc)
1722 {
1723         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1724
1725         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1726 }
1727
1728 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1729                                             int offset)
1730 {
1731         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1732         int block, block_offset;
1733
1734         /* The header is at the front of the bucket */
1735         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1736         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1737
1738         return bucket_block(bucket, block) + block_offset;
1739 }
1740
1741 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1742                                      struct ocfs2_xattr_info *xi)
1743 {
1744         return namevalue_size_xe(loc->xl_entry) >=
1745                 namevalue_size_xi(xi);
1746 }
1747
1748 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1749 {
1750         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1751         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1752 }
1753
1754 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1755                                          int free_start, int size)
1756 {
1757         /*
1758          * We need to make sure that the name+value pair fits within
1759          * one block.
1760          */
1761         if (((free_start - size) >> sb->s_blocksize_bits) !=
1762             ((free_start - 1) >> sb->s_blocksize_bits))
1763                 free_start -= free_start % sb->s_blocksize;
1764
1765         return free_start;
1766 }
1767
1768 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1769                                        struct ocfs2_xattr_info *xi)
1770 {
1771         int rc;
1772         int count = le16_to_cpu(loc->xl_header->xh_count);
1773         int free_start = ocfs2_xa_get_free_start(loc);
1774         int needed_space = ocfs2_xi_entry_usage(xi);
1775         int size = namevalue_size_xi(xi);
1776         struct super_block *sb = loc->xl_inode->i_sb;
1777
1778         /*
1779          * Bucket storage does not reclaim name+value pairs it cannot
1780          * reuse.  They live as holes until the bucket fills, and then
1781          * the bucket is defragmented.  However, the bucket can reclaim
1782          * the ocfs2_xattr_entry.
1783          */
1784         if (loc->xl_entry) {
1785                 /* Don't need space if we're reusing! */
1786                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1787                         needed_space = 0;
1788                 else
1789                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1790         }
1791         BUG_ON(needed_space < 0);
1792
1793         if (free_start < size) {
1794                 if (needed_space)
1795                         return -ENOSPC;
1796         } else {
1797                 /*
1798                  * First we check if it would fit in the first place.
1799                  * Below, we align the free start to a block.  This may
1800                  * slide us below the minimum gap.  By checking unaligned
1801                  * first, we avoid that error.
1802                  */
1803                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1804                                                  count);
1805                 if (rc)
1806                         return rc;
1807                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1808                                                            size);
1809         }
1810         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1811 }
1812
1813 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1814 {
1815         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1816                      -namevalue_size_xe(loc->xl_entry));
1817 }
1818
1819 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1820 {
1821         struct ocfs2_xattr_header *xh = loc->xl_header;
1822         int count = le16_to_cpu(xh->xh_count);
1823         int low = 0, high = count - 1, tmp;
1824         struct ocfs2_xattr_entry *tmp_xe;
1825
1826         /*
1827          * We keep buckets sorted by name_hash, so we need to find
1828          * our insert place.
1829          */
1830         while (low <= high && count) {
1831                 tmp = (low + high) / 2;
1832                 tmp_xe = &xh->xh_entries[tmp];
1833
1834                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1835                         low = tmp + 1;
1836                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1837                         high = tmp - 1;
1838                 else {
1839                         low = tmp;
1840                         break;
1841                 }
1842         }
1843
1844         if (low != count)
1845                 memmove(&xh->xh_entries[low + 1],
1846                         &xh->xh_entries[low],
1847                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1848
1849         le16_add_cpu(&xh->xh_count, 1);
1850         loc->xl_entry = &xh->xh_entries[low];
1851         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1852 }
1853
1854 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1855 {
1856         int free_start = ocfs2_xa_get_free_start(loc);
1857         struct ocfs2_xattr_header *xh = loc->xl_header;
1858         struct super_block *sb = loc->xl_inode->i_sb;
1859         int nameval_offset;
1860
1861         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1862         nameval_offset = free_start - size;
1863         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1864         xh->xh_free_start = cpu_to_le16(nameval_offset);
1865         le16_add_cpu(&xh->xh_name_value_len, size);
1866
1867 }
1868
1869 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1870                                            struct ocfs2_xattr_value_buf *vb)
1871 {
1872         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1873         struct super_block *sb = loc->xl_inode->i_sb;
1874         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1875         int size = namevalue_size_xe(loc->xl_entry);
1876         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1877
1878         /* Values are not allowed to straddle block boundaries */
1879         BUG_ON(block_offset !=
1880                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1881         /* We expect the bucket to be filled in */
1882         BUG_ON(!bucket->bu_bhs[block_offset]);
1883
1884         vb->vb_access = ocfs2_journal_access;
1885         vb->vb_bh = bucket->bu_bhs[block_offset];
1886 }
1887
1888 /* Operations for xattrs stored in buckets. */
1889 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1890         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1891         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1892         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1893         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1894         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1895         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1896         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1897         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1898         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1899         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1900 };
1901
1902 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1903 {
1904         struct ocfs2_xattr_value_buf vb;
1905
1906         if (ocfs2_xattr_is_local(loc->xl_entry))
1907                 return 0;
1908
1909         ocfs2_xa_fill_value_buf(loc, &vb);
1910         return le32_to_cpu(vb.vb_xv->xr_clusters);
1911 }
1912
1913 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1914                                    struct ocfs2_xattr_set_ctxt *ctxt)
1915 {
1916         int trunc_rc, access_rc;
1917         struct ocfs2_xattr_value_buf vb;
1918
1919         ocfs2_xa_fill_value_buf(loc, &vb);
1920         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1921                                               ctxt);
1922
1923         /*
1924          * The caller of ocfs2_xa_value_truncate() has already called
1925          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1926          * calls ocfs2_extend_trans().  This may commit the previous
1927          * transaction and open a new one.  If this is a bucket, truncate
1928          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1929          * the caller is expecting to dirty the entire bucket.  So we must
1930          * reset the journal work.  We do this even if truncate has failed,
1931          * as it could have failed after committing the extend.
1932          */
1933         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1934                                             OCFS2_JOURNAL_ACCESS_WRITE);
1935
1936         /* Errors in truncate take precedence */
1937         return trunc_rc ? trunc_rc : access_rc;
1938 }
1939
1940 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1941 {
1942         int index, count;
1943         struct ocfs2_xattr_header *xh = loc->xl_header;
1944         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1945
1946         ocfs2_xa_wipe_namevalue(loc);
1947         loc->xl_entry = NULL;
1948
1949         le16_add_cpu(&xh->xh_count, -1);
1950         count = le16_to_cpu(xh->xh_count);
1951
1952         /*
1953          * Only zero out the entry if there are more remaining.  This is
1954          * important for an empty bucket, as it keeps track of the
1955          * bucket's hash value.  It doesn't hurt empty block storage.
1956          */
1957         if (count) {
1958                 index = ((char *)entry - (char *)&xh->xh_entries) /
1959                         sizeof(struct ocfs2_xattr_entry);
1960                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1961                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1962                 memset(&xh->xh_entries[count], 0,
1963                        sizeof(struct ocfs2_xattr_entry));
1964         }
1965 }
1966
1967 /*
1968  * If we have a problem adjusting the size of an external value during
1969  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1970  * in an intermediate state.  For example, the value may be partially
1971  * truncated.
1972  *
1973  * If the value tree hasn't changed, the extend/truncate went nowhere.
1974  * We have nothing to do.  The caller can treat it as a straight error.
1975  *
1976  * If the value tree got partially truncated, we now have a corrupted
1977  * extended attribute.  We're going to wipe its entry and leak the
1978  * clusters.  Better to leak some storage than leave a corrupt entry.
1979  *
1980  * If the value tree grew, it obviously didn't grow enough for the
1981  * new entry.  We're not going to try and reclaim those clusters either.
1982  * If there was already an external value there (orig_clusters != 0),
1983  * the new clusters are attached safely and we can just leave the old
1984  * value in place.  If there was no external value there, we remove
1985  * the entry.
1986  *
1987  * This way, the xattr block we store in the journal will be consistent.
1988  * If the size change broke because of the journal, no changes will hit
1989  * disk anyway.
1990  */
1991 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1992                                             const char *what,
1993                                             unsigned int orig_clusters)
1994 {
1995         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1996         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1997                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1998
1999         if (new_clusters < orig_clusters) {
2000                 mlog(ML_ERROR,
2001                      "Partial truncate while %s xattr %.*s.  Leaking "
2002                      "%u clusters and removing the entry\n",
2003                      what, loc->xl_entry->xe_name_len, nameval_buf,
2004                      orig_clusters - new_clusters);
2005                 ocfs2_xa_remove_entry(loc);
2006         } else if (!orig_clusters) {
2007                 mlog(ML_ERROR,
2008                      "Unable to allocate an external value for xattr "
2009                      "%.*s safely.  Leaking %u clusters and removing the "
2010                      "entry\n",
2011                      loc->xl_entry->xe_name_len, nameval_buf,
2012                      new_clusters - orig_clusters);
2013                 ocfs2_xa_remove_entry(loc);
2014         } else if (new_clusters > orig_clusters)
2015                 mlog(ML_ERROR,
2016                      "Unable to grow xattr %.*s safely.  %u new clusters "
2017                      "have been added, but the value will not be "
2018                      "modified\n",
2019                      loc->xl_entry->xe_name_len, nameval_buf,
2020                      new_clusters - orig_clusters);
2021 }
2022
2023 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2024                            struct ocfs2_xattr_set_ctxt *ctxt)
2025 {
2026         int rc = 0;
2027         unsigned int orig_clusters;
2028
2029         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2030                 orig_clusters = ocfs2_xa_value_clusters(loc);
2031                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2032                 if (rc) {
2033                         mlog_errno(rc);
2034                         /*
2035                          * Since this is remove, we can return 0 if
2036                          * ocfs2_xa_cleanup_value_truncate() is going to
2037                          * wipe the entry anyway.  So we check the
2038                          * cluster count as well.
2039                          */
2040                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2041                                 rc = 0;
2042                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2043                                                         orig_clusters);
2044                         if (rc)
2045                                 goto out;
2046                 }
2047         }
2048
2049         ocfs2_xa_remove_entry(loc);
2050
2051 out:
2052         return rc;
2053 }
2054
2055 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2056 {
2057         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2058         char *nameval_buf;
2059
2060         nameval_buf = ocfs2_xa_offset_pointer(loc,
2061                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2062         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2063 }
2064
2065 /*
2066  * Take an existing entry and make it ready for the new value.  This
2067  * won't allocate space, but it may free space.  It should be ready for
2068  * ocfs2_xa_prepare_entry() to finish the work.
2069  */
2070 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2071                                 struct ocfs2_xattr_info *xi,
2072                                 struct ocfs2_xattr_set_ctxt *ctxt)
2073 {
2074         int rc = 0;
2075         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2076         unsigned int orig_clusters;
2077         char *nameval_buf;
2078         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2079         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2080
2081         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2082                name_size);
2083
2084         nameval_buf = ocfs2_xa_offset_pointer(loc,
2085                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2086         if (xe_local) {
2087                 memset(nameval_buf + name_size, 0,
2088                        namevalue_size_xe(loc->xl_entry) - name_size);
2089                 if (!xi_local)
2090                         ocfs2_xa_install_value_root(loc);
2091         } else {
2092                 orig_clusters = ocfs2_xa_value_clusters(loc);
2093                 if (xi_local) {
2094                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2095                         if (rc < 0)
2096                                 mlog_errno(rc);
2097                         else
2098                                 memset(nameval_buf + name_size, 0,
2099                                        namevalue_size_xe(loc->xl_entry) -
2100                                        name_size);
2101                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2102                            xi->xi_value_len) {
2103                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2104                                                      ctxt);
2105                         if (rc < 0)
2106                                 mlog_errno(rc);
2107                 }
2108
2109                 if (rc) {
2110                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2111                                                         orig_clusters);
2112                         goto out;
2113                 }
2114         }
2115
2116         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2117         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2118
2119 out:
2120         return rc;
2121 }
2122
2123 /*
2124  * Prepares loc->xl_entry to receive the new xattr.  This includes
2125  * properly setting up the name+value pair region.  If loc->xl_entry
2126  * already exists, it will take care of modifying it appropriately.
2127  *
2128  * Note that this modifies the data.  You did journal_access already,
2129  * right?
2130  */
2131 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2132                                   struct ocfs2_xattr_info *xi,
2133                                   u32 name_hash,
2134                                   struct ocfs2_xattr_set_ctxt *ctxt)
2135 {
2136         int rc = 0;
2137         unsigned int orig_clusters;
2138         __le64 orig_value_size = 0;
2139
2140         rc = ocfs2_xa_check_space(loc, xi);
2141         if (rc)
2142                 goto out;
2143
2144         if (loc->xl_entry) {
2145                 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2146                         orig_value_size = loc->xl_entry->xe_value_size;
2147                         rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2148                         if (rc)
2149                                 goto out;
2150                         goto alloc_value;
2151                 }
2152
2153                 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2154                         orig_clusters = ocfs2_xa_value_clusters(loc);
2155                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2156                         if (rc) {
2157                                 mlog_errno(rc);
2158                                 ocfs2_xa_cleanup_value_truncate(loc,
2159                                                                 "overwriting",
2160                                                                 orig_clusters);
2161                                 goto out;
2162                         }
2163                 }
2164                 ocfs2_xa_wipe_namevalue(loc);
2165         } else
2166                 ocfs2_xa_add_entry(loc, name_hash);
2167
2168         /*
2169          * If we get here, we have a blank entry.  Fill it.  We grow our
2170          * name+value pair back from the end.
2171          */
2172         ocfs2_xa_add_namevalue(loc, xi);
2173         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2174                 ocfs2_xa_install_value_root(loc);
2175
2176 alloc_value:
2177         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2178                 orig_clusters = ocfs2_xa_value_clusters(loc);
2179                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2180                 if (rc < 0) {
2181                         ctxt->set_abort = 1;
2182                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2183                                                         orig_clusters);
2184                         /*
2185                          * If we were growing an existing value,
2186                          * ocfs2_xa_cleanup_value_truncate() won't remove
2187                          * the entry. We need to restore the original value
2188                          * size.
2189                          */
2190                         if (loc->xl_entry) {
2191                                 BUG_ON(!orig_value_size);
2192                                 loc->xl_entry->xe_value_size = orig_value_size;
2193                         }
2194                         mlog_errno(rc);
2195                 }
2196         }
2197
2198 out:
2199         return rc;
2200 }
2201
2202 /*
2203  * Store the value portion of the name+value pair.  This will skip
2204  * values that are stored externally.  Their tree roots were set up
2205  * by ocfs2_xa_prepare_entry().
2206  */
2207 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2208                                 struct ocfs2_xattr_info *xi,
2209                                 struct ocfs2_xattr_set_ctxt *ctxt)
2210 {
2211         int rc = 0;
2212         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2213         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2214         char *nameval_buf;
2215         struct ocfs2_xattr_value_buf vb;
2216
2217         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2218         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2219                 ocfs2_xa_fill_value_buf(loc, &vb);
2220                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2221                                                      ctxt->handle, &vb,
2222                                                      xi->xi_value,
2223                                                      xi->xi_value_len);
2224         } else
2225                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2226
2227         return rc;
2228 }
2229
2230 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2231                         struct ocfs2_xattr_info *xi,
2232                         struct ocfs2_xattr_set_ctxt *ctxt)
2233 {
2234         int ret;
2235         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2236                                               xi->xi_name_len);
2237
2238         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2239                                       OCFS2_JOURNAL_ACCESS_WRITE);
2240         if (ret) {
2241                 mlog_errno(ret);
2242                 goto out;
2243         }
2244
2245         /*
2246          * From here on out, everything is going to modify the buffer a
2247          * little.  Errors are going to leave the xattr header in a
2248          * sane state.  Thus, even with errors we dirty the sucker.
2249          */
2250
2251         /* Don't worry, we are never called with !xi_value and !xl_entry */
2252         if (!xi->xi_value) {
2253                 ret = ocfs2_xa_remove(loc, ctxt);
2254                 goto out_dirty;
2255         }
2256
2257         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2258         if (ret) {
2259                 if (ret != -ENOSPC)
2260                         mlog_errno(ret);
2261                 goto out_dirty;
2262         }
2263
2264         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2265         if (ret)
2266                 mlog_errno(ret);
2267
2268 out_dirty:
2269         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2270
2271 out:
2272         return ret;
2273 }
2274
2275 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2276                                      struct inode *inode,
2277                                      struct buffer_head *bh,
2278                                      struct ocfs2_xattr_entry *entry)
2279 {
2280         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2281
2282         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2283
2284         loc->xl_inode = inode;
2285         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2286         loc->xl_storage = bh;
2287         loc->xl_entry = entry;
2288         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2289         loc->xl_header =
2290                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2291                                               loc->xl_size);
2292 }
2293
2294 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2295                                           struct inode *inode,
2296                                           struct buffer_head *bh,
2297                                           struct ocfs2_xattr_entry *entry)
2298 {
2299         struct ocfs2_xattr_block *xb =
2300                 (struct ocfs2_xattr_block *)bh->b_data;
2301
2302         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2303
2304         loc->xl_inode = inode;
2305         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2306         loc->xl_storage = bh;
2307         loc->xl_header = &(xb->xb_attrs.xb_header);
2308         loc->xl_entry = entry;
2309         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2310                                              xb_attrs.xb_header);
2311 }
2312
2313 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2314                                            struct ocfs2_xattr_bucket *bucket,
2315                                            struct ocfs2_xattr_entry *entry)
2316 {
2317         loc->xl_inode = bucket->bu_inode;
2318         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2319         loc->xl_storage = bucket;
2320         loc->xl_header = bucket_xh(bucket);
2321         loc->xl_entry = entry;
2322         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2323 }
2324
2325 /*
2326  * In xattr remove, if it is stored outside and refcounted, we may have
2327  * the chance to split the refcount tree. So need the allocators.
2328  */
2329 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2330                                         struct ocfs2_xattr_value_root *xv,
2331                                         struct ocfs2_caching_info *ref_ci,
2332                                         struct buffer_head *ref_root_bh,
2333                                         struct ocfs2_alloc_context **meta_ac,
2334                                         int *ref_credits)
2335 {
2336         int ret, meta_add = 0;
2337         u32 p_cluster, num_clusters;
2338         unsigned int ext_flags;
2339
2340         *ref_credits = 0;
2341         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2342                                        &num_clusters,
2343                                        &xv->xr_list,
2344                                        &ext_flags);
2345         if (ret) {
2346                 mlog_errno(ret);
2347                 goto out;
2348         }
2349
2350         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2351                 goto out;
2352
2353         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2354                                                  ref_root_bh, xv,
2355                                                  &meta_add, ref_credits);
2356         if (ret) {
2357                 mlog_errno(ret);
2358                 goto out;
2359         }
2360
2361         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2362                                                 meta_add, meta_ac);
2363         if (ret)
2364                 mlog_errno(ret);
2365
2366 out:
2367         return ret;
2368 }
2369
2370 static int ocfs2_remove_value_outside(struct inode*inode,
2371                                       struct ocfs2_xattr_value_buf *vb,
2372                                       struct ocfs2_xattr_header *header,
2373                                       struct ocfs2_caching_info *ref_ci,
2374                                       struct buffer_head *ref_root_bh)
2375 {
2376         int ret = 0, i, ref_credits;
2377         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2378         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2379         void *val;
2380
2381         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2382
2383         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2384                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2385
2386                 if (ocfs2_xattr_is_local(entry))
2387                         continue;
2388
2389                 val = (void *)header +
2390                         le16_to_cpu(entry->xe_name_offset);
2391                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2392                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2393
2394                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2395                                                          ref_ci, ref_root_bh,
2396                                                          &ctxt.meta_ac,
2397                                                          &ref_credits);
2398
2399                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2400                                         ocfs2_remove_extent_credits(osb->sb));
2401                 if (IS_ERR(ctxt.handle)) {
2402                         ret = PTR_ERR(ctxt.handle);
2403                         mlog_errno(ret);
2404                         break;
2405                 }
2406
2407                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2408
2409                 ocfs2_commit_trans(osb, ctxt.handle);
2410                 if (ctxt.meta_ac) {
2411                         ocfs2_free_alloc_context(ctxt.meta_ac);
2412                         ctxt.meta_ac = NULL;
2413                 }
2414
2415                 if (ret < 0) {
2416                         mlog_errno(ret);
2417                         break;
2418                 }
2419
2420         }
2421
2422         if (ctxt.meta_ac)
2423                 ocfs2_free_alloc_context(ctxt.meta_ac);
2424         ocfs2_schedule_truncate_log_flush(osb, 1);
2425         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2426         return ret;
2427 }
2428
2429 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2430                                     struct buffer_head *di_bh,
2431                                     struct ocfs2_caching_info *ref_ci,
2432                                     struct buffer_head *ref_root_bh)
2433 {
2434
2435         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2436         struct ocfs2_xattr_header *header;
2437         int ret;
2438         struct ocfs2_xattr_value_buf vb = {
2439                 .vb_bh = di_bh,
2440                 .vb_access = ocfs2_journal_access_di,
2441         };
2442
2443         header = (struct ocfs2_xattr_header *)
2444                  ((void *)di + inode->i_sb->s_blocksize -
2445                  le16_to_cpu(di->i_xattr_inline_size));
2446
2447         ret = ocfs2_remove_value_outside(inode, &vb, header,
2448                                          ref_ci, ref_root_bh);
2449
2450         return ret;
2451 }
2452
2453 struct ocfs2_rm_xattr_bucket_para {
2454         struct ocfs2_caching_info *ref_ci;
2455         struct buffer_head *ref_root_bh;
2456 };
2457
2458 static int ocfs2_xattr_block_remove(struct inode *inode,
2459                                     struct buffer_head *blk_bh,
2460                                     struct ocfs2_caching_info *ref_ci,
2461                                     struct buffer_head *ref_root_bh)
2462 {
2463         struct ocfs2_xattr_block *xb;
2464         int ret = 0;
2465         struct ocfs2_xattr_value_buf vb = {
2466                 .vb_bh = blk_bh,
2467                 .vb_access = ocfs2_journal_access_xb,
2468         };
2469         struct ocfs2_rm_xattr_bucket_para args = {
2470                 .ref_ci = ref_ci,
2471                 .ref_root_bh = ref_root_bh,
2472         };
2473
2474         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2475         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2476                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2477                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2478                                                  ref_ci, ref_root_bh);
2479         } else
2480                 ret = ocfs2_iterate_xattr_index_block(inode,
2481                                                 blk_bh,
2482                                                 ocfs2_rm_xattr_cluster,
2483                                                 &args);
2484
2485         return ret;
2486 }
2487
2488 static int ocfs2_xattr_free_block(struct inode *inode,
2489                                   u64 block,
2490                                   struct ocfs2_caching_info *ref_ci,
2491                                   struct buffer_head *ref_root_bh)
2492 {
2493         struct inode *xb_alloc_inode;
2494         struct buffer_head *xb_alloc_bh = NULL;
2495         struct buffer_head *blk_bh = NULL;
2496         struct ocfs2_xattr_block *xb;
2497         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2498         handle_t *handle;
2499         int ret = 0;
2500         u64 blk, bg_blkno;
2501         u16 bit;
2502
2503         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2504         if (ret < 0) {
2505                 mlog_errno(ret);
2506                 goto out;
2507         }
2508
2509         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2510         if (ret < 0) {
2511                 mlog_errno(ret);
2512                 goto out;
2513         }
2514
2515         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2516         blk = le64_to_cpu(xb->xb_blkno);
2517         bit = le16_to_cpu(xb->xb_suballoc_bit);
2518         if (xb->xb_suballoc_loc)
2519                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2520         else
2521                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2522
2523         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2524                                 EXTENT_ALLOC_SYSTEM_INODE,
2525                                 le16_to_cpu(xb->xb_suballoc_slot));
2526         if (!xb_alloc_inode) {
2527                 ret = -ENOMEM;
2528                 mlog_errno(ret);
2529                 goto out;
2530         }
2531         inode_lock(xb_alloc_inode);
2532
2533         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2534         if (ret < 0) {
2535                 mlog_errno(ret);
2536                 goto out_mutex;
2537         }
2538
2539         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2540         if (IS_ERR(handle)) {
2541                 ret = PTR_ERR(handle);
2542                 mlog_errno(ret);
2543                 goto out_unlock;
2544         }
2545
2546         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2547                                        bit, bg_blkno, 1);
2548         if (ret < 0)
2549                 mlog_errno(ret);
2550
2551         ocfs2_commit_trans(osb, handle);
2552 out_unlock:
2553         ocfs2_inode_unlock(xb_alloc_inode, 1);
2554         brelse(xb_alloc_bh);
2555 out_mutex:
2556         inode_unlock(xb_alloc_inode);
2557         iput(xb_alloc_inode);
2558 out:
2559         brelse(blk_bh);
2560         return ret;
2561 }
2562
2563 /*
2564  * ocfs2_xattr_remove()
2565  *
2566  * Free extended attribute resources associated with this inode.
2567  */
2568 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2569 {
2570         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2571         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2572         struct ocfs2_refcount_tree *ref_tree = NULL;
2573         struct buffer_head *ref_root_bh = NULL;
2574         struct ocfs2_caching_info *ref_ci = NULL;
2575         handle_t *handle;
2576         int ret;
2577
2578         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2579                 return 0;
2580
2581         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2582                 return 0;
2583
2584         if (ocfs2_is_refcount_inode(inode)) {
2585                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2586                                                le64_to_cpu(di->i_refcount_loc),
2587                                                1, &ref_tree, &ref_root_bh);
2588                 if (ret) {
2589                         mlog_errno(ret);
2590                         goto out;
2591                 }
2592                 ref_ci = &ref_tree->rf_ci;
2593
2594         }
2595
2596         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2597                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2598                                                ref_ci, ref_root_bh);
2599                 if (ret < 0) {
2600                         mlog_errno(ret);
2601                         goto out;
2602                 }
2603         }
2604
2605         if (di->i_xattr_loc) {
2606                 ret = ocfs2_xattr_free_block(inode,
2607                                              le64_to_cpu(di->i_xattr_loc),
2608                                              ref_ci, ref_root_bh);
2609                 if (ret < 0) {
2610                         mlog_errno(ret);
2611                         goto out;
2612                 }
2613         }
2614
2615         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2616                                    OCFS2_INODE_UPDATE_CREDITS);
2617         if (IS_ERR(handle)) {
2618                 ret = PTR_ERR(handle);
2619                 mlog_errno(ret);
2620                 goto out;
2621         }
2622         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2623                                       OCFS2_JOURNAL_ACCESS_WRITE);
2624         if (ret) {
2625                 mlog_errno(ret);
2626                 goto out_commit;
2627         }
2628
2629         di->i_xattr_loc = 0;
2630
2631         spin_lock(&oi->ip_lock);
2632         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2633         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2634         spin_unlock(&oi->ip_lock);
2635         ocfs2_update_inode_fsync_trans(handle, inode, 0);
2636
2637         ocfs2_journal_dirty(handle, di_bh);
2638 out_commit:
2639         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2640 out:
2641         if (ref_tree)
2642                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2643         brelse(ref_root_bh);
2644         return ret;
2645 }
2646
2647 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2648                                         struct ocfs2_dinode *di)
2649 {
2650         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2651         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2652         int free;
2653
2654         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2655                 return 0;
2656
2657         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2658                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2659                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2660         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2661                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2662                         le64_to_cpu(di->i_size);
2663         } else {
2664                 struct ocfs2_extent_list *el = &di->id2.i_list;
2665                 free = (le16_to_cpu(el->l_count) -
2666                         le16_to_cpu(el->l_next_free_rec)) *
2667                         sizeof(struct ocfs2_extent_rec);
2668         }
2669         if (free >= xattrsize)
2670                 return 1;
2671
2672         return 0;
2673 }
2674
2675 /*
2676  * ocfs2_xattr_ibody_find()
2677  *
2678  * Find extended attribute in inode block and
2679  * fill search info into struct ocfs2_xattr_search.
2680  */
2681 static int ocfs2_xattr_ibody_find(struct inode *inode,
2682                                   int name_index,
2683                                   const char *name,
2684                                   struct ocfs2_xattr_search *xs)
2685 {
2686         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2687         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2688         int ret;
2689         int has_space = 0;
2690
2691         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2692                 return 0;
2693
2694         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2695                 down_read(&oi->ip_alloc_sem);
2696                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2697                 up_read(&oi->ip_alloc_sem);
2698                 if (!has_space)
2699                         return 0;
2700         }
2701
2702         xs->xattr_bh = xs->inode_bh;
2703         xs->end = (void *)di + inode->i_sb->s_blocksize;
2704         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2705                 xs->header = (struct ocfs2_xattr_header *)
2706                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2707         else
2708                 xs->header = (struct ocfs2_xattr_header *)
2709                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2710         xs->base = (void *)xs->header;
2711         xs->here = xs->header->xh_entries;
2712
2713         /* Find the named attribute. */
2714         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2715                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2716                 if (ret && ret != -ENODATA)
2717                         return ret;
2718                 xs->not_found = ret;
2719         }
2720
2721         return 0;
2722 }
2723
2724 static int ocfs2_xattr_ibody_init(struct inode *inode,
2725                                   struct buffer_head *di_bh,
2726                                   struct ocfs2_xattr_set_ctxt *ctxt)
2727 {
2728         int ret;
2729         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2730         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2731         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2732         unsigned int xattrsize = osb->s_xattr_inline_size;
2733
2734         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2735                 ret = -ENOSPC;
2736                 goto out;
2737         }
2738
2739         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2740                                       OCFS2_JOURNAL_ACCESS_WRITE);
2741         if (ret) {
2742                 mlog_errno(ret);
2743                 goto out;
2744         }
2745
2746         /*
2747          * Adjust extent record count or inline data size
2748          * to reserve space for extended attribute.
2749          */
2750         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2751                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2752                 le16_add_cpu(&idata->id_count, -xattrsize);
2753         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2754                 struct ocfs2_extent_list *el = &di->id2.i_list;
2755                 le16_add_cpu(&el->l_count, -(xattrsize /
2756                                              sizeof(struct ocfs2_extent_rec)));
2757         }
2758         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2759
2760         spin_lock(&oi->ip_lock);
2761         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2762         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2763         spin_unlock(&oi->ip_lock);
2764
2765         ocfs2_journal_dirty(ctxt->handle, di_bh);
2766
2767 out:
2768         return ret;
2769 }
2770
2771 /*
2772  * ocfs2_xattr_ibody_set()
2773  *
2774  * Set, replace or remove an extended attribute into inode block.
2775  *
2776  */
2777 static int ocfs2_xattr_ibody_set(struct inode *inode,
2778                                  struct ocfs2_xattr_info *xi,
2779                                  struct ocfs2_xattr_search *xs,
2780                                  struct ocfs2_xattr_set_ctxt *ctxt)
2781 {
2782         int ret;
2783         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2784         struct ocfs2_xa_loc loc;
2785
2786         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2787                 return -ENOSPC;
2788
2789         down_write(&oi->ip_alloc_sem);
2790         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2791                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2792                 if (ret) {
2793                         if (ret != -ENOSPC)
2794                                 mlog_errno(ret);
2795                         goto out;
2796                 }
2797         }
2798
2799         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2800                                  xs->not_found ? NULL : xs->here);
2801         ret = ocfs2_xa_set(&loc, xi, ctxt);
2802         if (ret) {
2803                 if (ret != -ENOSPC)
2804                         mlog_errno(ret);
2805                 goto out;
2806         }
2807         xs->here = loc.xl_entry;
2808
2809 out:
2810         up_write(&oi->ip_alloc_sem);
2811
2812         return ret;
2813 }
2814
2815 /*
2816  * ocfs2_xattr_block_find()
2817  *
2818  * Find extended attribute in external block and
2819  * fill search info into struct ocfs2_xattr_search.
2820  */
2821 static int ocfs2_xattr_block_find(struct inode *inode,
2822                                   int name_index,
2823                                   const char *name,
2824                                   struct ocfs2_xattr_search *xs)
2825 {
2826         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2827         struct buffer_head *blk_bh = NULL;
2828         struct ocfs2_xattr_block *xb;
2829         int ret = 0;
2830
2831         if (!di->i_xattr_loc)
2832                 return ret;
2833
2834         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2835                                      &blk_bh);
2836         if (ret < 0) {
2837                 mlog_errno(ret);
2838                 return ret;
2839         }
2840
2841         xs->xattr_bh = blk_bh;
2842         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2843
2844         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2845                 xs->header = &xb->xb_attrs.xb_header;
2846                 xs->base = (void *)xs->header;
2847                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2848                 xs->here = xs->header->xh_entries;
2849
2850                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2851         } else
2852                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2853                                                    name_index,
2854                                                    name, xs);
2855
2856         if (ret && ret != -ENODATA) {
2857                 xs->xattr_bh = NULL;
2858                 goto cleanup;
2859         }
2860         xs->not_found = ret;
2861         return 0;
2862 cleanup:
2863         brelse(blk_bh);
2864
2865         return ret;
2866 }
2867
2868 static int ocfs2_create_xattr_block(struct inode *inode,
2869                                     struct buffer_head *inode_bh,
2870                                     struct ocfs2_xattr_set_ctxt *ctxt,
2871                                     int indexed,
2872                                     struct buffer_head **ret_bh)
2873 {
2874         int ret;
2875         u16 suballoc_bit_start;
2876         u32 num_got;
2877         u64 suballoc_loc, first_blkno;
2878         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2879         struct buffer_head *new_bh = NULL;
2880         struct ocfs2_xattr_block *xblk;
2881
2882         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2883                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2884         if (ret < 0) {
2885                 mlog_errno(ret);
2886                 goto end;
2887         }
2888
2889         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2890                                    &suballoc_loc, &suballoc_bit_start,
2891                                    &num_got, &first_blkno);
2892         if (ret < 0) {
2893                 mlog_errno(ret);
2894                 goto end;
2895         }
2896
2897         new_bh = sb_getblk(inode->i_sb, first_blkno);
2898         if (!new_bh) {
2899                 ret = -ENOMEM;
2900                 mlog_errno(ret);
2901                 goto end;
2902         }
2903
2904         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2905
2906         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2907                                       new_bh,
2908                                       OCFS2_JOURNAL_ACCESS_CREATE);
2909         if (ret < 0) {
2910                 mlog_errno(ret);
2911                 goto end;
2912         }
2913
2914         /* Initialize ocfs2_xattr_block */
2915         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2916         memset(xblk, 0, inode->i_sb->s_blocksize);
2917         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2918         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2919         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2920         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2921         xblk->xb_fs_generation =
2922                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2923         xblk->xb_blkno = cpu_to_le64(first_blkno);
2924         if (indexed) {
2925                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2926                 xr->xt_clusters = cpu_to_le32(1);
2927                 xr->xt_last_eb_blk = 0;
2928                 xr->xt_list.l_tree_depth = 0;
2929                 xr->xt_list.l_count = cpu_to_le16(
2930                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2931                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2932                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2933         }
2934         ocfs2_journal_dirty(ctxt->handle, new_bh);
2935
2936         /* Add it to the inode */
2937         di->i_xattr_loc = cpu_to_le64(first_blkno);
2938
2939         spin_lock(&OCFS2_I(inode)->ip_lock);
2940         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2941         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2942         spin_unlock(&OCFS2_I(inode)->ip_lock);
2943
2944         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2945
2946         *ret_bh = new_bh;
2947         new_bh = NULL;
2948
2949 end:
2950         brelse(new_bh);
2951         return ret;
2952 }
2953
2954 /*
2955  * ocfs2_xattr_block_set()
2956  *
2957  * Set, replace or remove an extended attribute into external block.
2958  *
2959  */
2960 static int ocfs2_xattr_block_set(struct inode *inode,
2961                                  struct ocfs2_xattr_info *xi,
2962                                  struct ocfs2_xattr_search *xs,
2963                                  struct ocfs2_xattr_set_ctxt *ctxt)
2964 {
2965         struct buffer_head *new_bh = NULL;
2966         struct ocfs2_xattr_block *xblk = NULL;
2967         int ret;
2968         struct ocfs2_xa_loc loc;
2969
2970         if (!xs->xattr_bh) {
2971                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2972                                                0, &new_bh);
2973                 if (ret) {
2974                         mlog_errno(ret);
2975                         goto end;
2976                 }
2977
2978                 xs->xattr_bh = new_bh;
2979                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2980                 xs->header = &xblk->xb_attrs.xb_header;
2981                 xs->base = (void *)xs->header;
2982                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2983                 xs->here = xs->header->xh_entries;
2984         } else
2985                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2986
2987         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2988                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2989                                               xs->not_found ? NULL : xs->here);
2990
2991                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2992                 if (!ret)
2993                         xs->here = loc.xl_entry;
2994                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2995                         goto end;
2996                 else {
2997                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2998                         if (ret)
2999                                 goto end;
3000                 }
3001         }
3002
3003         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
3004                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
3005
3006 end:
3007         return ret;
3008 }
3009
3010 /* Check whether the new xattr can be inserted into the inode. */
3011 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
3012                                        struct ocfs2_xattr_info *xi,
3013                                        struct ocfs2_xattr_search *xs)
3014 {
3015         struct ocfs2_xattr_entry *last;
3016         int free, i;
3017         size_t min_offs = xs->end - xs->base;
3018
3019         if (!xs->header)
3020                 return 0;
3021
3022         last = xs->header->xh_entries;
3023
3024         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3025                 size_t offs = le16_to_cpu(last->xe_name_offset);
3026                 if (offs < min_offs)
3027                         min_offs = offs;
3028                 last += 1;
3029         }
3030
3031         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3032         if (free < 0)
3033                 return 0;
3034
3035         BUG_ON(!xs->not_found);
3036
3037         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3038                 return 1;
3039
3040         return 0;
3041 }
3042
3043 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3044                                      struct ocfs2_dinode *di,
3045                                      struct ocfs2_xattr_info *xi,
3046                                      struct ocfs2_xattr_search *xis,
3047                                      struct ocfs2_xattr_search *xbs,
3048                                      int *clusters_need,
3049                                      int *meta_need,
3050                                      int *credits_need)
3051 {
3052         int ret = 0, old_in_xb = 0;
3053         int clusters_add = 0, meta_add = 0, credits = 0;
3054         struct buffer_head *bh = NULL;
3055         struct ocfs2_xattr_block *xb = NULL;
3056         struct ocfs2_xattr_entry *xe = NULL;
3057         struct ocfs2_xattr_value_root *xv = NULL;
3058         char *base = NULL;
3059         int name_offset, name_len = 0;
3060         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3061                                                     xi->xi_value_len);
3062         u64 value_size;
3063
3064         /*
3065          * Calculate the clusters we need to write.
3066          * No matter whether we replace an old one or add a new one,
3067          * we need this for writing.
3068          */
3069         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3070                 credits += new_clusters *
3071                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3072
3073         if (xis->not_found && xbs->not_found) {
3074                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3075
3076                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3077                         clusters_add += new_clusters;
3078                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3079                                                         &def_xv.xv.xr_list);
3080                 }
3081
3082                 goto meta_guess;
3083         }
3084
3085         if (!xis->not_found) {
3086                 xe = xis->here;
3087                 name_offset = le16_to_cpu(xe->xe_name_offset);
3088                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3089                 base = xis->base;
3090                 credits += OCFS2_INODE_UPDATE_CREDITS;
3091         } else {
3092                 int i, block_off = 0;
3093                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3094                 xe = xbs->here;
3095                 name_offset = le16_to_cpu(xe->xe_name_offset);
3096                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3097                 i = xbs->here - xbs->header->xh_entries;
3098                 old_in_xb = 1;
3099
3100                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3101                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3102                                                         bucket_xh(xbs->bucket),
3103                                                         i, &block_off,
3104                                                         &name_offset);
3105                         base = bucket_block(xbs->bucket, block_off);
3106                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3107                 } else {
3108                         base = xbs->base;
3109                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3110                 }
3111         }
3112
3113         /*
3114          * delete a xattr doesn't need metadata and cluster allocation.
3115          * so just calculate the credits and return.
3116          *
3117          * The credits for removing the value tree will be extended
3118          * by ocfs2_remove_extent itself.
3119          */
3120         if (!xi->xi_value) {
3121                 if (!ocfs2_xattr_is_local(xe))
3122                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3123
3124                 goto out;
3125         }
3126
3127         /* do cluster allocation guess first. */
3128         value_size = le64_to_cpu(xe->xe_value_size);
3129
3130         if (old_in_xb) {
3131                 /*
3132                  * In xattr set, we always try to set the xe in inode first,
3133                  * so if it can be inserted into inode successfully, the old
3134                  * one will be removed from the xattr block, and this xattr
3135                  * will be inserted into inode as a new xattr in inode.
3136                  */
3137                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3138                         clusters_add += new_clusters;
3139                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3140                                     OCFS2_INODE_UPDATE_CREDITS;
3141                         if (!ocfs2_xattr_is_local(xe))
3142                                 credits += ocfs2_calc_extend_credits(
3143                                                         inode->i_sb,
3144                                                         &def_xv.xv.xr_list);
3145                         goto out;
3146                 }
3147         }
3148
3149         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3150                 /* the new values will be stored outside. */
3151                 u32 old_clusters = 0;
3152
3153                 if (!ocfs2_xattr_is_local(xe)) {
3154                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3155                                                                  value_size);
3156                         xv = (struct ocfs2_xattr_value_root *)
3157                              (base + name_offset + name_len);
3158                         value_size = OCFS2_XATTR_ROOT_SIZE;
3159                 } else
3160                         xv = &def_xv.xv;
3161
3162                 if (old_clusters >= new_clusters) {
3163                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3164                         goto out;
3165                 } else {
3166                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3167                         clusters_add += new_clusters - old_clusters;
3168                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3169                                                              &xv->xr_list);
3170                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3171                                 goto out;
3172                 }
3173         } else {
3174                 /*
3175                  * Now the new value will be stored inside. So if the new
3176                  * value is smaller than the size of value root or the old
3177                  * value, we don't need any allocation, otherwise we have
3178                  * to guess metadata allocation.
3179                  */
3180                 if ((ocfs2_xattr_is_local(xe) &&
3181                      (value_size >= xi->xi_value_len)) ||
3182                     (!ocfs2_xattr_is_local(xe) &&
3183                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3184                         goto out;
3185         }
3186
3187 meta_guess:
3188         /* calculate metadata allocation. */
3189         if (di->i_xattr_loc) {
3190                 if (!xbs->xattr_bh) {
3191                         ret = ocfs2_read_xattr_block(inode,
3192                                                      le64_to_cpu(di->i_xattr_loc),
3193                                                      &bh);
3194                         if (ret) {
3195                                 mlog_errno(ret);
3196                                 goto out;
3197                         }
3198
3199                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3200                 } else
3201                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3202
3203                 /*
3204                  * If there is already an xattr tree, good, we can calculate
3205                  * like other b-trees. Otherwise we may have the chance of
3206                  * create a tree, the credit calculation is borrowed from
3207                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3208                  * new tree will be cluster based, so no meta is needed.
3209                  */
3210                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3211                         struct ocfs2_extent_list *el =
3212                                  &xb->xb_attrs.xb_root.xt_list;
3213                         meta_add += ocfs2_extend_meta_needed(el);
3214                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3215                                                              el);
3216                 } else
3217                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3218
3219                 /*
3220                  * This cluster will be used either for new bucket or for
3221                  * new xattr block.
3222                  * If the cluster size is the same as the bucket size, one
3223                  * more is needed since we may need to extend the bucket
3224                  * also.
3225                  */
3226                 clusters_add += 1;
3227                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3228                 if (OCFS2_XATTR_BUCKET_SIZE ==
3229                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3230                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3231                         clusters_add += 1;
3232                 }
3233         } else {
3234                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3235                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3236                         struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3237                         meta_add += ocfs2_extend_meta_needed(el);
3238                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3239                                                              el);
3240                 } else {
3241                         meta_add += 1;
3242                 }
3243         }
3244 out:
3245         if (clusters_need)
3246                 *clusters_need = clusters_add;
3247         if (meta_need)
3248                 *meta_need = meta_add;
3249         if (credits_need)
3250                 *credits_need = credits;
3251         brelse(bh);
3252         return ret;
3253 }
3254
3255 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3256                                      struct ocfs2_dinode *di,
3257                                      struct ocfs2_xattr_info *xi,
3258                                      struct ocfs2_xattr_search *xis,
3259                                      struct ocfs2_xattr_search *xbs,
3260                                      struct ocfs2_xattr_set_ctxt *ctxt,
3261                                      int extra_meta,
3262                                      int *credits)
3263 {
3264         int clusters_add, meta_add, ret;
3265         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3266
3267         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3268
3269         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3270
3271         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3272                                         &clusters_add, &meta_add, credits);
3273         if (ret) {
3274                 mlog_errno(ret);
3275                 return ret;
3276         }
3277
3278         meta_add += extra_meta;
3279         trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3280                                         clusters_add, *credits);
3281
3282         if (meta_add) {
3283                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3284                                                         &ctxt->meta_ac);
3285                 if (ret) {
3286                         mlog_errno(ret);
3287                         goto out;
3288                 }
3289         }
3290
3291         if (clusters_add) {
3292                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3293                 if (ret)
3294                         mlog_errno(ret);
3295         }
3296 out:
3297         if (ret) {
3298                 if (ctxt->meta_ac) {
3299                         ocfs2_free_alloc_context(ctxt->meta_ac);
3300                         ctxt->meta_ac = NULL;
3301                 }
3302
3303                 /*
3304                  * We cannot have an error and a non null ctxt->data_ac.
3305                  */
3306         }
3307
3308         return ret;
3309 }
3310
3311 static int __ocfs2_xattr_set_handle(struct inode *inode,
3312                                     struct ocfs2_dinode *di,
3313                                     struct ocfs2_xattr_info *xi,
3314                                     struct ocfs2_xattr_search *xis,
3315                                     struct ocfs2_xattr_search *xbs,
3316                                     struct ocfs2_xattr_set_ctxt *ctxt)
3317 {
3318         int ret = 0, credits, old_found;
3319
3320         if (!xi->xi_value) {
3321                 /* Remove existing extended attribute */
3322                 if (!xis->not_found)
3323                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3324                 else if (!xbs->not_found)
3325                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3326         } else {
3327                 /* We always try to set extended attribute into inode first*/
3328                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3329                 if (!ret && !xbs->not_found) {
3330                         /*
3331                          * If succeed and that extended attribute existing in
3332                          * external block, then we will remove it.
3333                          */
3334                         xi->xi_value = NULL;
3335                         xi->xi_value_len = 0;
3336
3337                         old_found = xis->not_found;
3338                         xis->not_found = -ENODATA;
3339                         ret = ocfs2_calc_xattr_set_need(inode,
3340                                                         di,
3341                                                         xi,
3342                                                         xis,
3343                                                         xbs,
3344                                                         NULL,
3345                                                         NULL,
3346                                                         &credits);
3347                         xis->not_found = old_found;
3348                         if (ret) {
3349                                 mlog_errno(ret);
3350                                 goto out;
3351                         }
3352
3353                         ret = ocfs2_extend_trans(ctxt->handle, credits);
3354                         if (ret) {
3355                                 mlog_errno(ret);
3356                                 goto out;
3357                         }
3358                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3359                 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3360                         if (di->i_xattr_loc && !xbs->xattr_bh) {
3361                                 ret = ocfs2_xattr_block_find(inode,
3362                                                              xi->xi_name_index,
3363                                                              xi->xi_name, xbs);
3364                                 if (ret)
3365                                         goto out;
3366
3367                                 old_found = xis->not_found;
3368                                 xis->not_found = -ENODATA;
3369                                 ret = ocfs2_calc_xattr_set_need(inode,
3370                                                                 di,
3371                                                                 xi,
3372                                                                 xis,
3373                                                                 xbs,
3374                                                                 NULL,
3375                                                                 NULL,
3376                                                                 &credits);
3377                                 xis->not_found = old_found;
3378                                 if (ret) {
3379                                         mlog_errno(ret);
3380                                         goto out;
3381                                 }
3382
3383                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3384                                 if (ret) {
3385                                         mlog_errno(ret);
3386                                         goto out;
3387                                 }
3388                         }
3389                         /*
3390                          * If no space in inode, we will set extended attribute
3391                          * into external block.
3392                          */
3393                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3394                         if (ret)
3395                                 goto out;
3396                         if (!xis->not_found) {
3397                                 /*
3398                                  * If succeed and that extended attribute
3399                                  * existing in inode, we will remove it.
3400                                  */
3401                                 xi->xi_value = NULL;
3402                                 xi->xi_value_len = 0;
3403                                 xbs->not_found = -ENODATA;
3404                                 ret = ocfs2_calc_xattr_set_need(inode,
3405                                                                 di,
3406                                                                 xi,
3407                                                                 xis,
3408                                                                 xbs,
3409                                                                 NULL,
3410                                                                 NULL,
3411                                                                 &credits);
3412                                 if (ret) {
3413                                         mlog_errno(ret);
3414                                         goto out;
3415                                 }
3416
3417                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3418                                 if (ret) {
3419                                         mlog_errno(ret);
3420                                         goto out;
3421                                 }
3422                                 ret = ocfs2_xattr_ibody_set(inode, xi,
3423                                                             xis, ctxt);
3424                         }
3425                 }
3426         }
3427
3428         if (!ret) {
3429                 /* Update inode ctime. */
3430                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3431                                               xis->inode_bh,
3432                                               OCFS2_JOURNAL_ACCESS_WRITE);
3433                 if (ret) {
3434                         mlog_errno(ret);
3435                         goto out;
3436                 }
3437
3438                 inode->i_ctime = current_time(inode);
3439                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3440                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3441                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3442         }
3443 out:
3444         return ret;
3445 }
3446
3447 /*
3448  * This function only called duing creating inode
3449  * for init security/acl xattrs of the new inode.
3450  * All transanction credits have been reserved in mknod.
3451  */
3452 int ocfs2_xattr_set_handle(handle_t *handle,
3453                            struct inode *inode,
3454                            struct buffer_head *di_bh,
3455                            int name_index,
3456                            const char *name,
3457                            const void *value,
3458                            size_t value_len,
3459                            int flags,
3460                            struct ocfs2_alloc_context *meta_ac,
3461                            struct ocfs2_alloc_context *data_ac)
3462 {
3463         struct ocfs2_dinode *di;
3464         int ret;
3465
3466         struct ocfs2_xattr_info xi = {
3467                 .xi_name_index = name_index,
3468                 .xi_name = name,
3469                 .xi_name_len = strlen(name),
3470                 .xi_value = value,
3471                 .xi_value_len = value_len,
3472         };
3473
3474         struct ocfs2_xattr_search xis = {
3475                 .not_found = -ENODATA,
3476         };
3477
3478         struct ocfs2_xattr_search xbs = {
3479                 .not_found = -ENODATA,
3480         };
3481
3482         struct ocfs2_xattr_set_ctxt ctxt = {
3483                 .handle = handle,
3484                 .meta_ac = meta_ac,
3485                 .data_ac = data_ac,
3486         };
3487
3488         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3489                 return -EOPNOTSUPP;
3490
3491         /*
3492          * In extreme situation, may need xattr bucket when
3493          * block size is too small. And we have already reserved
3494          * the credits for bucket in mknod.
3495          */
3496         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3497                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3498                 if (!xbs.bucket) {
3499                         mlog_errno(-ENOMEM);
3500                         return -ENOMEM;
3501                 }
3502         }
3503
3504         xis.inode_bh = xbs.inode_bh = di_bh;
3505         di = (struct ocfs2_dinode *)di_bh->b_data;
3506
3507         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3508
3509         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3510         if (ret)
3511                 goto cleanup;
3512         if (xis.not_found) {
3513                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3514                 if (ret)
3515                         goto cleanup;
3516         }
3517
3518         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3519
3520 cleanup:
3521         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3522         brelse(xbs.xattr_bh);
3523         ocfs2_xattr_bucket_free(xbs.bucket);
3524
3525         return ret;
3526 }
3527
3528 /*
3529  * ocfs2_xattr_set()
3530  *
3531  * Set, replace or remove an extended attribute for this inode.
3532  * value is NULL to remove an existing extended attribute, else either
3533  * create or replace an extended attribute.
3534  */
3535 int ocfs2_xattr_set(struct inode *inode,
3536                     int name_index,
3537                     const char *name,
3538                     const void *value,
3539                     size_t value_len,
3540                     int flags)
3541 {
3542         struct buffer_head *di_bh = NULL;
3543         struct ocfs2_dinode *di;
3544         int ret, credits, had_lock, ref_meta = 0, ref_credits = 0;
3545         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3546         struct inode *tl_inode = osb->osb_tl_inode;
3547         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3548         struct ocfs2_refcount_tree *ref_tree = NULL;
3549         struct ocfs2_lock_holder oh;
3550
3551         struct ocfs2_xattr_info xi = {
3552                 .xi_name_index = name_index,
3553                 .xi_name = name,
3554                 .xi_name_len = strlen(name),
3555                 .xi_value = value,
3556                 .xi_value_len = value_len,
3557         };
3558
3559         struct ocfs2_xattr_search xis = {
3560                 .not_found = -ENODATA,
3561         };
3562
3563         struct ocfs2_xattr_search xbs = {
3564                 .not_found = -ENODATA,
3565         };
3566
3567         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3568                 return -EOPNOTSUPP;
3569
3570         /*
3571          * Only xbs will be used on indexed trees.  xis doesn't need a
3572          * bucket.
3573          */
3574         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3575         if (!xbs.bucket) {
3576                 mlog_errno(-ENOMEM);
3577                 return -ENOMEM;
3578         }
3579
3580         had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 1, &oh);
3581         if (had_lock < 0) {
3582                 ret = had_lock;
3583                 mlog_errno(ret);
3584                 goto cleanup_nolock;
3585         }
3586         xis.inode_bh = xbs.inode_bh = di_bh;
3587         di = (struct ocfs2_dinode *)di_bh->b_data;
3588
3589         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3590         /*
3591          * Scan inode and external block to find the same name
3592          * extended attribute and collect search information.
3593          */
3594         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3595         if (ret)
3596                 goto cleanup;
3597         if (xis.not_found) {
3598                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3599                 if (ret)
3600                         goto cleanup;
3601         }
3602
3603         if (xis.not_found && xbs.not_found) {
3604                 ret = -ENODATA;
3605                 if (flags & XATTR_REPLACE)
3606                         goto cleanup;
3607                 ret = 0;
3608                 if (!value)
3609                         goto cleanup;
3610         } else {
3611                 ret = -EEXIST;
3612                 if (flags & XATTR_CREATE)
3613                         goto cleanup;
3614         }
3615
3616         /* Check whether the value is refcounted and do some preparation. */
3617         if (ocfs2_is_refcount_inode(inode) &&
3618             (!xis.not_found || !xbs.not_found)) {
3619                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3620                                                    &xis, &xbs, &ref_tree,
3621                                                    &ref_meta, &ref_credits);
3622                 if (ret) {
3623                         mlog_errno(ret);
3624                         goto cleanup;
3625                 }
3626         }
3627
3628         inode_lock(tl_inode);
3629
3630         if (ocfs2_truncate_log_needs_flush(osb)) {
3631                 ret = __ocfs2_flush_truncate_log(osb);
3632                 if (ret < 0) {
3633                         inode_unlock(tl_inode);
3634                         mlog_errno(ret);
3635                         goto cleanup;
3636                 }
3637         }
3638         inode_unlock(tl_inode);
3639
3640         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3641                                         &xbs, &ctxt, ref_meta, &credits);
3642         if (ret) {
3643                 mlog_errno(ret);
3644                 goto cleanup;
3645         }
3646
3647         /* we need to update inode's ctime field, so add credit for it. */
3648         credits += OCFS2_INODE_UPDATE_CREDITS;
3649         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3650         if (IS_ERR(ctxt.handle)) {
3651                 ret = PTR_ERR(ctxt.handle);
3652                 mlog_errno(ret);
3653                 goto out_free_ac;
3654         }
3655
3656         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3657         ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3658
3659         ocfs2_commit_trans(osb, ctxt.handle);
3660
3661 out_free_ac:
3662         if (ctxt.data_ac)
3663                 ocfs2_free_alloc_context(ctxt.data_ac);
3664         if (ctxt.meta_ac)
3665                 ocfs2_free_alloc_context(ctxt.meta_ac);
3666         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3667                 ocfs2_schedule_truncate_log_flush(osb, 1);
3668         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3669
3670 cleanup:
3671         if (ref_tree)
3672                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3673         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3674         if (!value && !ret) {
3675                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3676                 if (ret)
3677                         mlog_errno(ret);
3678         }
3679         ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
3680 cleanup_nolock:
3681         brelse(di_bh);
3682         brelse(xbs.xattr_bh);
3683         ocfs2_xattr_bucket_free(xbs.bucket);
3684
3685         return ret;
3686 }
3687
3688 /*
3689  * Find the xattr extent rec which may contains name_hash.
3690  * e_cpos will be the first name hash of the xattr rec.
3691  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3692  */
3693 static int ocfs2_xattr_get_rec(struct inode *inode,
3694                                u32 name_hash,
3695                                u64 *p_blkno,
3696                                u32 *e_cpos,
3697                                u32 *num_clusters,
3698                                struct ocfs2_extent_list *el)
3699 {
3700         int ret = 0, i;
3701         struct buffer_head *eb_bh = NULL;
3702         struct ocfs2_extent_block *eb;
3703         struct ocfs2_extent_rec *rec = NULL;
3704         u64 e_blkno = 0;
3705
3706         if (el->l_tree_depth) {
3707                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3708                                       &eb_bh);
3709                 if (ret) {
3710                         mlog_errno(ret);
3711                         goto out;
3712                 }
3713
3714                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3715                 el = &eb->h_list;
3716
3717                 if (el->l_tree_depth) {
3718                         ret = ocfs2_error(inode->i_sb,
3719                                           "Inode %lu has non zero tree depth in xattr tree block %llu\n",
3720                                           inode->i_ino,
3721                                           (unsigned long long)eb_bh->b_blocknr);
3722                         goto out;
3723                 }
3724         }
3725
3726         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3727                 rec = &el->l_recs[i];
3728
3729                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3730                         e_blkno = le64_to_cpu(rec->e_blkno);
3731                         break;
3732                 }
3733         }
3734
3735         if (!e_blkno) {
3736                 ret = ocfs2_error(inode->i_sb, "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
3737                                   inode->i_ino,
3738                                   le32_to_cpu(rec->e_cpos),
3739                                   ocfs2_rec_clusters(el, rec));
3740                 goto out;
3741         }
3742
3743         *p_blkno = le64_to_cpu(rec->e_blkno);
3744         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3745         if (e_cpos)
3746                 *e_cpos = le32_to_cpu(rec->e_cpos);
3747 out:
3748         brelse(eb_bh);
3749         return ret;
3750 }
3751
3752 typedef int (xattr_bucket_func)(struct inode *inode,
3753                                 struct ocfs2_xattr_bucket *bucket,
3754                                 void *para);
3755
3756 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3757                                    struct ocfs2_xattr_bucket *bucket,
3758                                    int name_index,
3759                                    const char *name,
3760                                    u32 name_hash,
3761                                    u16 *xe_index,
3762                                    int *found)
3763 {
3764         int i, ret = 0, cmp = 1, block_off, new_offset;
3765         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3766         size_t name_len = strlen(name);
3767         struct ocfs2_xattr_entry *xe = NULL;
3768         char *xe_name;
3769
3770         /*
3771          * We don't use binary search in the bucket because there
3772          * may be multiple entries with the same name hash.
3773          */
3774         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3775                 xe = &xh->xh_entries[i];
3776
3777                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3778                         continue;
3779                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3780                         break;
3781
3782                 cmp = name_index - ocfs2_xattr_get_type(xe);
3783                 if (!cmp)
3784                         cmp = name_len - xe->xe_name_len;
3785                 if (cmp)
3786                         continue;
3787
3788                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3789                                                         xh,
3790                                                         i,
3791                                                         &block_off,
3792                                                         &new_offset);
3793                 if (ret) {
3794                         mlog_errno(ret);
3795                         break;
3796                 }
3797
3798
3799                 xe_name = bucket_block(bucket, block_off) + new_offset;
3800                 if (!memcmp(name, xe_name, name_len)) {
3801                         *xe_index = i;
3802                         *found = 1;
3803                         ret = 0;
3804                         break;
3805                 }
3806         }
3807
3808         return ret;
3809 }
3810
3811 /*
3812  * Find the specified xattr entry in a series of buckets.
3813  * This series start from p_blkno and last for num_clusters.
3814  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3815  * the num of the valid buckets.
3816  *
3817  * Return the buffer_head this xattr should reside in. And if the xattr's
3818  * hash is in the gap of 2 buckets, return the lower bucket.
3819  */
3820 static int ocfs2_xattr_bucket_find(struct inode *inode,
3821                                    int name_index,
3822                                    const char *name,
3823                                    u32 name_hash,
3824                                    u64 p_blkno,
3825                                    u32 first_hash,
3826                                    u32 num_clusters,
3827                                    struct ocfs2_xattr_search *xs)
3828 {
3829         int ret, found = 0;
3830         struct ocfs2_xattr_header *xh = NULL;
3831         struct ocfs2_xattr_entry *xe = NULL;
3832         u16 index = 0;
3833         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3834         int low_bucket = 0, bucket, high_bucket;
3835         struct ocfs2_xattr_bucket *search;
3836         u32 last_hash;
3837         u64 blkno, lower_blkno = 0;
3838
3839         search = ocfs2_xattr_bucket_new(inode);
3840         if (!search) {
3841                 ret = -ENOMEM;
3842                 mlog_errno(ret);
3843                 goto out;
3844         }
3845
3846         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3847         if (ret) {
3848                 mlog_errno(ret);
3849                 goto out;
3850         }
3851
3852         xh = bucket_xh(search);
3853         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3854         while (low_bucket <= high_bucket) {
3855                 ocfs2_xattr_bucket_relse(search);
3856
3857                 bucket = (low_bucket + high_bucket) / 2;
3858                 blkno = p_blkno + bucket * blk_per_bucket;
3859                 ret = ocfs2_read_xattr_bucket(search, blkno);
3860                 if (ret) {
3861                         mlog_errno(ret);
3862                         goto out;
3863                 }
3864
3865                 xh = bucket_xh(search);
3866                 xe = &xh->xh_entries[0];
3867                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3868                         high_bucket = bucket - 1;
3869                         continue;
3870                 }
3871
3872                 /*
3873                  * Check whether the hash of the last entry in our
3874                  * bucket is larger than the search one. for an empty
3875                  * bucket, the last one is also the first one.
3876                  */
3877                 if (xh->xh_count)
3878                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3879
3880                 last_hash = le32_to_cpu(xe->xe_name_hash);
3881
3882                 /* record lower_blkno which may be the insert place. */
3883                 lower_blkno = blkno;
3884
3885                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3886                         low_bucket = bucket + 1;
3887                         continue;
3888                 }
3889
3890                 /* the searched xattr should reside in this bucket if exists. */
3891                 ret = ocfs2_find_xe_in_bucket(inode, search,
3892                                               name_index, name, name_hash,
3893                                               &index, &found);
3894                 if (ret) {
3895                         mlog_errno(ret);
3896                         goto out;
3897                 }
3898                 break;
3899         }
3900
3901         /*
3902          * Record the bucket we have found.
3903          * When the xattr's hash value is in the gap of 2 buckets, we will
3904          * always set it to the previous bucket.
3905          */
3906         if (!lower_blkno)
3907                 lower_blkno = p_blkno;
3908
3909         /* This should be in cache - we just read it during the search */
3910         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3911         if (ret) {
3912                 mlog_errno(ret);
3913                 goto out;
3914         }
3915
3916         xs->header = bucket_xh(xs->bucket);
3917         xs->base = bucket_block(xs->bucket, 0);
3918         xs->end = xs->base + inode->i_sb->s_blocksize;
3919
3920         if (found) {
3921                 xs->here = &xs->header->xh_entries[index];
3922                 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3923                         name, name_index, name_hash,
3924                         (unsigned long long)bucket_blkno(xs->bucket),
3925                         index);
3926         } else
3927                 ret = -ENODATA;
3928
3929 out:
3930         ocfs2_xattr_bucket_free(search);
3931         return ret;
3932 }
3933
3934 static int ocfs2_xattr_index_block_find(struct inode *inode,
3935                                         struct buffer_head *root_bh,
3936                                         int name_index,
3937                                         const char *name,
3938                                         struct ocfs2_xattr_search *xs)
3939 {
3940         int ret;
3941         struct ocfs2_xattr_block *xb =
3942                         (struct ocfs2_xattr_block *)root_bh->b_data;
3943         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3944         struct ocfs2_extent_list *el = &xb_root->xt_list;
3945         u64 p_blkno = 0;
3946         u32 first_hash, num_clusters = 0;
3947         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3948
3949         if (le16_to_cpu(el->l_next_free_rec) == 0)
3950                 return -ENODATA;
3951
3952         trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3953                                         name, name_index, name_hash,
3954                                         (unsigned long long)root_bh->b_blocknr,
3955                                         -1);
3956
3957         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3958                                   &num_clusters, el);
3959         if (ret) {
3960                 mlog_errno(ret);
3961                 goto out;
3962         }
3963
3964         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3965
3966         trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3967                                         name, name_index, first_hash,
3968                                         (unsigned long long)p_blkno,
3969                                         num_clusters);
3970
3971         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3972                                       p_blkno, first_hash, num_clusters, xs);
3973
3974 out:
3975         return ret;
3976 }
3977
3978 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3979                                        u64 blkno,
3980                                        u32 clusters,
3981                                        xattr_bucket_func *func,
3982                                        void *para)
3983 {
3984         int i, ret = 0;
3985         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3986         u32 num_buckets = clusters * bpc;
3987         struct ocfs2_xattr_bucket *bucket;
3988
3989         bucket = ocfs2_xattr_bucket_new(inode);
3990         if (!bucket) {
3991                 mlog_errno(-ENOMEM);
3992                 return -ENOMEM;
3993         }
3994
3995         trace_ocfs2_iterate_xattr_buckets(
3996                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3997                 (unsigned long long)blkno, clusters);
3998
3999         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
4000                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
4001                 if (ret) {
4002                         mlog_errno(ret);
4003                         break;
4004                 }
4005
4006                 /*
4007                  * The real bucket num in this series of blocks is stored
4008                  * in the 1st bucket.
4009                  */
4010                 if (i == 0)
4011                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
4012
4013                 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
4014                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
4015                 if (func) {
4016                         ret = func(inode, bucket, para);
4017                         if (ret && ret != -ERANGE)
4018                                 mlog_errno(ret);
4019                         /* Fall through to bucket_relse() */
4020                 }
4021
4022                 ocfs2_xattr_bucket_relse(bucket);
4023                 if (ret)
4024                         break;
4025         }
4026
4027         ocfs2_xattr_bucket_free(bucket);
4028         return ret;
4029 }
4030
4031 struct ocfs2_xattr_tree_list {
4032         char *buffer;
4033         size_t buffer_size;
4034         size_t result;
4035 };
4036
4037 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4038                                              struct ocfs2_xattr_header *xh,
4039                                              int index,
4040                                              int *block_off,
4041                                              int *new_offset)
4042 {
4043         u16 name_offset;
4044
4045         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4046                 return -EINVAL;
4047
4048         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4049
4050         *block_off = name_offset >> sb->s_blocksize_bits;
4051         *new_offset = name_offset % sb->s_blocksize;
4052
4053         return 0;
4054 }
4055
4056 static int ocfs2_list_xattr_bucket(struct inode *inode,
4057                                    struct ocfs2_xattr_bucket *bucket,
4058                                    void *para)
4059 {
4060         int ret = 0, type;
4061         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4062         int i, block_off, new_offset;
4063         const char *name;
4064
4065         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4066                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4067                 type = ocfs2_xattr_get_type(entry);
4068
4069                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4070                                                         bucket_xh(bucket),
4071                                                         i,
4072                                                         &block_off,
4073                                                         &new_offset);
4074                 if (ret)
4075                         break;
4076
4077                 name = (const char *)bucket_block(bucket, block_off) +
4078                         new_offset;
4079                 ret = ocfs2_xattr_list_entry(inode->i_sb,
4080                                              xl->buffer,
4081                                              xl->buffer_size,
4082                                              &xl->result,
4083                                              type, name,
4084                                              entry->xe_name_len);
4085                 if (ret)
4086                         break;
4087         }
4088
4089         return ret;
4090 }
4091
4092 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4093                                            struct buffer_head *blk_bh,
4094                                            xattr_tree_rec_func *rec_func,
4095                                            void *para)
4096 {
4097         struct ocfs2_xattr_block *xb =
4098                         (struct ocfs2_xattr_block *)blk_bh->b_data;
4099         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4100         int ret = 0;
4101         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4102         u64 p_blkno = 0;
4103
4104         if (!el->l_next_free_rec || !rec_func)
4105                 return 0;
4106
4107         while (name_hash > 0) {
4108                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4109                                           &e_cpos, &num_clusters, el);
4110                 if (ret) {
4111                         mlog_errno(ret);
4112                         break;
4113                 }
4114
4115                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4116                                num_clusters, para);
4117                 if (ret) {
4118                         if (ret != -ERANGE)
4119                                 mlog_errno(ret);
4120                         break;
4121                 }
4122
4123                 if (e_cpos == 0)
4124                         break;
4125
4126                 name_hash = e_cpos - 1;
4127         }
4128
4129         return ret;
4130
4131 }
4132
4133 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4134                                      struct buffer_head *root_bh,
4135                                      u64 blkno, u32 cpos, u32 len, void *para)
4136 {
4137         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4138                                            ocfs2_list_xattr_bucket, para);
4139 }
4140
4141 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4142                                              struct buffer_head *blk_bh,
4143                                              char *buffer,
4144                                              size_t buffer_size)
4145 {
4146         int ret;
4147         struct ocfs2_xattr_tree_list xl = {
4148                 .buffer = buffer,
4149                 .buffer_size = buffer_size,
4150                 .result = 0,
4151         };
4152
4153         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4154                                               ocfs2_list_xattr_tree_rec, &xl);
4155         if (ret) {
4156                 mlog_errno(ret);
4157                 goto out;
4158         }
4159
4160         ret = xl.result;
4161 out:
4162         return ret;
4163 }
4164
4165 static int cmp_xe(const void *a, const void *b)
4166 {
4167         const struct ocfs2_xattr_entry *l = a, *r = b;
4168         u32 l_hash = le32_to_cpu(l->xe_name_hash);
4169         u32 r_hash = le32_to_cpu(r->xe_name_hash);
4170
4171         if (l_hash > r_hash)
4172                 return 1;
4173         if (l_hash < r_hash)
4174                 return -1;
4175         return 0;
4176 }
4177
4178 static void swap_xe(void *a, void *b, int size)
4179 {
4180         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4181
4182         tmp = *l;
4183         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4184         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4185 }
4186
4187 /*
4188  * When the ocfs2_xattr_block is filled up, new bucket will be created
4189  * and all the xattr entries will be moved to the new bucket.
4190  * The header goes at the start of the bucket, and the names+values are
4191  * filled from the end.  This is why *target starts as the last buffer.
4192  * Note: we need to sort the entries since they are not saved in order
4193  * in the ocfs2_xattr_block.
4194  */
4195 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4196                                            struct buffer_head *xb_bh,
4197                                            struct ocfs2_xattr_bucket *bucket)
4198 {
4199         int i, blocksize = inode->i_sb->s_blocksize;
4200         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4201         u16 offset, size, off_change;
4202         struct ocfs2_xattr_entry *xe;
4203         struct ocfs2_xattr_block *xb =
4204                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
4205         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4206         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4207         u16 count = le16_to_cpu(xb_xh->xh_count);
4208         char *src = xb_bh->b_data;
4209         char *target = bucket_block(bucket, blks - 1);
4210
4211         trace_ocfs2_cp_xattr_block_to_bucket_begin(
4212                                 (unsigned long long)xb_bh->b_blocknr,
4213                                 (unsigned long long)bucket_blkno(bucket));
4214
4215         for (i = 0; i < blks; i++)
4216                 memset(bucket_block(bucket, i), 0, blocksize);
4217
4218         /*
4219          * Since the xe_name_offset is based on ocfs2_xattr_header,
4220          * there is a offset change corresponding to the change of
4221          * ocfs2_xattr_header's position.
4222          */
4223         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4224         xe = &xb_xh->xh_entries[count - 1];
4225         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4226         size = blocksize - offset;
4227
4228         /* copy all the names and values. */
4229         memcpy(target + offset, src + offset, size);
4230
4231         /* Init new header now. */
4232         xh->xh_count = xb_xh->xh_count;
4233         xh->xh_num_buckets = cpu_to_le16(1);
4234         xh->xh_name_value_len = cpu_to_le16(size);
4235         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4236
4237         /* copy all the entries. */
4238         target = bucket_block(bucket, 0);
4239         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4240         size = count * sizeof(struct ocfs2_xattr_entry);
4241         memcpy(target + offset, (char *)xb_xh + offset, size);
4242
4243         /* Change the xe offset for all the xe because of the move. */
4244         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4245                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4246         for (i = 0; i < count; i++)
4247                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4248
4249         trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4250
4251         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4252              cmp_xe, swap_xe);
4253 }
4254
4255 /*
4256  * After we move xattr from block to index btree, we have to
4257  * update ocfs2_xattr_search to the new xe and base.
4258  *
4259  * When the entry is in xattr block, xattr_bh indicates the storage place.
4260  * While if the entry is in index b-tree, "bucket" indicates the
4261  * real place of the xattr.
4262  */
4263 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4264                                             struct ocfs2_xattr_search *xs,
4265                                             struct buffer_head *old_bh)
4266 {
4267         char *buf = old_bh->b_data;
4268         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4269         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4270         int i;
4271
4272         xs->header = bucket_xh(xs->bucket);
4273         xs->base = bucket_block(xs->bucket, 0);
4274         xs->end = xs->base + inode->i_sb->s_blocksize;
4275
4276         if (xs->not_found)
4277                 return;
4278
4279         i = xs->here - old_xh->xh_entries;
4280         xs->here = &xs->header->xh_entries[i];
4281 }
4282
4283 static int ocfs2_xattr_create_index_block(struct inode *inode,
4284                                           struct ocfs2_xattr_search *xs,
4285                                           struct ocfs2_xattr_set_ctxt *ctxt)
4286 {
4287         int ret;
4288         u32 bit_off, len;
4289         u64 blkno;
4290         handle_t *handle = ctxt->handle;
4291         struct ocfs2_inode_info *oi = OCFS2_I(inode);
4292         struct buffer_head *xb_bh = xs->xattr_bh;
4293         struct ocfs2_xattr_block *xb =
4294                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4295         struct ocfs2_xattr_tree_root *xr;
4296         u16 xb_flags = le16_to_cpu(xb->xb_flags);
4297
4298         trace_ocfs2_xattr_create_index_block_begin(
4299                                 (unsigned long long)xb_bh->b_blocknr);
4300
4301         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4302         BUG_ON(!xs->bucket);
4303
4304         /*
4305          * XXX:
4306          * We can use this lock for now, and maybe move to a dedicated mutex
4307          * if performance becomes a problem later.
4308          */
4309         down_write(&oi->ip_alloc_sem);
4310
4311         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4312                                       OCFS2_JOURNAL_ACCESS_WRITE);
4313         if (ret) {
4314                 mlog_errno(ret);
4315                 goto out;
4316         }
4317
4318         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4319                                      1, 1, &bit_off, &len);
4320         if (ret) {
4321                 mlog_errno(ret);
4322                 goto out;
4323         }
4324
4325         /*
4326          * The bucket may spread in many blocks, and
4327          * we will only touch the 1st block and the last block
4328          * in the whole bucket(one for entry and one for data).
4329          */
4330         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4331
4332         trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4333
4334         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4335         if (ret) {
4336                 mlog_errno(ret);
4337                 goto out;
4338         }
4339
4340         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4341                                                 OCFS2_JOURNAL_ACCESS_CREATE);
4342         if (ret) {
4343                 mlog_errno(ret);
4344                 goto out;
4345         }
4346
4347         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4348         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4349
4350         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4351
4352         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4353         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4354                offsetof(struct ocfs2_xattr_block, xb_attrs));
4355
4356         xr = &xb->xb_attrs.xb_root;
4357         xr->xt_clusters = cpu_to_le32(1);
4358         xr->xt_last_eb_blk = 0;
4359         xr->xt_list.l_tree_depth = 0;
4360         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4361         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4362
4363         xr->xt_list.l_recs[0].e_cpos = 0;
4364         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4365         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4366
4367         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4368
4369         ocfs2_journal_dirty(handle, xb_bh);
4370
4371 out:
4372         up_write(&oi->ip_alloc_sem);
4373
4374         return ret;
4375 }
4376
4377 static int cmp_xe_offset(const void *a, const void *b)
4378 {
4379         const struct ocfs2_xattr_entry *l = a, *r = b;
4380         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4381         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4382
4383         if (l_name_offset < r_name_offset)
4384                 return 1;
4385         if (l_name_offset > r_name_offset)
4386                 return -1;
4387         return 0;
4388 }
4389
4390 /*
4391  * defrag a xattr bucket if we find that the bucket has some
4392  * holes beteen name/value pairs.
4393  * We will move all the name/value pairs to the end of the bucket
4394  * so that we can spare some space for insertion.
4395  */
4396 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4397                                      handle_t *handle,
4398                                      struct ocfs2_xattr_bucket *bucket)
4399 {
4400         int ret, i;
4401         size_t end, offset, len;
4402         struct ocfs2_xattr_header *xh;
4403         char *entries, *buf, *bucket_buf = NULL;
4404         u64 blkno = bucket_blkno(bucket);
4405         u16 xh_free_start;
4406         size_t blocksize = inode->i_sb->s_blocksize;
4407         struct ocfs2_xattr_entry *xe;
4408
4409         /*
4410          * In order to make the operation more efficient and generic,
4411          * we copy all the blocks into a contiguous memory and do the
4412          * defragment there, so if anything is error, we will not touch
4413          * the real block.
4414          */
4415         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4416         if (!bucket_buf) {
4417                 ret = -EIO;
4418                 goto out;
4419         }
4420
4421         buf = bucket_buf;
4422         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4423                 memcpy(buf, bucket_block(bucket, i), blocksize);
4424
4425         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4426                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4427         if (ret < 0) {
4428                 mlog_errno(ret);
4429                 goto out;
4430         }
4431
4432         xh = (struct ocfs2_xattr_header *)bucket_buf;
4433         entries = (char *)xh->xh_entries;
4434         xh_free_start = le16_to_cpu(xh->xh_free_start);
4435
4436         trace_ocfs2_defrag_xattr_bucket(
4437              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4438              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4439
4440         /*
4441          * sort all the entries by their offset.
4442          * the largest will be the first, so that we can
4443          * move them to the end one by one.
4444          */
4445         sort(entries, le16_to_cpu(xh->xh_count),
4446              sizeof(struct ocfs2_xattr_entry),
4447              cmp_xe_offset, swap_xe);
4448
4449         /* Move all name/values to the end of the bucket. */
4450         xe = xh->xh_entries;
4451         end = OCFS2_XATTR_BUCKET_SIZE;
4452         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4453                 offset = le16_to_cpu(xe->xe_name_offset);
4454                 len = namevalue_size_xe(xe);
4455
4456                 /*
4457                  * We must make sure that the name/value pair
4458                  * exist in the same block. So adjust end to
4459                  * the previous block end if needed.
4460                  */
4461                 if (((end - len) / blocksize !=
4462                         (end - 1) / blocksize))
4463                         end = end - end % blocksize;
4464
4465                 if (end > offset + len) {
4466                         memmove(bucket_buf + end - len,
4467                                 bucket_buf + offset, len);
4468                         xe->xe_name_offset = cpu_to_le16(end - len);
4469                 }
4470
4471                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4472                                 "bucket %llu\n", (unsigned long long)blkno);
4473
4474                 end -= len;
4475         }
4476
4477         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4478                         "bucket %llu\n", (unsigned long long)blkno);
4479
4480         if (xh_free_start == end)
4481                 goto out;
4482
4483         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4484         xh->xh_free_start = cpu_to_le16(end);
4485
4486         /* sort the entries by their name_hash. */
4487         sort(entries, le16_to_cpu(xh->xh_count),
4488              sizeof(struct ocfs2_xattr_entry),
4489              cmp_xe, swap_xe);
4490
4491         buf = bucket_buf;
4492         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4493                 memcpy(bucket_block(bucket, i), buf, blocksize);
4494         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4495
4496 out:
4497         kfree(bucket_buf);
4498         return ret;
4499 }
4500
4501 /*
4502  * prev_blkno points to the start of an existing extent.  new_blkno
4503  * points to a newly allocated extent.  Because we know each of our
4504  * clusters contains more than bucket, we can easily split one cluster
4505  * at a bucket boundary.  So we take the last cluster of the existing
4506  * extent and split it down the middle.  We move the last half of the
4507  * buckets in the last cluster of the existing extent over to the new
4508  * extent.
4509  *
4510  * first_bh is the buffer at prev_blkno so we can update the existing
4511  * extent's bucket count.  header_bh is the bucket were we were hoping
4512  * to insert our xattr.  If the bucket move places the target in the new
4513  * extent, we'll update first_bh and header_bh after modifying the old
4514  * extent.
4515  *
4516  * first_hash will be set as the 1st xe's name_hash in the new extent.
4517  */
4518 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4519                                                handle_t *handle,
4520                                                struct ocfs2_xattr_bucket *first,
4521                                                struct ocfs2_xattr_bucket *target,
4522                                                u64 new_blkno,
4523                                                u32 num_clusters,
4524                                                u32 *first_hash)
4525 {
4526         int ret;
4527         struct super_block *sb = inode->i_sb;
4528         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4529         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4530         int to_move = num_buckets / 2;
4531         u64 src_blkno;
4532         u64 last_cluster_blkno = bucket_blkno(first) +
4533                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4534
4535         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4536         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4537
4538         trace_ocfs2_mv_xattr_bucket_cross_cluster(
4539                                 (unsigned long long)last_cluster_blkno,
4540                                 (unsigned long long)new_blkno);
4541
4542         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4543                                      last_cluster_blkno, new_blkno,
4544                                      to_move, first_hash);
4545         if (ret) {
4546                 mlog_errno(ret);
4547                 goto out;
4548         }
4549
4550         /* This is the first bucket that got moved */
4551         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4552
4553         /*
4554          * If the target bucket was part of the moved buckets, we need to
4555          * update first and target.
4556          */
4557         if (bucket_blkno(target) >= src_blkno) {
4558                 /* Find the block for the new target bucket */
4559                 src_blkno = new_blkno +
4560                         (bucket_blkno(target) - src_blkno);
4561
4562                 ocfs2_xattr_bucket_relse(first);
4563                 ocfs2_xattr_bucket_relse(target);
4564
4565                 /*
4566                  * These shouldn't fail - the buffers are in the
4567                  * journal from ocfs2_cp_xattr_bucket().
4568                  */
4569                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4570                 if (ret) {
4571                         mlog_errno(ret);
4572                         goto out;
4573                 }
4574                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4575                 if (ret)
4576                         mlog_errno(ret);
4577
4578         }
4579
4580 out:
4581         return ret;
4582 }
4583
4584 /*
4585  * Find the suitable pos when we divide a bucket into 2.
4586  * We have to make sure the xattrs with the same hash value exist
4587  * in the same bucket.
4588  *
4589  * If this ocfs2_xattr_header covers more than one hash value, find a
4590  * place where the hash value changes.  Try to find the most even split.
4591  * The most common case is that all entries have different hash values,
4592  * and the first check we make will find a place to split.
4593  */
4594 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4595 {
4596         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4597         int count = le16_to_cpu(xh->xh_count);
4598         int delta, middle = count / 2;
4599
4600         /*
4601          * We start at the middle.  Each step gets farther away in both
4602          * directions.  We therefore hit the change in hash value
4603          * nearest to the middle.  Note that this loop does not execute for
4604          * count < 2.
4605          */
4606         for (delta = 0; delta < middle; delta++) {
4607                 /* Let's check delta earlier than middle */
4608                 if (cmp_xe(&entries[middle - delta - 1],
4609                            &entries[middle - delta]))
4610                         return middle - delta;
4611
4612                 /* For even counts, don't walk off the end */
4613                 if ((middle + delta + 1) == count)
4614                         continue;
4615
4616                 /* Now try delta past middle */
4617                 if (cmp_xe(&entries[middle + delta],
4618                            &entries[middle + delta + 1]))
4619                         return middle + delta + 1;
4620         }
4621
4622         /* Every entry had the same hash */
4623         return count;
4624 }
4625
4626 /*
4627  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4628  * first_hash will record the 1st hash of the new bucket.
4629  *
4630  * Normally half of the xattrs will be moved.  But we have to make
4631  * sure that the xattrs with the same hash value are stored in the
4632  * same bucket. If all the xattrs in this bucket have the same hash
4633  * value, the new bucket will be initialized as an empty one and the
4634  * first_hash will be initialized as (hash_value+1).
4635  */
4636 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4637                                     handle_t *handle,
4638                                     u64 blk,
4639                                     u64 new_blk,
4640                                     u32 *first_hash,
4641                                     int new_bucket_head)
4642 {
4643         int ret, i;
4644         int count, start, len, name_value_len = 0, name_offset = 0;
4645         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4646         struct ocfs2_xattr_header *xh;
4647         struct ocfs2_xattr_entry *xe;
4648         int blocksize = inode->i_sb->s_blocksize;
4649
4650         trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4651                                               (unsigned long long)new_blk);
4652
4653         s_bucket = ocfs2_xattr_bucket_new(inode);
4654         t_bucket = ocfs2_xattr_bucket_new(inode);
4655         if (!s_bucket || !t_bucket) {
4656                 ret = -ENOMEM;
4657                 mlog_errno(ret);
4658                 goto out;
4659         }
4660
4661         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4662         if (ret) {
4663                 mlog_errno(ret);
4664                 goto out;
4665         }
4666
4667         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4668                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4669         if (ret) {
4670                 mlog_errno(ret);
4671                 goto out;
4672         }
4673
4674         /*
4675          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4676          * there's no need to read it.
4677          */
4678         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4679         if (ret) {
4680                 mlog_errno(ret);
4681                 goto out;
4682         }
4683
4684         /*
4685          * Hey, if we're overwriting t_bucket, what difference does
4686          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4687          * same part of ocfs2_cp_xattr_bucket().
4688          */
4689         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4690                                                 new_bucket_head ?
4691                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4692                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4693         if (ret) {
4694                 mlog_errno(ret);
4695                 goto out;
4696         }
4697
4698         xh = bucket_xh(s_bucket);
4699         count = le16_to_cpu(xh->xh_count);
4700         start = ocfs2_xattr_find_divide_pos(xh);
4701
4702         if (start == count) {
4703                 xe = &xh->xh_entries[start-1];
4704
4705                 /*
4706                  * initialized a new empty bucket here.
4707                  * The hash value is set as one larger than
4708                  * that of the last entry in the previous bucket.
4709                  */
4710                 for (i = 0; i < t_bucket->bu_blocks; i++)
4711                         memset(bucket_block(t_bucket, i), 0, blocksize);
4712
4713                 xh = bucket_xh(t_bucket);
4714                 xh->xh_free_start = cpu_to_le16(blocksize);
4715                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4716                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4717
4718                 goto set_num_buckets;
4719         }
4720
4721         /* copy the whole bucket to the new first. */
4722         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4723
4724         /* update the new bucket. */
4725         xh = bucket_xh(t_bucket);
4726
4727         /*
4728          * Calculate the total name/value len and xh_free_start for
4729          * the old bucket first.
4730          */
4731         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4732         name_value_len = 0;
4733         for (i = 0; i < start; i++) {
4734                 xe = &xh->xh_entries[i];
4735                 name_value_len += namevalue_size_xe(xe);
4736                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4737                         name_offset = le16_to_cpu(xe->xe_name_offset);
4738         }
4739
4740         /*
4741          * Now begin the modification to the new bucket.
4742          *
4743          * In the new bucket, We just move the xattr entry to the beginning
4744          * and don't touch the name/value. So there will be some holes in the
4745          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4746          * called.
4747          */
4748         xe = &xh->xh_entries[start];
4749         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4750         trace_ocfs2_divide_xattr_bucket_move(len,
4751                         (int)((char *)xe - (char *)xh),
4752                         (int)((char *)xh->xh_entries - (char *)xh));
4753         memmove((char *)xh->xh_entries, (char *)xe, len);
4754         xe = &xh->xh_entries[count - start];
4755         len = sizeof(struct ocfs2_xattr_entry) * start;
4756         memset((char *)xe, 0, len);
4757
4758         le16_add_cpu(&xh->xh_count, -start);
4759         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4760
4761         /* Calculate xh_free_start for the new bucket. */
4762         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4763         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4764                 xe = &xh->xh_entries[i];
4765                 if (le16_to_cpu(xe->xe_name_offset) <
4766                     le16_to_cpu(xh->xh_free_start))
4767                         xh->xh_free_start = xe->xe_name_offset;
4768         }
4769
4770 set_num_buckets:
4771         /* set xh->xh_num_buckets for the new xh. */
4772         if (new_bucket_head)
4773                 xh->xh_num_buckets = cpu_to_le16(1);
4774         else
4775                 xh->xh_num_buckets = 0;
4776
4777         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4778
4779         /* store the first_hash of the new bucket. */
4780         if (first_hash)
4781                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4782
4783         /*
4784          * Now only update the 1st block of the old bucket.  If we
4785          * just added a new empty bucket, there is no need to modify
4786          * it.
4787          */
4788         if (start == count)
4789                 goto out;
4790
4791         xh = bucket_xh(s_bucket);
4792         memset(&xh->xh_entries[start], 0,
4793                sizeof(struct ocfs2_xattr_entry) * (count - start));
4794         xh->xh_count = cpu_to_le16(start);
4795         xh->xh_free_start = cpu_to_le16(name_offset);
4796         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4797
4798         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4799
4800 out:
4801         ocfs2_xattr_bucket_free(s_bucket);
4802         ocfs2_xattr_bucket_free(t_bucket);
4803
4804         return ret;
4805 }
4806
4807 /*
4808  * Copy xattr from one bucket to another bucket.
4809  *
4810  * The caller must make sure that the journal transaction
4811  * has enough space for journaling.
4812  */
4813 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4814                                  handle_t *handle,
4815                                  u64 s_blkno,
4816                                  u64 t_blkno,
4817                                  int t_is_new)
4818 {
4819         int ret;
4820         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4821
4822         BUG_ON(s_blkno == t_blkno);
4823
4824         trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4825                                     (unsigned long long)t_blkno,
4826                                     t_is_new);
4827
4828         s_bucket = ocfs2_xattr_bucket_new(inode);
4829         t_bucket = ocfs2_xattr_bucket_new(inode);
4830         if (!s_bucket || !t_bucket) {
4831                 ret = -ENOMEM;
4832                 mlog_errno(ret);
4833                 goto out;
4834         }
4835
4836         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4837         if (ret)
4838                 goto out;
4839
4840         /*
4841          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4842          * there's no need to read it.
4843          */
4844         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4845         if (ret)
4846                 goto out;
4847
4848         /*
4849          * Hey, if we're overwriting t_bucket, what difference does
4850          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4851          * cluster to fill, we came here from
4852          * ocfs2_mv_xattr_buckets(), and it is really new -
4853          * ACCESS_CREATE is required.  But we also might have moved data
4854          * out of t_bucket before extending back into it.
4855          * ocfs2_add_new_xattr_bucket() can do this - its call to
4856          * ocfs2_add_new_xattr_cluster() may have created a new extent
4857          * and copied out the end of the old extent.  Then it re-extends
4858          * the old extent back to create space for new xattrs.  That's
4859          * how we get here, and the bucket isn't really new.
4860          */
4861         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4862                                                 t_is_new ?
4863                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4864                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4865         if (ret)
4866                 goto out;
4867
4868         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4869         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4870
4871 out:
4872         ocfs2_xattr_bucket_free(t_bucket);
4873         ocfs2_xattr_bucket_free(s_bucket);
4874
4875         return ret;
4876 }
4877
4878 /*
4879  * src_blk points to the start of an existing extent.  last_blk points to
4880  * last cluster in that extent.  to_blk points to a newly allocated
4881  * extent.  We copy the buckets from the cluster at last_blk to the new
4882  * extent.  If start_bucket is non-zero, we skip that many buckets before
4883  * we start copying.  The new extent's xh_num_buckets gets set to the
4884  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4885  * by the same amount.
4886  */
4887 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4888                                   u64 src_blk, u64 last_blk, u64 to_blk,
4889                                   unsigned int start_bucket,
4890                                   u32 *first_hash)
4891 {
4892         int i, ret, credits;
4893         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4894         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4895         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4896         struct ocfs2_xattr_bucket *old_first, *new_first;
4897
4898         trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4899                                      (unsigned long long)to_blk);
4900
4901         BUG_ON(start_bucket >= num_buckets);
4902         if (start_bucket) {
4903                 num_buckets -= start_bucket;
4904                 last_blk += (start_bucket * blks_per_bucket);
4905         }
4906
4907         /* The first bucket of the original extent */
4908         old_first = ocfs2_xattr_bucket_new(inode);
4909         /* The first bucket of the new extent */
4910         new_first = ocfs2_xattr_bucket_new(inode);
4911         if (!old_first || !new_first) {
4912                 ret = -ENOMEM;
4913                 mlog_errno(ret);
4914                 goto out;
4915         }
4916
4917         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4918         if (ret) {
4919                 mlog_errno(ret);
4920                 goto out;
4921         }
4922
4923         /*
4924          * We need to update the first bucket of the old extent and all
4925          * the buckets going to the new extent.
4926          */
4927         credits = ((num_buckets + 1) * blks_per_bucket);
4928         ret = ocfs2_extend_trans(handle, credits);
4929         if (ret) {
4930                 mlog_errno(ret);
4931                 goto out;
4932         }
4933
4934         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4935                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4936         if (ret) {
4937                 mlog_errno(ret);
4938                 goto out;
4939         }
4940
4941         for (i = 0; i < num_buckets; i++) {
4942                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4943                                             last_blk + (i * blks_per_bucket),
4944                                             to_blk + (i * blks_per_bucket),
4945                                             1);
4946                 if (ret) {
4947                         mlog_errno(ret);
4948                         goto out;
4949                 }
4950         }
4951
4952         /*
4953          * Get the new bucket ready before we dirty anything
4954          * (This actually shouldn't fail, because we already dirtied
4955          * it once in ocfs2_cp_xattr_bucket()).
4956          */
4957         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4958         if (ret) {
4959                 mlog_errno(ret);
4960                 goto out;
4961         }
4962         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4963                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4964         if (ret) {
4965                 mlog_errno(ret);
4966                 goto out;
4967         }
4968
4969         /* Now update the headers */
4970         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4971         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4972
4973         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4974         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4975
4976         if (first_hash)
4977                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4978
4979 out:
4980         ocfs2_xattr_bucket_free(new_first);
4981         ocfs2_xattr_bucket_free(old_first);
4982         return ret;
4983 }
4984
4985 /*
4986  * Move some xattrs in this cluster to the new cluster.
4987  * This function should only be called when bucket size == cluster size.
4988  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4989  */
4990 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4991                                       handle_t *handle,
4992                                       u64 prev_blk,
4993                                       u64 new_blk,
4994                                       u32 *first_hash)
4995 {
4996         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4997         int ret, credits = 2 * blk_per_bucket;
4998
4999         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
5000
5001         ret = ocfs2_extend_trans(handle, credits);
5002         if (ret) {
5003                 mlog_errno(ret);
5004                 return ret;
5005         }
5006
5007         /* Move half of the xattr in start_blk to the next bucket. */
5008         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
5009                                           new_blk, first_hash, 1);
5010 }
5011
5012 /*
5013  * Move some xattrs from the old cluster to the new one since they are not
5014  * contiguous in ocfs2 xattr tree.
5015  *
5016  * new_blk starts a new separate cluster, and we will move some xattrs from
5017  * prev_blk to it. v_start will be set as the first name hash value in this
5018  * new cluster so that it can be used as e_cpos during tree insertion and
5019  * don't collide with our original b-tree operations. first_bh and header_bh
5020  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5021  * to extend the insert bucket.
5022  *
5023  * The problem is how much xattr should we move to the new one and when should
5024  * we update first_bh and header_bh?
5025  * 1. If cluster size > bucket size, that means the previous cluster has more
5026  *    than 1 bucket, so just move half nums of bucket into the new cluster and
5027  *    update the first_bh and header_bh if the insert bucket has been moved
5028  *    to the new cluster.
5029  * 2. If cluster_size == bucket_size:
5030  *    a) If the previous extent rec has more than one cluster and the insert
5031  *       place isn't in the last cluster, copy the entire last cluster to the
5032  *       new one. This time, we don't need to upate the first_bh and header_bh
5033  *       since they will not be moved into the new cluster.
5034  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
5035  *       the new one. And we set the extend flag to zero if the insert place is
5036  *       moved into the new allocated cluster since no extend is needed.
5037  */
5038 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5039                                             handle_t *handle,
5040                                             struct ocfs2_xattr_bucket *first,
5041                                             struct ocfs2_xattr_bucket *target,
5042                                             u64 new_blk,
5043                                             u32 prev_clusters,
5044                                             u32 *v_start,
5045                                             int *extend)
5046 {
5047         int ret;
5048
5049         trace_ocfs2_adjust_xattr_cross_cluster(
5050                         (unsigned long long)bucket_blkno(first),
5051                         (unsigned long long)new_blk, prev_clusters);
5052
5053         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5054                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5055                                                           handle,
5056                                                           first, target,
5057                                                           new_blk,
5058                                                           prev_clusters,
5059                                                           v_start);
5060                 if (ret)
5061                         mlog_errno(ret);
5062         } else {
5063                 /* The start of the last cluster in the first extent */
5064                 u64 last_blk = bucket_blkno(first) +
5065                         ((prev_clusters - 1) *
5066                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
5067
5068                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5069                         ret = ocfs2_mv_xattr_buckets(inode, handle,
5070                                                      bucket_blkno(first),
5071                                                      last_blk, new_blk, 0,
5072                                                      v_start);
5073                         if (ret)
5074                                 mlog_errno(ret);
5075                 } else {
5076                         ret = ocfs2_divide_xattr_cluster(inode, handle,
5077                                                          last_blk, new_blk,
5078                                                          v_start);
5079                         if (ret)
5080                                 mlog_errno(ret);
5081
5082                         if ((bucket_blkno(target) == last_blk) && extend)
5083                                 *extend = 0;
5084                 }
5085         }
5086
5087         return ret;
5088 }
5089
5090 /*
5091  * Add a new cluster for xattr storage.
5092  *
5093  * If the new cluster is contiguous with the previous one, it will be
5094  * appended to the same extent record, and num_clusters will be updated.
5095  * If not, we will insert a new extent for it and move some xattrs in
5096  * the last cluster into the new allocated one.
5097  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5098  * lose the benefits of hashing because we'll have to search large leaves.
5099  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5100  * if it's bigger).
5101  *
5102  * first_bh is the first block of the previous extent rec and header_bh
5103  * indicates the bucket we will insert the new xattrs. They will be updated
5104  * when the header_bh is moved into the new cluster.
5105  */
5106 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5107                                        struct buffer_head *root_bh,
5108                                        struct ocfs2_xattr_bucket *first,
5109                                        struct ocfs2_xattr_bucket *target,
5110                                        u32 *num_clusters,
5111                                        u32 prev_cpos,
5112                                        int *extend,
5113                                        struct ocfs2_xattr_set_ctxt *ctxt)
5114 {
5115         int ret;
5116         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5117         u32 prev_clusters = *num_clusters;
5118         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5119         u64 block;
5120         handle_t *handle = ctxt->handle;
5121         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5122         struct ocfs2_extent_tree et;
5123
5124         trace_ocfs2_add_new_xattr_cluster_begin(
5125                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5126                 (unsigned long long)bucket_blkno(first),
5127                 prev_cpos, prev_clusters);
5128
5129         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5130
5131         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5132                                       OCFS2_JOURNAL_ACCESS_WRITE);
5133         if (ret < 0) {
5134                 mlog_errno(ret);
5135                 goto leave;
5136         }
5137
5138         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5139                                      clusters_to_add, &bit_off, &num_bits);
5140         if (ret < 0) {
5141                 if (ret != -ENOSPC)
5142                         mlog_errno(ret);
5143                 goto leave;
5144         }
5145
5146         BUG_ON(num_bits > clusters_to_add);
5147
5148         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5149         trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5150
5151         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5152             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5153              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5154                 /*
5155                  * If this cluster is contiguous with the old one and
5156                  * adding this new cluster, we don't surpass the limit of
5157                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5158                  * initialized and used like other buckets in the previous
5159                  * cluster.
5160                  * So add it as a contiguous one. The caller will handle
5161                  * its init process.
5162                  */
5163                 v_start = prev_cpos + prev_clusters;
5164                 *num_clusters = prev_clusters + num_bits;
5165         } else {
5166                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5167                                                        handle,
5168                                                        first,
5169                                                        target,
5170                                                        block,
5171                                                        prev_clusters,
5172                                                        &v_start,
5173                                                        extend);
5174                 if (ret) {
5175                         mlog_errno(ret);
5176                         goto leave;
5177                 }
5178         }
5179
5180         trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5181                                                  v_start, num_bits);
5182         ret = ocfs2_insert_extent(handle, &et, v_start, block,
5183                                   num_bits, 0, ctxt->meta_ac);
5184         if (ret < 0) {
5185                 mlog_errno(ret);
5186                 goto leave;
5187         }
5188
5189         ocfs2_journal_dirty(handle, root_bh);
5190
5191 leave:
5192         return ret;
5193 }
5194
5195 /*
5196  * We are given an extent.  'first' is the bucket at the very front of
5197  * the extent.  The extent has space for an additional bucket past
5198  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5199  * of the target bucket.  We wish to shift every bucket past the target
5200  * down one, filling in that additional space.  When we get back to the
5201  * target, we split the target between itself and the now-empty bucket
5202  * at target+1 (aka, target_blkno + blks_per_bucket).
5203  */
5204 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5205                                      handle_t *handle,
5206                                      struct ocfs2_xattr_bucket *first,
5207                                      u64 target_blk,
5208                                      u32 num_clusters)
5209 {
5210         int ret, credits;
5211         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5212         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5213         u64 end_blk;
5214         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5215
5216         trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5217                                         (unsigned long long)bucket_blkno(first),
5218                                         num_clusters, new_bucket);
5219
5220         /* The extent must have room for an additional bucket */
5221         BUG_ON(new_bucket >=
5222                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5223
5224         /* end_blk points to the last existing bucket */
5225         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5226
5227         /*
5228          * end_blk is the start of the last existing bucket.
5229          * Thus, (end_blk - target_blk) covers the target bucket and
5230          * every bucket after it up to, but not including, the last
5231          * existing bucket.  Then we add the last existing bucket, the
5232          * new bucket, and the first bucket (3 * blk_per_bucket).
5233          */
5234         credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5235         ret = ocfs2_extend_trans(handle, credits);
5236         if (ret) {
5237                 mlog_errno(ret);
5238                 goto out;
5239         }
5240
5241         ret = ocfs2_xattr_bucket_journal_access(handle, first,
5242                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5243         if (ret) {
5244                 mlog_errno(ret);
5245                 goto out;
5246         }
5247
5248         while (end_blk != target_blk) {
5249                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5250                                             end_blk + blk_per_bucket, 0);
5251                 if (ret)
5252                         goto out;
5253                 end_blk -= blk_per_bucket;
5254         }
5255
5256         /* Move half of the xattr in target_blkno to the next bucket. */
5257         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5258                                         target_blk + blk_per_bucket, NULL, 0);
5259
5260         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5261         ocfs2_xattr_bucket_journal_dirty(handle, first);
5262
5263 out:
5264         return ret;
5265 }
5266
5267 /*
5268  * Add new xattr bucket in an extent record and adjust the buckets
5269  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5270  * bucket we want to insert into.
5271  *
5272  * In the easy case, we will move all the buckets after target down by
5273  * one. Half of target's xattrs will be moved to the next bucket.
5274  *
5275  * If current cluster is full, we'll allocate a new one.  This may not
5276  * be contiguous.  The underlying calls will make sure that there is
5277  * space for the insert, shifting buckets around if necessary.
5278  * 'target' may be moved by those calls.
5279  */
5280 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5281                                       struct buffer_head *xb_bh,
5282                                       struct ocfs2_xattr_bucket *target,
5283                                       struct ocfs2_xattr_set_ctxt *ctxt)
5284 {
5285         struct ocfs2_xattr_block *xb =
5286                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5287         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5288         struct ocfs2_extent_list *el = &xb_root->xt_list;
5289         u32 name_hash =
5290                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5291         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5292         int ret, num_buckets, extend = 1;
5293         u64 p_blkno;
5294         u32 e_cpos, num_clusters;
5295         /* The bucket at the front of the extent */
5296         struct ocfs2_xattr_bucket *first;
5297
5298         trace_ocfs2_add_new_xattr_bucket(
5299                                 (unsigned long long)bucket_blkno(target));
5300
5301         /* The first bucket of the original extent */
5302         first = ocfs2_xattr_bucket_new(inode);
5303         if (!first) {
5304                 ret = -ENOMEM;
5305                 mlog_errno(ret);
5306                 goto out;
5307         }
5308
5309         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5310                                   &num_clusters, el);
5311         if (ret) {
5312                 mlog_errno(ret);
5313                 goto out;
5314         }
5315
5316         ret = ocfs2_read_xattr_bucket(first, p_blkno);
5317         if (ret) {
5318                 mlog_errno(ret);
5319                 goto out;
5320         }
5321
5322         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5323         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5324                 /*
5325                  * This can move first+target if the target bucket moves
5326                  * to the new extent.
5327                  */
5328                 ret = ocfs2_add_new_xattr_cluster(inode,
5329                                                   xb_bh,
5330                                                   first,
5331                                                   target,
5332                                                   &num_clusters,
5333                                                   e_cpos,
5334                                                   &extend,
5335                                                   ctxt);
5336                 if (ret) {
5337                         mlog_errno(ret);
5338                         goto out;
5339                 }
5340         }
5341
5342         if (extend) {
5343                 ret = ocfs2_extend_xattr_bucket(inode,
5344                                                 ctxt->handle,
5345                                                 first,
5346                                                 bucket_blkno(target),
5347                                                 num_clusters);
5348                 if (ret)
5349                         mlog_errno(ret);
5350         }
5351
5352 out:
5353         ocfs2_xattr_bucket_free(first);
5354
5355         return ret;
5356 }
5357
5358 /*
5359  * Truncate the specified xe_off entry in xattr bucket.
5360  * bucket is indicated by header_bh and len is the new length.
5361  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5362  *
5363  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5364  */
5365 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5366                                              struct ocfs2_xattr_bucket *bucket,
5367                                              int xe_off,
5368                                              int len,
5369                                              struct ocfs2_xattr_set_ctxt *ctxt)
5370 {
5371         int ret, offset;
5372         u64 value_blk;
5373         struct ocfs2_xattr_entry *xe;
5374         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5375         size_t blocksize = inode->i_sb->s_blocksize;
5376         struct ocfs2_xattr_value_buf vb = {
5377                 .vb_access = ocfs2_journal_access,
5378         };
5379
5380         xe = &xh->xh_entries[xe_off];
5381
5382         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5383
5384         offset = le16_to_cpu(xe->xe_name_offset) +
5385                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5386
5387         value_blk = offset / blocksize;
5388
5389         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5390         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5391
5392         vb.vb_bh = bucket->bu_bhs[value_blk];
5393         BUG_ON(!vb.vb_bh);
5394
5395         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5396                 (vb.vb_bh->b_data + offset % blocksize);
5397
5398         /*
5399          * From here on out we have to dirty the bucket.  The generic
5400          * value calls only modify one of the bucket's bhs, but we need
5401          * to send the bucket at once.  So if they error, they *could* have
5402          * modified something.  We have to assume they did, and dirty
5403          * the whole bucket.  This leaves us in a consistent state.
5404          */
5405         trace_ocfs2_xattr_bucket_value_truncate(
5406                         (unsigned long long)bucket_blkno(bucket), xe_off, len);
5407         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5408         if (ret) {
5409                 mlog_errno(ret);
5410                 goto out;
5411         }
5412
5413         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5414                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5415         if (ret) {
5416                 mlog_errno(ret);
5417                 goto out;
5418         }
5419
5420         xe->xe_value_size = cpu_to_le64(len);
5421
5422         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5423
5424 out:
5425         return ret;
5426 }
5427
5428 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5429                                   struct buffer_head *root_bh,
5430                                   u64 blkno,
5431                                   u32 cpos,
5432                                   u32 len,
5433                                   void *para)
5434 {
5435         int ret;
5436         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5437         struct inode *tl_inode = osb->osb_tl_inode;
5438         handle_t *handle;
5439         struct ocfs2_xattr_block *xb =
5440                         (struct ocfs2_xattr_block *)root_bh->b_data;
5441         struct ocfs2_alloc_context *meta_ac = NULL;
5442         struct ocfs2_cached_dealloc_ctxt dealloc;
5443         struct ocfs2_extent_tree et;
5444
5445         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5446                                           ocfs2_delete_xattr_in_bucket, para);
5447         if (ret) {
5448                 mlog_errno(ret);
5449                 return ret;
5450         }
5451
5452         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5453
5454         ocfs2_init_dealloc_ctxt(&dealloc);
5455
5456         trace_ocfs2_rm_xattr_cluster(
5457                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
5458                         (unsigned long long)blkno, cpos, len);
5459
5460         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5461                                                len);
5462
5463         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5464         if (ret) {
5465                 mlog_errno(ret);
5466                 return ret;
5467         }
5468
5469         inode_lock(tl_inode);
5470
5471         if (ocfs2_truncate_log_needs_flush(osb)) {
5472                 ret = __ocfs2_flush_truncate_log(osb);
5473                 if (ret < 0) {
5474                         mlog_errno(ret);
5475                         goto out;
5476                 }
5477         }
5478
5479         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5480         if (IS_ERR(handle)) {
5481                 ret = -ENOMEM;
5482                 mlog_errno(ret);
5483                 goto out;
5484         }
5485
5486         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5487                                       OCFS2_JOURNAL_ACCESS_WRITE);
5488         if (ret) {
5489                 mlog_errno(ret);
5490                 goto out_commit;
5491         }
5492
5493         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5494                                   &dealloc);
5495         if (ret) {
5496                 mlog_errno(ret);
5497                 goto out_commit;
5498         }
5499
5500         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5501         ocfs2_journal_dirty(handle, root_bh);
5502
5503         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5504         if (ret)
5505                 mlog_errno(ret);
5506         ocfs2_update_inode_fsync_trans(handle, inode, 0);
5507
5508 out_commit:
5509         ocfs2_commit_trans(osb, handle);
5510 out:
5511         ocfs2_schedule_truncate_log_flush(osb, 1);
5512
5513         inode_unlock(tl_inode);
5514
5515         if (meta_ac)
5516                 ocfs2_free_alloc_context(meta_ac);
5517
5518         ocfs2_run_deallocs(osb, &dealloc);
5519
5520         return ret;
5521 }
5522
5523 /*
5524  * check whether the xattr bucket is filled up with the same hash value.
5525  * If we want to insert the xattr with the same hash, return -ENOSPC.
5526  * If we want to insert a xattr with different hash value, go ahead
5527  * and ocfs2_divide_xattr_bucket will handle this.
5528  */
5529 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5530                                               struct ocfs2_xattr_bucket *bucket,
5531                                               const char *name)
5532 {
5533         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5534         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5535
5536         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5537                 return 0;
5538
5539         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5540             xh->xh_entries[0].xe_name_hash) {
5541                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5542                      "hash = %u\n",
5543                      (unsigned long long)bucket_blkno(bucket),
5544                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5545                 return -ENOSPC;
5546         }
5547
5548         return 0;
5549 }
5550
5551 /*
5552  * Try to set the entry in the current bucket.  If we fail, the caller
5553  * will handle getting us another bucket.
5554  */
5555 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5556                                         struct ocfs2_xattr_info *xi,
5557                                         struct ocfs2_xattr_search *xs,
5558                                         struct ocfs2_xattr_set_ctxt *ctxt)
5559 {
5560         int ret;
5561         struct ocfs2_xa_loc loc;
5562
5563         trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5564
5565         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5566                                        xs->not_found ? NULL : xs->here);
5567         ret = ocfs2_xa_set(&loc, xi, ctxt);
5568         if (!ret) {
5569                 xs->here = loc.xl_entry;
5570                 goto out;
5571         }
5572         if (ret != -ENOSPC) {
5573                 mlog_errno(ret);
5574                 goto out;
5575         }
5576
5577         /* Ok, we need space.  Let's try defragmenting the bucket. */
5578         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5579                                         xs->bucket);
5580         if (ret) {
5581                 mlog_errno(ret);
5582                 goto out;
5583         }
5584
5585         ret = ocfs2_xa_set(&loc, xi, ctxt);
5586         if (!ret) {
5587                 xs->here = loc.xl_entry;
5588                 goto out;
5589         }
5590         if (ret != -ENOSPC)
5591                 mlog_errno(ret);
5592
5593
5594 out:
5595         return ret;
5596 }
5597
5598 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5599                                              struct ocfs2_xattr_info *xi,
5600                                              struct ocfs2_xattr_search *xs,
5601                                              struct ocfs2_xattr_set_ctxt *ctxt)
5602 {
5603         int ret;
5604
5605         trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5606
5607         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5608         if (!ret)
5609                 goto out;
5610         if (ret != -ENOSPC) {
5611                 mlog_errno(ret);
5612                 goto out;
5613         }
5614
5615         /* Ack, need more space.  Let's try to get another bucket! */
5616
5617         /*
5618          * We do not allow for overlapping ranges between buckets. And
5619          * the maximum number of collisions we will allow for then is
5620          * one bucket's worth, so check it here whether we need to
5621          * add a new bucket for the insert.
5622          */
5623         ret = ocfs2_check_xattr_bucket_collision(inode,
5624                                                  xs->bucket,
5625                                                  xi->xi_name);
5626         if (ret) {
5627                 mlog_errno(ret);
5628                 goto out;
5629         }
5630
5631         ret = ocfs2_add_new_xattr_bucket(inode,
5632                                          xs->xattr_bh,
5633                                          xs->bucket,
5634                                          ctxt);
5635         if (ret) {
5636                 mlog_errno(ret);
5637                 goto out;
5638         }
5639
5640         /*
5641          * ocfs2_add_new_xattr_bucket() will have updated
5642          * xs->bucket if it moved, but it will not have updated
5643          * any of the other search fields.  Thus, we drop it and
5644          * re-search.  Everything should be cached, so it'll be
5645          * quick.
5646          */
5647         ocfs2_xattr_bucket_relse(xs->bucket);
5648         ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5649                                            xi->xi_name_index,
5650                                            xi->xi_name, xs);
5651         if (ret && ret != -ENODATA)
5652                 goto out;
5653         xs->not_found = ret;
5654
5655         /* Ok, we have a new bucket, let's try again */
5656         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5657         if (ret && (ret != -ENOSPC))
5658                 mlog_errno(ret);
5659
5660 out:
5661         return ret;
5662 }
5663
5664 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5665                                         struct ocfs2_xattr_bucket *bucket,
5666                                         void *para)
5667 {
5668         int ret = 0, ref_credits;
5669         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5670         u16 i;
5671         struct ocfs2_xattr_entry *xe;
5672         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5673         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5674         int credits = ocfs2_remove_extent_credits(osb->sb) +
5675                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5676         struct ocfs2_xattr_value_root *xv;
5677         struct ocfs2_rm_xattr_bucket_para *args =
5678                         (struct ocfs2_rm_xattr_bucket_para *)para;
5679
5680         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5681
5682         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5683                 xe = &xh->xh_entries[i];
5684                 if (ocfs2_xattr_is_local(xe))
5685                         continue;
5686
5687                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5688                                                       i, &xv, NULL);
5689                 if (ret) {
5690                         mlog_errno(ret);
5691                         break;
5692                 }
5693
5694                 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5695                                                          args->ref_ci,
5696                                                          args->ref_root_bh,
5697                                                          &ctxt.meta_ac,
5698                                                          &ref_credits);
5699
5700                 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5701                 if (IS_ERR(ctxt.handle)) {
5702                         ret = PTR_ERR(ctxt.handle);
5703                         mlog_errno(ret);
5704                         break;
5705                 }
5706
5707                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5708                                                         i, 0, &ctxt);
5709
5710                 ocfs2_commit_trans(osb, ctxt.handle);
5711                 if (ctxt.meta_ac) {
5712                         ocfs2_free_alloc_context(ctxt.meta_ac);
5713                         ctxt.meta_ac = NULL;
5714                 }
5715                 if (ret) {
5716                         mlog_errno(ret);
5717                         break;
5718                 }
5719         }
5720
5721         if (ctxt.meta_ac)
5722                 ocfs2_free_alloc_context(ctxt.meta_ac);
5723         ocfs2_schedule_truncate_log_flush(osb, 1);
5724         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5725         return ret;
5726 }
5727
5728 /*
5729  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5730  * or change the extent record flag), we need to recalculate
5731  * the metaecc for the whole bucket. So it is done here.
5732  *
5733  * Note:
5734  * We have to give the extra credits for the caller.
5735  */
5736 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5737                                             handle_t *handle,
5738                                             void *para)
5739 {
5740         int ret;
5741         struct ocfs2_xattr_bucket *bucket =
5742                         (struct ocfs2_xattr_bucket *)para;
5743
5744         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5745                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5746         if (ret) {
5747                 mlog_errno(ret);
5748                 return ret;
5749         }
5750
5751         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5752
5753         return 0;
5754 }
5755
5756 /*
5757  * Special action we need if the xattr value is refcounted.
5758  *
5759  * 1. If the xattr is refcounted, lock the tree.
5760  * 2. CoW the xattr if we are setting the new value and the value
5761  *    will be stored outside.
5762  * 3. In other case, decrease_refcount will work for us, so just
5763  *    lock the refcount tree, calculate the meta and credits is OK.
5764  *
5765  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5766  * currently CoW is a completed transaction, while this function
5767  * will also lock the allocators and let us deadlock. So we will
5768  * CoW the whole xattr value.
5769  */
5770 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5771                                         struct ocfs2_dinode *di,
5772                                         struct ocfs2_xattr_info *xi,
5773                                         struct ocfs2_xattr_search *xis,
5774                                         struct ocfs2_xattr_search *xbs,
5775                                         struct ocfs2_refcount_tree **ref_tree,
5776                                         int *meta_add,
5777                                         int *credits)
5778 {
5779         int ret = 0;
5780         struct ocfs2_xattr_block *xb;
5781         struct ocfs2_xattr_entry *xe;
5782         char *base;
5783         u32 p_cluster, num_clusters;
5784         unsigned int ext_flags;
5785         int name_offset, name_len;
5786         struct ocfs2_xattr_value_buf vb;
5787         struct ocfs2_xattr_bucket *bucket = NULL;
5788         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5789         struct ocfs2_post_refcount refcount;
5790         struct ocfs2_post_refcount *p = NULL;
5791         struct buffer_head *ref_root_bh = NULL;
5792
5793         if (!xis->not_found) {
5794                 xe = xis->here;
5795                 name_offset = le16_to_cpu(xe->xe_name_offset);
5796                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5797                 base = xis->base;
5798                 vb.vb_bh = xis->inode_bh;
5799                 vb.vb_access = ocfs2_journal_access_di;
5800         } else {
5801                 int i, block_off = 0;
5802                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5803                 xe = xbs->here;
5804                 name_offset = le16_to_cpu(xe->xe_name_offset);
5805                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5806                 i = xbs->here - xbs->header->xh_entries;
5807
5808                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5809                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5810                                                         bucket_xh(xbs->bucket),
5811                                                         i, &block_off,
5812                                                         &name_offset);
5813                         if (ret) {
5814                                 mlog_errno(ret);
5815                                 goto out;
5816                         }
5817                         base = bucket_block(xbs->bucket, block_off);
5818                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5819                         vb.vb_access = ocfs2_journal_access;
5820
5821                         if (ocfs2_meta_ecc(osb)) {
5822                                 /*create parameters for ocfs2_post_refcount. */
5823                                 bucket = xbs->bucket;
5824                                 refcount.credits = bucket->bu_blocks;
5825                                 refcount.para = bucket;
5826                                 refcount.func =
5827                                         ocfs2_xattr_bucket_post_refcount;
5828                                 p = &refcount;
5829                         }
5830                 } else {
5831                         base = xbs->base;
5832                         vb.vb_bh = xbs->xattr_bh;
5833                         vb.vb_access = ocfs2_journal_access_xb;
5834                 }
5835         }
5836
5837         if (ocfs2_xattr_is_local(xe))
5838                 goto out;
5839
5840         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5841                                 (base + name_offset + name_len);
5842
5843         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5844                                        &num_clusters, &vb.vb_xv->xr_list,
5845                                        &ext_flags);
5846         if (ret) {
5847                 mlog_errno(ret);
5848                 goto out;
5849         }
5850
5851         /*
5852          * We just need to check the 1st extent record, since we always
5853          * CoW the whole xattr. So there shouldn't be a xattr with
5854          * some REFCOUNT extent recs after the 1st one.
5855          */
5856         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5857                 goto out;
5858
5859         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5860                                        1, ref_tree, &ref_root_bh);
5861         if (ret) {
5862                 mlog_errno(ret);
5863                 goto out;
5864         }
5865
5866         /*
5867          * If we are deleting the xattr or the new size will be stored inside,
5868          * cool, leave it there, the xattr truncate process will remove them
5869          * for us(it still needs the refcount tree lock and the meta, credits).
5870          * And the worse case is that every cluster truncate will split the
5871          * refcount tree, and make the original extent become 3. So we will need
5872          * 2 * cluster more extent recs at most.
5873          */
5874         if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5875
5876                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5877                                                          &(*ref_tree)->rf_ci,
5878                                                          ref_root_bh, vb.vb_xv,
5879                                                          meta_add, credits);
5880                 if (ret)
5881                         mlog_errno(ret);
5882                 goto out;
5883         }
5884
5885         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5886                                        *ref_tree, ref_root_bh, 0,
5887                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5888         if (ret)
5889                 mlog_errno(ret);
5890
5891 out:
5892         brelse(ref_root_bh);
5893         return ret;
5894 }
5895
5896 /*
5897  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5898  * The physical clusters will be added to refcount tree.
5899  */
5900 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5901                                 struct ocfs2_xattr_value_root *xv,
5902                                 struct ocfs2_extent_tree *value_et,
5903                                 struct ocfs2_caching_info *ref_ci,
5904                                 struct buffer_head *ref_root_bh,
5905                                 struct ocfs2_cached_dealloc_ctxt *dealloc,
5906                                 struct ocfs2_post_refcount *refcount)
5907 {
5908         int ret = 0;
5909         u32 clusters = le32_to_cpu(xv->xr_clusters);
5910         u32 cpos, p_cluster, num_clusters;
5911         struct ocfs2_extent_list *el = &xv->xr_list;
5912         unsigned int ext_flags;
5913
5914         cpos = 0;
5915         while (cpos < clusters) {
5916                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5917                                                &num_clusters, el, &ext_flags);
5918                 if (ret) {
5919                         mlog_errno(ret);
5920                         break;
5921                 }
5922
5923                 cpos += num_clusters;
5924                 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5925                         continue;
5926
5927                 BUG_ON(!p_cluster);
5928
5929                 ret = ocfs2_add_refcount_flag(inode, value_et,
5930                                               ref_ci, ref_root_bh,
5931                                               cpos - num_clusters,
5932                                               p_cluster, num_clusters,
5933                                               dealloc, refcount);
5934                 if (ret) {
5935                         mlog_errno(ret);
5936                         break;
5937                 }
5938         }
5939
5940         return ret;
5941 }
5942
5943 /*
5944  * Given a normal ocfs2_xattr_header, refcount all the entries which
5945  * have value stored outside.
5946  * Used for xattrs stored in inode and ocfs2_xattr_block.
5947  */
5948 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5949                                 struct ocfs2_xattr_value_buf *vb,
5950                                 struct ocfs2_xattr_header *header,
5951                                 struct ocfs2_caching_info *ref_ci,
5952                                 struct buffer_head *ref_root_bh,
5953                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5954 {
5955
5956         struct ocfs2_xattr_entry *xe;
5957         struct ocfs2_xattr_value_root *xv;
5958         struct ocfs2_extent_tree et;
5959         int i, ret = 0;
5960
5961         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5962                 xe = &header->xh_entries[i];
5963
5964                 if (ocfs2_xattr_is_local(xe))
5965                         continue;
5966
5967                 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5968                         le16_to_cpu(xe->xe_name_offset) +
5969                         OCFS2_XATTR_SIZE(xe->xe_name_len));
5970
5971                 vb->vb_xv = xv;
5972                 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5973
5974                 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5975                                                         ref_ci, ref_root_bh,
5976                                                         dealloc, NULL);
5977                 if (ret) {
5978                         mlog_errno(ret);
5979                         break;
5980                 }
5981         }
5982
5983         return ret;
5984 }
5985
5986 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5987                                 struct buffer_head *fe_bh,
5988                                 struct ocfs2_caching_info *ref_ci,
5989                                 struct buffer_head *ref_root_bh,
5990                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5991 {
5992         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5993         struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5994                                 (fe_bh->b_data + inode->i_sb->s_blocksize -
5995                                 le16_to_cpu(di->i_xattr_inline_size));
5996         struct ocfs2_xattr_value_buf vb = {
5997                 .vb_bh = fe_bh,
5998                 .vb_access = ocfs2_journal_access_di,
5999         };
6000
6001         return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6002                                                   ref_ci, ref_root_bh, dealloc);
6003 }
6004
6005 struct ocfs2_xattr_tree_value_refcount_para {
6006         struct ocfs2_caching_info *ref_ci;
6007         struct buffer_head *ref_root_bh;
6008         struct ocfs2_cached_dealloc_ctxt *dealloc;
6009 };
6010
6011 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
6012                                            struct ocfs2_xattr_bucket *bucket,
6013                                            int offset,
6014                                            struct ocfs2_xattr_value_root **xv,
6015                                            struct buffer_head **bh)
6016 {
6017         int ret, block_off, name_offset;
6018         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6019         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6020         void *base;
6021
6022         ret = ocfs2_xattr_bucket_get_name_value(sb,
6023                                                 bucket_xh(bucket),
6024                                                 offset,
6025                                                 &block_off,
6026                                                 &name_offset);
6027         if (ret) {
6028                 mlog_errno(ret);
6029                 goto out;
6030         }
6031
6032         base = bucket_block(bucket, block_off);
6033
6034         *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6035                          OCFS2_XATTR_SIZE(xe->xe_name_len));
6036
6037         if (bh)
6038                 *bh = bucket->bu_bhs[block_off];
6039 out:
6040         return ret;
6041 }
6042
6043 /*
6044  * For a given xattr bucket, refcount all the entries which
6045  * have value stored outside.
6046  */
6047 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6048                                              struct ocfs2_xattr_bucket *bucket,
6049                                              void *para)
6050 {
6051         int i, ret = 0;
6052         struct ocfs2_extent_tree et;
6053         struct ocfs2_xattr_tree_value_refcount_para *ref =
6054                         (struct ocfs2_xattr_tree_value_refcount_para *)para;
6055         struct ocfs2_xattr_header *xh =
6056                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6057         struct ocfs2_xattr_entry *xe;
6058         struct ocfs2_xattr_value_buf vb = {
6059                 .vb_access = ocfs2_journal_access,
6060         };
6061         struct ocfs2_post_refcount refcount = {
6062                 .credits = bucket->bu_blocks,
6063                 .para = bucket,
6064                 .func = ocfs2_xattr_bucket_post_refcount,
6065         };
6066         struct ocfs2_post_refcount *p = NULL;
6067
6068         /* We only need post_refcount if we support metaecc. */
6069         if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6070                 p = &refcount;
6071
6072         trace_ocfs2_xattr_bucket_value_refcount(
6073                                 (unsigned long long)bucket_blkno(bucket),
6074                                 le16_to_cpu(xh->xh_count));
6075         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6076                 xe = &xh->xh_entries[i];
6077
6078                 if (ocfs2_xattr_is_local(xe))
6079                         continue;
6080
6081                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6082                                                       &vb.vb_xv, &vb.vb_bh);
6083                 if (ret) {
6084                         mlog_errno(ret);
6085                         break;
6086                 }
6087
6088                 ocfs2_init_xattr_value_extent_tree(&et,
6089                                                    INODE_CACHE(inode), &vb);
6090
6091                 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6092                                                         &et, ref->ref_ci,
6093                                                         ref->ref_root_bh,
6094                                                         ref->dealloc, p);
6095                 if (ret) {
6096                         mlog_errno(ret);
6097                         break;
6098                 }
6099         }
6100
6101         return ret;
6102
6103 }
6104
6105 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6106                                      struct buffer_head *root_bh,
6107                                      u64 blkno, u32 cpos, u32 len, void *para)
6108 {
6109         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6110                                            ocfs2_xattr_bucket_value_refcount,
6111                                            para);
6112 }
6113
6114 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6115                                 struct buffer_head *blk_bh,
6116                                 struct ocfs2_caching_info *ref_ci,
6117                                 struct buffer_head *ref_root_bh,
6118                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
6119 {
6120         int ret = 0;
6121         struct ocfs2_xattr_block *xb =
6122                                 (struct ocfs2_xattr_block *)blk_bh->b_data;
6123
6124         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6125                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6126                 struct ocfs2_xattr_value_buf vb = {
6127                         .vb_bh = blk_bh,
6128                         .vb_access = ocfs2_journal_access_xb,
6129                 };
6130
6131                 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6132                                                          ref_ci, ref_root_bh,
6133                                                          dealloc);
6134         } else {
6135                 struct ocfs2_xattr_tree_value_refcount_para para = {
6136                         .ref_ci = ref_ci,
6137                         .ref_root_bh = ref_root_bh,
6138                         .dealloc = dealloc,
6139                 };
6140
6141                 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6142                                                 ocfs2_refcount_xattr_tree_rec,
6143                                                 &para);
6144         }
6145
6146         return ret;
6147 }
6148
6149 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6150                                      struct buffer_head *fe_bh,
6151                                      struct ocfs2_caching_info *ref_ci,
6152                                      struct buffer_head *ref_root_bh,
6153                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
6154 {
6155         int ret = 0;
6156         struct ocfs2_inode_info *oi = OCFS2_I(inode);
6157         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6158         struct buffer_head *blk_bh = NULL;
6159
6160         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6161                 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6162                                                          ref_ci, ref_root_bh,
6163                                                          dealloc);
6164                 if (ret) {
6165                         mlog_errno(ret);
6166                         goto out;
6167                 }
6168         }
6169
6170         if (!di->i_xattr_loc)
6171                 goto out;
6172
6173         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6174                                      &blk_bh);
6175         if (ret < 0) {
6176                 mlog_errno(ret);
6177                 goto out;
6178         }
6179
6180         ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6181                                                 ref_root_bh, dealloc);
6182         if (ret)
6183                 mlog_errno(ret);
6184
6185         brelse(blk_bh);
6186 out:
6187
6188         return ret;
6189 }
6190
6191 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6192 /*
6193  * Store the information we need in xattr reflink.
6194  * old_bh and new_bh are inode bh for the old and new inode.
6195  */
6196 struct ocfs2_xattr_reflink {
6197         struct inode *old_inode;
6198         struct inode *new_inode;
6199         struct buffer_head *old_bh;
6200         struct buffer_head *new_bh;
6201         struct ocfs2_caching_info *ref_ci;
6202         struct buffer_head *ref_root_bh;
6203         struct ocfs2_cached_dealloc_ctxt *dealloc;
6204         should_xattr_reflinked *xattr_reflinked;
6205 };
6206
6207 /*
6208  * Given a xattr header and xe offset,
6209  * return the proper xv and the corresponding bh.
6210  * xattr in inode, block and xattr tree have different implementaions.
6211  */
6212 typedef int (get_xattr_value_root)(struct super_block *sb,
6213                                    struct buffer_head *bh,
6214                                    struct ocfs2_xattr_header *xh,
6215                                    int offset,
6216                                    struct ocfs2_xattr_value_root **xv,
6217                                    struct buffer_head **ret_bh,
6218                                    void *para);
6219
6220 /*
6221  * Calculate all the xattr value root metadata stored in this xattr header and
6222  * credits we need if we create them from the scratch.
6223  * We use get_xattr_value_root so that all types of xattr container can use it.
6224  */
6225 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6226                                              struct buffer_head *bh,
6227                                              struct ocfs2_xattr_header *xh,
6228                                              int *metas, int *credits,
6229                                              int *num_recs,
6230                                              get_xattr_value_root *func,
6231                                              void *para)
6232 {
6233         int i, ret = 0;
6234         struct ocfs2_xattr_value_root *xv;
6235         struct ocfs2_xattr_entry *xe;
6236
6237         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6238                 xe = &xh->xh_entries[i];
6239                 if (ocfs2_xattr_is_local(xe))
6240                         continue;
6241
6242                 ret = func(sb, bh, xh, i, &xv, NULL, para);
6243                 if (ret) {
6244                         mlog_errno(ret);
6245                         break;
6246                 }
6247
6248                 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6249                           le16_to_cpu(xv->xr_list.l_next_free_rec);
6250
6251                 *credits += ocfs2_calc_extend_credits(sb,
6252                                                 &def_xv.xv.xr_list);
6253
6254                 /*
6255                  * If the value is a tree with depth > 1, We don't go deep
6256                  * to the extent block, so just calculate a maximum record num.
6257                  */
6258                 if (!xv->xr_list.l_tree_depth)
6259                         *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6260                 else
6261                         *num_recs += ocfs2_clusters_for_bytes(sb,
6262                                                               XATTR_SIZE_MAX);
6263         }
6264
6265         return ret;
6266 }
6267
6268 /* Used by xattr inode and block to return the right xv and buffer_head. */
6269 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6270                                       struct buffer_head *bh,
6271                                       struct ocfs2_xattr_header *xh,
6272                                       int offset,
6273                                       struct ocfs2_xattr_value_root **xv,
6274                                       struct buffer_head **ret_bh,
6275                                       void *para)
6276 {
6277         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6278
6279         *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6280                 le16_to_cpu(xe->xe_name_offset) +
6281                 OCFS2_XATTR_SIZE(xe->xe_name_len));
6282
6283         if (ret_bh)
6284                 *ret_bh = bh;
6285
6286         return 0;
6287 }
6288
6289 /*
6290  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6291  * It is only used for inline xattr and xattr block.
6292  */
6293 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6294                                         struct ocfs2_xattr_header *xh,
6295                                         struct buffer_head *ref_root_bh,
6296                                         int *credits,
6297                                         struct ocfs2_alloc_context **meta_ac)
6298 {
6299         int ret, meta_add = 0, num_recs = 0;
6300         struct ocfs2_refcount_block *rb =
6301                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6302
6303         *credits = 0;
6304
6305         ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6306                                                 &meta_add, credits, &num_recs,
6307                                                 ocfs2_get_xattr_value_root,
6308                                                 NULL);
6309         if (ret) {
6310                 mlog_errno(ret);
6311                 goto out;
6312         }
6313
6314         /*
6315          * We need to add/modify num_recs in refcount tree, so just calculate
6316          * an approximate number we need for refcount tree change.
6317          * Sometimes we need to split the tree, and after split,  half recs
6318          * will be moved to the new block, and a new block can only provide
6319          * half number of recs. So we multiple new blocks by 2.
6320          */
6321         num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6322         meta_add += num_recs;
6323         *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6324         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6325                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6326                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6327         else
6328                 *credits += 1;
6329
6330         ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6331         if (ret)
6332                 mlog_errno(ret);
6333
6334 out:
6335         return ret;
6336 }
6337
6338 /*
6339  * Given a xattr header, reflink all the xattrs in this container.
6340  * It can be used for inode, block and bucket.
6341  *
6342  * NOTE:
6343  * Before we call this function, the caller has memcpy the xattr in
6344  * old_xh to the new_xh.
6345  *
6346  * If args.xattr_reflinked is set, call it to decide whether the xe should
6347  * be reflinked or not. If not, remove it from the new xattr header.
6348  */
6349 static int ocfs2_reflink_xattr_header(handle_t *handle,
6350                                       struct ocfs2_xattr_reflink *args,
6351                                       struct buffer_head *old_bh,
6352                                       struct ocfs2_xattr_header *xh,
6353                                       struct buffer_head *new_bh,
6354                                       struct ocfs2_xattr_header *new_xh,
6355                                       struct ocfs2_xattr_value_buf *vb,
6356                                       struct ocfs2_alloc_context *meta_ac,
6357                                       get_xattr_value_root *func,
6358                                       void *para)
6359 {
6360         int ret = 0, i, j;
6361         struct super_block *sb = args->old_inode->i_sb;
6362         struct buffer_head *value_bh;
6363         struct ocfs2_xattr_entry *xe, *last;
6364         struct ocfs2_xattr_value_root *xv, *new_xv;
6365         struct ocfs2_extent_tree data_et;
6366         u32 clusters, cpos, p_cluster, num_clusters;
6367         unsigned int ext_flags = 0;
6368
6369         trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6370                                          le16_to_cpu(xh->xh_count));
6371
6372         last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6373         for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6374                 xe = &xh->xh_entries[i];
6375
6376                 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6377                         xe = &new_xh->xh_entries[j];
6378
6379                         le16_add_cpu(&new_xh->xh_count, -1);
6380                         if (new_xh->xh_count) {
6381                                 memmove(xe, xe + 1,
6382                                         (void *)last - (void *)xe);
6383                                 memset(last, 0,
6384                                        sizeof(struct ocfs2_xattr_entry));
6385                         }
6386
6387                         /*
6388                          * We don't want j to increase in the next round since
6389                          * it is already moved ahead.
6390                          */
6391                         j--;
6392                         continue;
6393                 }
6394
6395                 if (ocfs2_xattr_is_local(xe))
6396                         continue;
6397
6398                 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6399                 if (ret) {
6400                         mlog_errno(ret);
6401                         break;
6402                 }
6403
6404                 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6405                 if (ret) {
6406                         mlog_errno(ret);
6407                         break;
6408                 }
6409
6410                 /*
6411                  * For the xattr which has l_tree_depth = 0, all the extent
6412                  * recs have already be copied to the new xh with the
6413                  * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6414                  * increase the refount count int the refcount tree.
6415                  *
6416                  * For the xattr which has l_tree_depth > 0, we need
6417                  * to initialize it to the empty default value root,
6418                  * and then insert the extents one by one.
6419                  */
6420                 if (xv->xr_list.l_tree_depth) {
6421                         memcpy(new_xv, &def_xv, OCFS2_XATTR_ROOT_SIZE);
6422                         vb->vb_xv = new_xv;
6423                         vb->vb_bh = value_bh;
6424                         ocfs2_init_xattr_value_extent_tree(&data_et,
6425                                         INODE_CACHE(args->new_inode), vb);
6426                 }
6427
6428                 clusters = le32_to_cpu(xv->xr_clusters);
6429                 cpos = 0;
6430                 while (cpos < clusters) {
6431                         ret = ocfs2_xattr_get_clusters(args->old_inode,
6432                                                        cpos,
6433                                                        &p_cluster,
6434                                                        &num_clusters,
6435                                                        &xv->xr_list,
6436                                                        &ext_flags);
6437                         if (ret) {
6438                                 mlog_errno(ret);
6439                                 goto out;
6440                         }
6441
6442                         BUG_ON(!p_cluster);
6443
6444                         if (xv->xr_list.l_tree_depth) {
6445                                 ret = ocfs2_insert_extent(handle,
6446                                                 &data_et, cpos,
6447                                                 ocfs2_clusters_to_blocks(
6448                                                         args->old_inode->i_sb,
6449                                                         p_cluster),
6450                                                 num_clusters, ext_flags,
6451                                                 meta_ac);
6452                                 if (ret) {
6453                                         mlog_errno(ret);
6454                                         goto out;
6455                                 }
6456                         }
6457
6458                         ret = ocfs2_increase_refcount(handle, args->ref_ci,
6459                                                       args->ref_root_bh,
6460                                                       p_cluster, num_clusters,
6461                                                       meta_ac, args->dealloc);
6462                         if (ret) {
6463                                 mlog_errno(ret);
6464                                 goto out;
6465                         }
6466
6467                         cpos += num_clusters;
6468                 }
6469         }
6470
6471 out:
6472         return ret;
6473 }
6474
6475 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6476 {
6477         int ret = 0, credits = 0;
6478         handle_t *handle;
6479         struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6480         struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6481         int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6482         int header_off = osb->sb->s_blocksize - inline_size;
6483         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6484                                         (args->old_bh->b_data + header_off);
6485         struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6486                                         (args->new_bh->b_data + header_off);
6487         struct ocfs2_alloc_context *meta_ac = NULL;
6488         struct ocfs2_inode_info *new_oi;
6489         struct ocfs2_dinode *new_di;
6490         struct ocfs2_xattr_value_buf vb = {
6491                 .vb_bh = args->new_bh,
6492                 .vb_access = ocfs2_journal_access_di,
6493         };
6494
6495         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6496                                                   &credits, &meta_ac);
6497         if (ret) {
6498                 mlog_errno(ret);
6499                 goto out;
6500         }
6501
6502         handle = ocfs2_start_trans(osb, credits);
6503         if (IS_ERR(handle)) {
6504                 ret = PTR_ERR(handle);
6505                 mlog_errno(ret);
6506                 goto out;
6507         }
6508
6509         ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6510                                       args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6511         if (ret) {
6512                 mlog_errno(ret);
6513                 goto out_commit;
6514         }
6515
6516         memcpy(args->new_bh->b_data + header_off,
6517                args->old_bh->b_data + header_off, inline_size);
6518
6519         new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6520         new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6521
6522         ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6523                                          args->new_bh, new_xh, &vb, meta_ac,
6524                                          ocfs2_get_xattr_value_root, NULL);
6525         if (ret) {
6526                 mlog_errno(ret);
6527                 goto out_commit;
6528         }
6529
6530         new_oi = OCFS2_I(args->new_inode);
6531         /*
6532          * Adjust extent record count to reserve space for extended attribute.
6533          * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6534          */
6535         if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6536             !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6537                 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6538                 le16_add_cpu(&el->l_count, -(inline_size /
6539                                         sizeof(struct ocfs2_extent_rec)));
6540         }
6541         spin_lock(&new_oi->ip_lock);
6542         new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6543         new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6544         spin_unlock(&new_oi->ip_lock);
6545
6546         ocfs2_journal_dirty(handle, args->new_bh);
6547
6548 out_commit:
6549         ocfs2_commit_trans(osb, handle);
6550
6551 out:
6552         if (meta_ac)
6553                 ocfs2_free_alloc_context(meta_ac);
6554         return ret;
6555 }
6556
6557 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6558                                           struct buffer_head *fe_bh,
6559                                           struct buffer_head **ret_bh,
6560                                           int indexed)
6561 {
6562         int ret;
6563         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6564         struct ocfs2_xattr_set_ctxt ctxt;
6565
6566         memset(&ctxt, 0, sizeof(ctxt));
6567         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6568         if (ret < 0) {
6569                 mlog_errno(ret);
6570                 return ret;
6571         }
6572
6573         ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6574         if (IS_ERR(ctxt.handle)) {
6575                 ret = PTR_ERR(ctxt.handle);
6576                 mlog_errno(ret);
6577                 goto out;
6578         }
6579
6580         trace_ocfs2_create_empty_xattr_block(
6581                                 (unsigned long long)fe_bh->b_blocknr, indexed);
6582         ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6583                                        ret_bh);
6584         if (ret)
6585                 mlog_errno(ret);
6586
6587         ocfs2_commit_trans(osb, ctxt.handle);
6588 out:
6589         ocfs2_free_alloc_context(ctxt.meta_ac);
6590         return ret;
6591 }
6592
6593 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6594                                      struct buffer_head *blk_bh,
6595                                      struct buffer_head *new_blk_bh)
6596 {
6597         int ret = 0, credits = 0;
6598         handle_t *handle;
6599         struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6600         struct ocfs2_dinode *new_di;
6601         struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6602         int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6603         struct ocfs2_xattr_block *xb =
6604                         (struct ocfs2_xattr_block *)blk_bh->b_data;
6605         struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6606         struct ocfs2_xattr_block *new_xb =
6607                         (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6608         struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6609         struct ocfs2_alloc_context *meta_ac;
6610         struct ocfs2_xattr_value_buf vb = {
6611                 .vb_bh = new_blk_bh,
6612                 .vb_access = ocfs2_journal_access_xb,
6613         };
6614
6615         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6616                                                   &credits, &meta_ac);
6617         if (ret) {
6618                 mlog_errno(ret);
6619                 return ret;
6620         }
6621
6622         /* One more credits in case we need to add xattr flags in new inode. */
6623         handle = ocfs2_start_trans(osb, credits + 1);
6624         if (IS_ERR(handle)) {
6625                 ret = PTR_ERR(handle);
6626                 mlog_errno(ret);
6627                 goto out;
6628         }
6629
6630         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6631                 ret = ocfs2_journal_access_di(handle,
6632                                               INODE_CACHE(args->new_inode),
6633                                               args->new_bh,
6634                                               OCFS2_JOURNAL_ACCESS_WRITE);
6635                 if (ret) {
6636                         mlog_errno(ret);
6637                         goto out_commit;
6638                 }
6639         }
6640
6641         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6642                                       new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6643         if (ret) {
6644                 mlog_errno(ret);
6645                 goto out_commit;
6646         }
6647
6648         memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6649                osb->sb->s_blocksize - header_off);
6650
6651         ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6652                                          new_blk_bh, new_xh, &vb, meta_ac,
6653                                          ocfs2_get_xattr_value_root, NULL);
6654         if (ret) {
6655                 mlog_errno(ret);
6656                 goto out_commit;
6657         }
6658
6659         ocfs2_journal_dirty(handle, new_blk_bh);
6660
6661         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6662                 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6663                 spin_lock(&new_oi->ip_lock);
6664                 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6665                 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6666                 spin_unlock(&new_oi->ip_lock);
6667
6668                 ocfs2_journal_dirty(handle, args->new_bh);
6669         }
6670
6671 out_commit:
6672         ocfs2_commit_trans(osb, handle);
6673
6674 out:
6675         ocfs2_free_alloc_context(meta_ac);
6676         return ret;
6677 }
6678
6679 struct ocfs2_reflink_xattr_tree_args {
6680         struct ocfs2_xattr_reflink *reflink;
6681         struct buffer_head *old_blk_bh;
6682         struct buffer_head *new_blk_bh;
6683         struct ocfs2_xattr_bucket *old_bucket;
6684         struct ocfs2_xattr_bucket *new_bucket;
6685 };
6686
6687 /*
6688  * NOTE:
6689  * We have to handle the case that both old bucket and new bucket
6690  * will call this function to get the right ret_bh.
6691  * So The caller must give us the right bh.
6692  */
6693 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6694                                         struct buffer_head *bh,
6695                                         struct ocfs2_xattr_header *xh,
6696                                         int offset,
6697                                         struct ocfs2_xattr_value_root **xv,
6698                                         struct buffer_head **ret_bh,
6699                                         void *para)
6700 {
6701         struct ocfs2_reflink_xattr_tree_args *args =
6702                         (struct ocfs2_reflink_xattr_tree_args *)para;
6703         struct ocfs2_xattr_bucket *bucket;
6704
6705         if (bh == args->old_bucket->bu_bhs[0])
6706                 bucket = args->old_bucket;
6707         else
6708                 bucket = args->new_bucket;
6709
6710         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6711                                                xv, ret_bh);
6712 }
6713
6714 struct ocfs2_value_tree_metas {
6715         int num_metas;
6716         int credits;
6717         int num_recs;
6718 };
6719
6720 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6721                                         struct buffer_head *bh,
6722                                         struct ocfs2_xattr_header *xh,
6723                                         int offset,
6724                                         struct ocfs2_xattr_value_root **xv,
6725                                         struct buffer_head **ret_bh,
6726                                         void *para)
6727 {
6728         struct ocfs2_xattr_bucket *bucket =
6729                                 (struct ocfs2_xattr_bucket *)para;
6730
6731         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6732                                                xv, ret_bh);
6733 }
6734
6735 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6736                                       struct ocfs2_xattr_bucket *bucket,
6737                                       void *para)
6738 {
6739         struct ocfs2_value_tree_metas *metas =
6740                         (struct ocfs2_value_tree_metas *)para;
6741         struct ocfs2_xattr_header *xh =
6742                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6743
6744         /* Add the credits for this bucket first. */
6745         metas->credits += bucket->bu_blocks;
6746         return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6747                                         xh, &metas->num_metas,
6748                                         &metas->credits, &metas->num_recs,
6749                                         ocfs2_value_tree_metas_in_bucket,
6750                                         bucket);
6751 }
6752
6753 /*
6754  * Given a xattr extent rec starting from blkno and having len clusters,
6755  * iterate all the buckets calculate how much metadata we need for reflinking
6756  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6757  */
6758 static int ocfs2_lock_reflink_xattr_rec_allocators(
6759                                 struct ocfs2_reflink_xattr_tree_args *args,
6760                                 struct ocfs2_extent_tree *xt_et,
6761                                 u64 blkno, u32 len, int *credits,
6762                                 struct ocfs2_alloc_context **meta_ac,
6763                                 struct ocfs2_alloc_context **data_ac)
6764 {
6765         int ret, num_free_extents;
6766         struct ocfs2_value_tree_metas metas;
6767         struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6768         struct ocfs2_refcount_block *rb;
6769
6770         memset(&metas, 0, sizeof(metas));
6771
6772         ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6773                                           ocfs2_calc_value_tree_metas, &metas);
6774         if (ret) {
6775                 mlog_errno(ret);
6776                 goto out;
6777         }
6778
6779         *credits = metas.credits;
6780
6781         /*
6782          * Calculate we need for refcount tree change.
6783          *
6784          * We need to add/modify num_recs in refcount tree, so just calculate
6785          * an approximate number we need for refcount tree change.
6786          * Sometimes we need to split the tree, and after split,  half recs
6787          * will be moved to the new block, and a new block can only provide
6788          * half number of recs. So we multiple new blocks by 2.
6789          * In the end, we have to add credits for modifying the already
6790          * existed refcount block.
6791          */
6792         rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6793         metas.num_recs =
6794                 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6795                  ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6796         metas.num_metas += metas.num_recs;
6797         *credits += metas.num_recs +
6798                     metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6799         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6800                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6801                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6802         else
6803                 *credits += 1;
6804
6805         /* count in the xattr tree change. */
6806         num_free_extents = ocfs2_num_free_extents(xt_et);
6807         if (num_free_extents < 0) {
6808                 ret = num_free_extents;
6809                 mlog_errno(ret);
6810                 goto out;
6811         }
6812
6813         if (num_free_extents < len)
6814                 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6815
6816         *credits += ocfs2_calc_extend_credits(osb->sb,
6817                                               xt_et->et_root_el);
6818
6819         if (metas.num_metas) {
6820                 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6821                                                         meta_ac);
6822                 if (ret) {
6823                         mlog_errno(ret);
6824                         goto out;
6825                 }
6826         }
6827
6828         if (len) {
6829                 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6830                 if (ret)
6831                         mlog_errno(ret);
6832         }
6833 out:
6834         if (ret) {
6835                 if (*meta_ac) {
6836                         ocfs2_free_alloc_context(*meta_ac);
6837                         *meta_ac = NULL;
6838                 }
6839         }
6840
6841         return ret;
6842 }
6843
6844 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6845                                 u64 blkno, u64 new_blkno, u32 clusters,
6846                                 u32 *cpos, int num_buckets,
6847                                 struct ocfs2_alloc_context *meta_ac,
6848                                 struct ocfs2_alloc_context *data_ac,
6849                                 struct ocfs2_reflink_xattr_tree_args *args)
6850 {
6851         int i, j, ret = 0;
6852         struct super_block *sb = args->reflink->old_inode->i_sb;
6853         int bpb = args->old_bucket->bu_blocks;
6854         struct ocfs2_xattr_value_buf vb = {
6855                 .vb_access = ocfs2_journal_access,
6856         };
6857
6858         for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6859                 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6860                 if (ret) {
6861                         mlog_errno(ret);
6862                         break;
6863                 }
6864
6865                 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6866                 if (ret) {
6867                         mlog_errno(ret);
6868                         break;
6869                 }
6870
6871                 ret = ocfs2_xattr_bucket_journal_access(handle,
6872                                                 args->new_bucket,
6873                                                 OCFS2_JOURNAL_ACCESS_CREATE);
6874                 if (ret) {
6875                         mlog_errno(ret);
6876                         break;
6877                 }
6878
6879                 for (j = 0; j < bpb; j++)
6880                         memcpy(bucket_block(args->new_bucket, j),
6881                                bucket_block(args->old_bucket, j),
6882                                sb->s_blocksize);
6883
6884                 /*
6885                  * Record the start cpos so that we can use it to initialize
6886                  * our xattr tree we also set the xh_num_bucket for the new
6887                  * bucket.
6888                  */
6889                 if (i == 0) {
6890                         *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6891                                             xh_entries[0].xe_name_hash);
6892                         bucket_xh(args->new_bucket)->xh_num_buckets =
6893                                 cpu_to_le16(num_buckets);
6894                 }
6895
6896                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6897
6898                 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6899                                         args->old_bucket->bu_bhs[0],
6900                                         bucket_xh(args->old_bucket),
6901                                         args->new_bucket->bu_bhs[0],
6902                                         bucket_xh(args->new_bucket),
6903                                         &vb, meta_ac,
6904                                         ocfs2_get_reflink_xattr_value_root,
6905                                         args);
6906                 if (ret) {
6907                         mlog_errno(ret);
6908                         break;
6909                 }
6910
6911                 /*
6912                  * Re-access and dirty the bucket to calculate metaecc.
6913                  * Because we may extend the transaction in reflink_xattr_header
6914                  * which will let the already accessed block gone.
6915                  */
6916                 ret = ocfs2_xattr_bucket_journal_access(handle,
6917                                                 args->new_bucket,
6918                                                 OCFS2_JOURNAL_ACCESS_WRITE);
6919                 if (ret) {
6920                         mlog_errno(ret);
6921                         break;
6922                 }
6923
6924                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6925
6926                 ocfs2_xattr_bucket_relse(args->old_bucket);
6927                 ocfs2_xattr_bucket_relse(args->new_bucket);
6928         }
6929
6930         ocfs2_xattr_bucket_relse(args->old_bucket);
6931         ocfs2_xattr_bucket_relse(args->new_bucket);
6932         return ret;
6933 }
6934
6935 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6936                                 struct inode *inode,
6937                                 struct ocfs2_reflink_xattr_tree_args *args,
6938                                 struct ocfs2_extent_tree *et,
6939                                 struct ocfs2_alloc_context *meta_ac,
6940                                 struct ocfs2_alloc_context *data_ac,
6941                                 u64 blkno, u32 cpos, u32 len)
6942 {
6943         int ret, first_inserted = 0;
6944         u32 p_cluster, num_clusters, reflink_cpos = 0;
6945         u64 new_blkno;
6946         unsigned int num_buckets, reflink_buckets;
6947         unsigned int bpc =
6948                 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6949
6950         ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6951         if (ret) {
6952                 mlog_errno(ret);
6953                 goto out;
6954         }
6955         num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6956         ocfs2_xattr_bucket_relse(args->old_bucket);
6957
6958         while (len && num_buckets) {
6959                 ret = ocfs2_claim_clusters(handle, data_ac,
6960                                            1, &p_cluster, &num_clusters);
6961                 if (ret) {
6962                         mlog_errno(ret);
6963                         goto out;
6964                 }
6965
6966                 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6967                 reflink_buckets = min(num_buckets, bpc * num_clusters);
6968
6969                 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6970                                                  new_blkno, num_clusters,
6971                                                  &reflink_cpos, reflink_buckets,
6972                                                  meta_ac, data_ac, args);
6973                 if (ret) {
6974                         mlog_errno(ret);
6975                         goto out;
6976                 }
6977
6978                 /*
6979                  * For the 1st allocated cluster, we make it use the same cpos
6980                  * so that the xattr tree looks the same as the original one
6981                  * in the most case.
6982                  */
6983                 if (!first_inserted) {
6984                         reflink_cpos = cpos;
6985                         first_inserted = 1;
6986                 }
6987                 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6988                                           num_clusters, 0, meta_ac);
6989                 if (ret)
6990                         mlog_errno(ret);
6991
6992                 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6993                                                   num_clusters, reflink_cpos);
6994
6995                 len -= num_clusters;
6996                 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6997                 num_buckets -= reflink_buckets;
6998         }
6999 out:
7000         return ret;
7001 }
7002
7003 /*
7004  * Create the same xattr extent record in the new inode's xattr tree.
7005  */
7006 static int ocfs2_reflink_xattr_rec(struct inode *inode,
7007                                    struct buffer_head *root_bh,
7008                                    u64 blkno,
7009                                    u32 cpos,
7010                                    u32 len,
7011                                    void *para)
7012 {
7013         int ret, credits = 0;
7014         handle_t *handle;
7015         struct ocfs2_reflink_xattr_tree_args *args =
7016                         (struct ocfs2_reflink_xattr_tree_args *)para;
7017         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7018         struct ocfs2_alloc_context *meta_ac = NULL;
7019         struct ocfs2_alloc_context *data_ac = NULL;
7020         struct ocfs2_extent_tree et;
7021
7022         trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
7023
7024         ocfs2_init_xattr_tree_extent_tree(&et,
7025                                           INODE_CACHE(args->reflink->new_inode),
7026                                           args->new_blk_bh);
7027
7028         ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7029                                                       len, &credits,
7030                                                       &meta_ac, &data_ac);
7031         if (ret) {
7032                 mlog_errno(ret);
7033                 goto out;
7034         }
7035
7036         handle = ocfs2_start_trans(osb, credits);
7037         if (IS_ERR(handle)) {
7038                 ret = PTR_ERR(handle);
7039                 mlog_errno(ret);
7040                 goto out;
7041         }
7042
7043         ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7044                                           meta_ac, data_ac,
7045                                           blkno, cpos, len);
7046         if (ret)
7047                 mlog_errno(ret);
7048
7049         ocfs2_commit_trans(osb, handle);
7050
7051 out:
7052         if (meta_ac)
7053                 ocfs2_free_alloc_context(meta_ac);
7054         if (data_ac)
7055                 ocfs2_free_alloc_context(data_ac);
7056         return ret;
7057 }
7058
7059 /*
7060  * Create reflinked xattr buckets.
7061  * We will add bucket one by one, and refcount all the xattrs in the bucket
7062  * if they are stored outside.
7063  */
7064 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7065                                     struct buffer_head *blk_bh,
7066                                     struct buffer_head *new_blk_bh)
7067 {
7068         int ret;
7069         struct ocfs2_reflink_xattr_tree_args para;
7070
7071         memset(&para, 0, sizeof(para));
7072         para.reflink = args;
7073         para.old_blk_bh = blk_bh;
7074         para.new_blk_bh = new_blk_bh;
7075
7076         para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7077         if (!para.old_bucket) {
7078                 mlog_errno(-ENOMEM);
7079                 return -ENOMEM;
7080         }
7081
7082         para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7083         if (!para.new_bucket) {
7084                 ret = -ENOMEM;
7085                 mlog_errno(ret);
7086                 goto out;
7087         }
7088
7089         ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7090                                               ocfs2_reflink_xattr_rec,
7091                                               &para);
7092         if (ret)
7093                 mlog_errno(ret);
7094
7095 out:
7096         ocfs2_xattr_bucket_free(para.old_bucket);
7097         ocfs2_xattr_bucket_free(para.new_bucket);
7098         return ret;
7099 }
7100
7101 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7102                                         struct buffer_head *blk_bh)
7103 {
7104         int ret, indexed = 0;
7105         struct buffer_head *new_blk_bh = NULL;
7106         struct ocfs2_xattr_block *xb =
7107                         (struct ocfs2_xattr_block *)blk_bh->b_data;
7108
7109
7110         if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7111                 indexed = 1;
7112
7113         ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7114                                              &new_blk_bh, indexed);
7115         if (ret) {
7116                 mlog_errno(ret);
7117                 goto out;
7118         }
7119
7120         if (!indexed)
7121                 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7122         else
7123                 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7124         if (ret)
7125                 mlog_errno(ret);
7126
7127 out:
7128         brelse(new_blk_bh);
7129         return ret;
7130 }
7131
7132 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7133 {
7134         int type = ocfs2_xattr_get_type(xe);
7135
7136         return type != OCFS2_XATTR_INDEX_SECURITY &&
7137                type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7138                type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7139 }
7140
7141 int ocfs2_reflink_xattrs(struct inode *old_inode,
7142                          struct buffer_head *old_bh,
7143                          struct inode *new_inode,
7144                          struct buffer_head *new_bh,
7145                          bool preserve_security)
7146 {
7147         int ret;
7148         struct ocfs2_xattr_reflink args;
7149         struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7150         struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7151         struct buffer_head *blk_bh = NULL;
7152         struct ocfs2_cached_dealloc_ctxt dealloc;
7153         struct ocfs2_refcount_tree *ref_tree;
7154         struct buffer_head *ref_root_bh = NULL;
7155
7156         ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7157                                        le64_to_cpu(di->i_refcount_loc),
7158                                        1, &ref_tree, &ref_root_bh);
7159         if (ret) {
7160                 mlog_errno(ret);
7161                 goto out;
7162         }
7163
7164         ocfs2_init_dealloc_ctxt(&dealloc);
7165
7166         args.old_inode = old_inode;
7167         args.new_inode = new_inode;
7168         args.old_bh = old_bh;
7169         args.new_bh = new_bh;
7170         args.ref_ci = &ref_tree->rf_ci;
7171         args.ref_root_bh = ref_root_bh;
7172         args.dealloc = &dealloc;
7173         if (preserve_security)
7174                 args.xattr_reflinked = NULL;
7175         else
7176                 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7177
7178         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7179                 ret = ocfs2_reflink_xattr_inline(&args);
7180                 if (ret) {
7181                         mlog_errno(ret);
7182                         goto out_unlock;
7183                 }
7184         }
7185
7186         if (!di->i_xattr_loc)
7187                 goto out_unlock;
7188
7189         ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7190                                      &blk_bh);
7191         if (ret < 0) {
7192                 mlog_errno(ret);
7193                 goto out_unlock;
7194         }
7195
7196         ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7197         if (ret)
7198                 mlog_errno(ret);
7199
7200         brelse(blk_bh);
7201
7202 out_unlock:
7203         ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7204                                    ref_tree, 1);
7205         brelse(ref_root_bh);
7206
7207         if (ocfs2_dealloc_has_cluster(&dealloc)) {
7208                 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7209                 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7210         }
7211
7212 out:
7213         return ret;
7214 }
7215
7216 /*
7217  * Initialize security and acl for a already created inode.
7218  * Used for reflink a non-preserve-security file.
7219  *
7220  * It uses common api like ocfs2_xattr_set, so the caller
7221  * must not hold any lock expect i_mutex.
7222  */
7223 int ocfs2_init_security_and_acl(struct inode *dir,
7224                                 struct inode *inode,
7225                                 const struct qstr *qstr)
7226 {
7227         int ret = 0;
7228         struct buffer_head *dir_bh = NULL;
7229
7230         ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7231         if (ret) {
7232                 mlog_errno(ret);
7233                 goto leave;
7234         }
7235
7236         ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7237         if (ret) {
7238                 mlog_errno(ret);
7239                 goto leave;
7240         }
7241         ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7242         if (ret)
7243                 mlog_errno(ret);
7244
7245         ocfs2_inode_unlock(dir, 0);
7246         brelse(dir_bh);
7247 leave:
7248         return ret;
7249 }
7250
7251 /*
7252  * 'security' attributes support
7253  */
7254 static int ocfs2_xattr_security_get(const struct xattr_handler *handler,
7255                                     struct dentry *unused, struct inode *inode,
7256                                     const char *name, void *buffer, size_t size)
7257 {
7258         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY,
7259                                name, buffer, size);
7260 }
7261
7262 static int ocfs2_xattr_security_set(const struct xattr_handler *handler,
7263                                     struct dentry *unused, struct inode *inode,
7264                                     const char *name, const void *value,
7265                                     size_t size, int flags)
7266 {
7267         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7268                                name, value, size, flags);
7269 }
7270
7271 static int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7272                      void *fs_info)
7273 {
7274         const struct xattr *xattr;
7275         int err = 0;
7276
7277         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7278                 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7279                                       xattr->name, xattr->value,
7280                                       xattr->value_len, XATTR_CREATE);
7281                 if (err)
7282                         break;
7283         }
7284         return err;
7285 }
7286
7287 int ocfs2_init_security_get(struct inode *inode,
7288                             struct inode *dir,
7289                             const struct qstr *qstr,
7290                             struct ocfs2_security_xattr_info *si)
7291 {
7292         /* check whether ocfs2 support feature xattr */
7293         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7294                 return -EOPNOTSUPP;
7295         if (si)
7296                 return security_old_inode_init_security(inode, dir, qstr,
7297                                                         &si->name, &si->value,
7298                                                         &si->value_len);
7299
7300         return security_inode_init_security(inode, dir, qstr,
7301                                             &ocfs2_initxattrs, NULL);
7302 }
7303
7304 int ocfs2_init_security_set(handle_t *handle,
7305                             struct inode *inode,
7306                             struct buffer_head *di_bh,
7307                             struct ocfs2_security_xattr_info *si,
7308                             struct ocfs2_alloc_context *xattr_ac,
7309                             struct ocfs2_alloc_context *data_ac)
7310 {
7311         return ocfs2_xattr_set_handle(handle, inode, di_bh,
7312                                      OCFS2_XATTR_INDEX_SECURITY,
7313                                      si->name, si->value, si->value_len, 0,
7314                                      xattr_ac, data_ac);
7315 }
7316
7317 const struct xattr_handler ocfs2_xattr_security_handler = {
7318         .prefix = XATTR_SECURITY_PREFIX,
7319         .get    = ocfs2_xattr_security_get,
7320         .set    = ocfs2_xattr_security_set,
7321 };
7322
7323 /*
7324  * 'trusted' attributes support
7325  */
7326 static int ocfs2_xattr_trusted_get(const struct xattr_handler *handler,
7327                                    struct dentry *unused, struct inode *inode,
7328                                    const char *name, void *buffer, size_t size)
7329 {
7330         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED,
7331                                name, buffer, size);
7332 }
7333
7334 static int ocfs2_xattr_trusted_set(const struct xattr_handler *handler,
7335                                    struct dentry *unused, struct inode *inode,
7336                                    const char *name, const void *value,
7337                                    size_t size, int flags)
7338 {
7339         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED,
7340                                name, value, size, flags);
7341 }
7342
7343 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7344         .prefix = XATTR_TRUSTED_PREFIX,
7345         .get    = ocfs2_xattr_trusted_get,
7346         .set    = ocfs2_xattr_trusted_set,
7347 };
7348
7349 /*
7350  * 'user' attributes support
7351  */
7352 static int ocfs2_xattr_user_get(const struct xattr_handler *handler,
7353                                 struct dentry *unused, struct inode *inode,
7354                                 const char *name, void *buffer, size_t size)
7355 {
7356         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7357
7358         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7359                 return -EOPNOTSUPP;
7360         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
7361                                buffer, size);
7362 }
7363
7364 static int ocfs2_xattr_user_set(const struct xattr_handler *handler,
7365                                 struct dentry *unused, struct inode *inode,
7366                                 const char *name, const void *value,
7367                                 size_t size, int flags)
7368 {
7369         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7370
7371         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7372                 return -EOPNOTSUPP;
7373
7374         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER,
7375                                name, value, size, flags);
7376 }
7377
7378 const struct xattr_handler ocfs2_xattr_user_handler = {
7379         .prefix = XATTR_USER_PREFIX,
7380         .get    = ocfs2_xattr_user_get,
7381         .set    = ocfs2_xattr_user_set,
7382 };