Pileus Git - ~andy/linux/blob - 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)
 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                         ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 388                                                       bucket->bu_bhs[i]);
 389         }
 390
 391         if (rc)
 392                 ocfs2_xattr_bucket_relse(bucket);
 393         return rc;
 394 }
 395
 396 /* Read the xattr bucket at xb_blkno */
 397 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 398                                    u64 xb_blkno)
 399 {
 400         int rc;
 401
 402         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 403                                bucket->bu_blocks, bucket->bu_bhs, 0,
 404                                NULL);
 405         if (!rc) {
 406                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 407                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 408                                                  bucket->bu_bhs,
 409                                                  bucket->bu_blocks,
 410                                                  &bucket_xh(bucket)->xh_check);
 411                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 412                 if (rc)
 413                         mlog_errno(rc);
 414         }
 415
 416         if (rc)
 417                 ocfs2_xattr_bucket_relse(bucket);
 418         return rc;
 419 }
 420
 421 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 422                                              struct ocfs2_xattr_bucket *bucket,
 423                                              int type)
 424 {
 425         int i, rc = 0;
 426
 427         for (i = 0; i < bucket->bu_blocks; i++) {
 428                 rc = ocfs2_journal_access(handle,
 429                                           INODE_CACHE(bucket->bu_inode),
 430                                           bucket->bu_bhs[i], type);
 431                 if (rc) {
 432                         mlog_errno(rc);
 433                         break;
 434                 }
 435         }
 436
 437         return rc;
 438 }
 439
 440 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 441                                              struct ocfs2_xattr_bucket *bucket)
 442 {
 443         int i;
 444
 445         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 446         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 447                                    bucket->bu_bhs, bucket->bu_blocks,
 448                                    &bucket_xh(bucket)->xh_check);
 449         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 450
 451         for (i = 0; i < bucket->bu_blocks; i++)
 452                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 453 }
 454
 455 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 456                                          struct ocfs2_xattr_bucket *src)
 457 {
 458         int i;
 459         int blocksize = src->bu_inode->i_sb->s_blocksize;
 460
 461         BUG_ON(dest->bu_blocks != src->bu_blocks);
 462         BUG_ON(dest->bu_inode != src->bu_inode);
 463
 464         for (i = 0; i < src->bu_blocks; i++) {
 465                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 466                        blocksize);
 467         }
 468 }
 469
 470 static int ocfs2_validate_xattr_block(struct super_block *sb,
 471                                       struct buffer_head *bh)
 472 {
 473         int rc;
 474         struct ocfs2_xattr_block *xb =
 475                 (struct ocfs2_xattr_block *)bh->b_data;
 476
 477         trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
 478
 479         BUG_ON(!buffer_uptodate(bh));
 480
 481         /*
 482          * If the ecc fails, we return the error but otherwise
 483          * leave the filesystem running.  We know any error is
 484          * local to this block.
 485          */
 486         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 487         if (rc)
 488                 return rc;
 489
 490         /*
 491          * Errors after here are fatal
 492          */
 493
 494         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 495                 ocfs2_error(sb,
 496                             "Extended attribute block #%llu has bad "
 497                             "signature %.*s",
 498                             (unsigned long long)bh->b_blocknr, 7,
 499                             xb->xb_signature);
 500                 return -EINVAL;
 501         }
 502
 503         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 504                 ocfs2_error(sb,
 505                             "Extended attribute block #%llu has an "
 506                             "invalid xb_blkno of %llu",
 507                             (unsigned long long)bh->b_blocknr,
 508                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 509                 return -EINVAL;
 510         }
 511
 512         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 513                 ocfs2_error(sb,
 514                             "Extended attribute block #%llu has an invalid "
 515                             "xb_fs_generation of #%u",
 516                             (unsigned long long)bh->b_blocknr,
 517                             le32_to_cpu(xb->xb_fs_generation));
 518                 return -EINVAL;
 519         }
 520
 521         return 0;
 522 }
 523
 524 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 525                                   struct buffer_head **bh)
 526 {
 527         int rc;
 528         struct buffer_head *tmp = *bh;
 529
 530         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 531                               ocfs2_validate_xattr_block);
 532
 533         /* If ocfs2_read_block() got us a new bh, pass it up. */
 534         if (!rc && !*bh)
 535                 *bh = tmp;
 536
 537         return rc;
 538 }
 539
 540 static inline const char *ocfs2_xattr_prefix(int name_index)
 541 {
 542         const struct xattr_handler *handler = NULL;
 543
 544         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 545                 handler = ocfs2_xattr_handler_map[name_index];
 546
 547         return handler ? handler->prefix : NULL;
 548 }
 549
 550 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 551                                  const char *name,
 552                                  int name_len)
 553 {
 554         /* Get hash value of uuid from super block */
 555         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 556         int i;
 557
 558         /* hash extended attribute name */
 559         for (i = 0; i < name_len; i++) {
 560                 hash = (hash << OCFS2_HASH_SHIFT) ^
 561                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 562                        *name++;
 563         }
 564
 565         return hash;
 566 }
 567
 568 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 569 {
 570         return namevalue_size(name_len, value_len) +
 571                 sizeof(struct ocfs2_xattr_entry);
 572 }
 573
 574 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 575 {
 576         return namevalue_size_xi(xi) +
 577                 sizeof(struct ocfs2_xattr_entry);
 578 }
 579
 580 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 581 {
 582         return namevalue_size_xe(xe) +
 583                 sizeof(struct ocfs2_xattr_entry);
 584 }
 585
 586 int ocfs2_calc_security_init(struct inode *dir,
 587                              struct ocfs2_security_xattr_info *si,
 588                              int *want_clusters,
 589                              int *xattr_credits,
 590                              struct ocfs2_alloc_context **xattr_ac)
 591 {
 592         int ret = 0;
 593         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 594         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 595                                                  si->value_len);
 596
 597         /*
 598          * The max space of security xattr taken inline is
 599          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 600          * So reserve one metadata block for it is ok.
 601          */
 602         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 603             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 604                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 605                 if (ret) {
 606                         mlog_errno(ret);
 607                         return ret;
 608                 }
 609                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 610         }
 611
 612         /* reserve clusters for xattr value which will be set in B tree*/
 613         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 614                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 615                                                             si->value_len);
 616
 617                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 618                                                            new_clusters);
 619                 *want_clusters += new_clusters;
 620         }
 621         return ret;
 622 }
 623
 624 int ocfs2_calc_xattr_init(struct inode *dir,
 625                           struct buffer_head *dir_bh,
 626                           umode_t mode,
 627                           struct ocfs2_security_xattr_info *si,
 628                           int *want_clusters,
 629                           int *xattr_credits,
 630                           int *want_meta)
 631 {
 632         int ret = 0;
 633         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 634         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 635
 636         if (si->enable)
 637                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 638                                                      si->value_len);
 639
 640         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 641                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 642                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 643                                         "", NULL, 0);
 644                 if (acl_len > 0) {
 645                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 646                         if (S_ISDIR(mode))
 647                                 a_size <<= 1;
 648                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 649                         mlog_errno(ret);
 650                         return ret;
 651                 }
 652         }
 653
 654         if (!(s_size + a_size))
 655                 return ret;
 656
 657         /*
 658          * The max space of security xattr taken inline is
 659          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 660          * The max space of acl xattr taken inline is
 661          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 662          * when blocksize = 512, may reserve one more cluser for
 663          * xattr bucket, otherwise reserve one metadata block
 664          * for them is ok.
 665          * If this is a new directory with inline data,
 666          * we choose to reserve the entire inline area for
 667          * directory contents and force an external xattr block.
 668          */
 669         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 670             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 671             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 672                 *want_meta = *want_meta + 1;
 673                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 674         }
 675
 676         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 677             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 678                 *want_clusters += 1;
 679                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 680         }
 681
 682         /*
 683          * reserve credits and clusters for xattrs which has large value
 684          * and have to be set outside
 685          */
 686         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 687                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 688                                                         si->value_len);
 689                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 690                                                            new_clusters);
 691                 *want_clusters += new_clusters;
 692         }
 693         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 694             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 695                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 696                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 697                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 698                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 699                                                            new_clusters);
 700                 *want_clusters += new_clusters;
 701         }
 702
 703         return ret;
 704 }
 705
 706 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 707                                          u32 clusters_to_add,
 708                                          struct ocfs2_xattr_value_buf *vb,
 709                                          struct ocfs2_xattr_set_ctxt *ctxt)
 710 {
 711         int status = 0, credits;
 712         handle_t *handle = ctxt->handle;
 713         enum ocfs2_alloc_restarted why;
 714         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 715         struct ocfs2_extent_tree et;
 716
 717         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 718
 719         while (clusters_to_add) {
 720                 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 721
 722                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 723                                        OCFS2_JOURNAL_ACCESS_WRITE);
 724                 if (status < 0) {
 725                         mlog_errno(status);
 726                         break;
 727                 }
 728
 729                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 730                 status = ocfs2_add_clusters_in_btree(handle,
 731                                                      &et,
 732                                                      &logical_start,
 733                                                      clusters_to_add,
 734                                                      0,
 735                                                      ctxt->data_ac,
 736                                                      ctxt->meta_ac,
 737                                                      &why);
 738                 if ((status < 0) && (status != -EAGAIN)) {
 739                         if (status != -ENOSPC)
 740                                 mlog_errno(status);
 741                         break;
 742                 }
 743
 744                 ocfs2_journal_dirty(handle, vb->vb_bh);
 745
 746                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 747                                          prev_clusters;
 748
 749                 if (why != RESTART_NONE && clusters_to_add) {
 750                         /*
 751                          * We can only fail in case the alloc file doesn't give
 752                          * up enough clusters.
 753                          */
 754                         BUG_ON(why == RESTART_META);
 755
 756                         credits = ocfs2_calc_extend_credits(inode->i_sb,
 757                                                             &vb->vb_xv->xr_list);
 758                         status = ocfs2_extend_trans(handle, credits);
 759                         if (status < 0) {
 760                                 status = -ENOMEM;
 761                                 mlog_errno(status);
 762                                 break;
 763                         }
 764                 }
 765         }
 766
 767         return status;
 768 }
 769
 770 static int __ocfs2_remove_xattr_range(struct inode *inode,
 771                                       struct ocfs2_xattr_value_buf *vb,
 772                                       u32 cpos, u32 phys_cpos, u32 len,
 773                                       unsigned int ext_flags,
 774                                       struct ocfs2_xattr_set_ctxt *ctxt)
 775 {
 776         int ret;
 777         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 778         handle_t *handle = ctxt->handle;
 779         struct ocfs2_extent_tree et;
 780
 781         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 782
 783         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 784                             OCFS2_JOURNAL_ACCESS_WRITE);
 785         if (ret) {
 786                 mlog_errno(ret);
 787                 goto out;
 788         }
 789
 790         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 791                                   &ctxt->dealloc);
 792         if (ret) {
 793                 mlog_errno(ret);
 794                 goto out;
 795         }
 796
 797         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 798         ocfs2_journal_dirty(handle, vb->vb_bh);
 799
 800         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 801                 ret = ocfs2_decrease_refcount(inode, handle,
 802                                         ocfs2_blocks_to_clusters(inode->i_sb,
 803                                                                  phys_blkno),
 804                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 805         else
 806                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 807                                                   phys_blkno, len);
 808         if (ret)
 809                 mlog_errno(ret);
 810
 811 out:
 812         return ret;
 813 }
 814
 815 static int ocfs2_xattr_shrink_size(struct inode *inode,
 816                                    u32 old_clusters,
 817                                    u32 new_clusters,
 818                                    struct ocfs2_xattr_value_buf *vb,
 819                                    struct ocfs2_xattr_set_ctxt *ctxt)
 820 {
 821         int ret = 0;
 822         unsigned int ext_flags;
 823         u32 trunc_len, cpos, phys_cpos, alloc_size;
 824         u64 block;
 825
 826         if (old_clusters <= new_clusters)
 827                 return 0;
 828
 829         cpos = new_clusters;
 830         trunc_len = old_clusters - new_clusters;
 831         while (trunc_len) {
 832                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 833                                                &alloc_size,
 834                                                &vb->vb_xv->xr_list, &ext_flags);
 835                 if (ret) {
 836                         mlog_errno(ret);
 837                         goto out;
 838                 }
 839
 840                 if (alloc_size > trunc_len)
 841                         alloc_size = trunc_len;
 842
 843                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 844                                                  phys_cpos, alloc_size,
 845                                                  ext_flags, ctxt);
 846                 if (ret) {
 847                         mlog_errno(ret);
 848                         goto out;
 849                 }
 850
 851                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 852                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 853                                                        block, alloc_size);
 854                 cpos += alloc_size;
 855                 trunc_len -= alloc_size;
 856         }
 857
 858 out:
 859         return ret;
 860 }
 861
 862 static int ocfs2_xattr_value_truncate(struct inode *inode,
 863                                       struct ocfs2_xattr_value_buf *vb,
 864                                       int len,
 865                                       struct ocfs2_xattr_set_ctxt *ctxt)
 866 {
 867         int ret;
 868         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 869         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 870
 871         if (new_clusters == old_clusters)
 872                 return 0;
 873
 874         if (new_clusters > old_clusters)
 875                 ret = ocfs2_xattr_extend_allocation(inode,
 876                                                     new_clusters - old_clusters,
 877                                                     vb, ctxt);
 878         else
 879                 ret = ocfs2_xattr_shrink_size(inode,
 880                                               old_clusters, new_clusters,
 881                                               vb, ctxt);
 882
 883         return ret;
 884 }
 885
 886 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 887                                   size_t *result, const char *prefix,
 888                                   const char *name, int name_len)
 889 {
 890         char *p = buffer + *result;
 891         int prefix_len = strlen(prefix);
 892         int total_len = prefix_len + name_len + 1;
 893
 894         *result += total_len;
 895
 896         /* we are just looking for how big our buffer needs to be */
 897         if (!size)
 898                 return 0;
 899
 900         if (*result > size)
 901                 return -ERANGE;
 902
 903         memcpy(p, prefix, prefix_len);
 904         memcpy(p + prefix_len, name, name_len);
 905         p[prefix_len + name_len] = '\0';
 906
 907         return 0;
 908 }
 909
 910 static int ocfs2_xattr_list_entries(struct inode *inode,
 911                                     struct ocfs2_xattr_header *header,
 912                                     char *buffer, size_t buffer_size)
 913 {
 914         size_t result = 0;
 915         int i, type, ret;
 916         const char *prefix, *name;
 917
 918         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 919                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 920                 type = ocfs2_xattr_get_type(entry);
 921                 prefix = ocfs2_xattr_prefix(type);
 922
 923                 if (prefix) {
 924                         name = (const char *)header +
 925                                 le16_to_cpu(entry->xe_name_offset);
 926
 927                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 928                                                      &result, prefix, name,
 929                                                      entry->xe_name_len);
 930                         if (ret)
 931                                 return ret;
 932                 }
 933         }
 934
 935         return result;
 936 }
 937
 938 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 939                                          struct ocfs2_dinode *di)
 940 {
 941         struct ocfs2_xattr_header *xh;
 942         int i;
 943
 944         xh = (struct ocfs2_xattr_header *)
 945                  ((void *)di + inode->i_sb->s_blocksize -
 946                  le16_to_cpu(di->i_xattr_inline_size));
 947
 948         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 949                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 950                         return 1;
 951
 952         return 0;
 953 }
 954
 955 static int ocfs2_xattr_ibody_list(struct inode *inode,
 956                                   struct ocfs2_dinode *di,
 957                                   char *buffer,
 958                                   size_t buffer_size)
 959 {
 960         struct ocfs2_xattr_header *header = NULL;
 961         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 962         int ret = 0;
 963
 964         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 965                 return ret;
 966
 967         header = (struct ocfs2_xattr_header *)
 968                  ((void *)di + inode->i_sb->s_blocksize -
 969                  le16_to_cpu(di->i_xattr_inline_size));
 970
 971         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 972
 973         return ret;
 974 }
 975
 976 static int ocfs2_xattr_block_list(struct inode *inode,
 977                                   struct ocfs2_dinode *di,
 978                                   char *buffer,
 979                                   size_t buffer_size)
 980 {
 981         struct buffer_head *blk_bh = NULL;
 982         struct ocfs2_xattr_block *xb;
 983         int ret = 0;
 984
 985         if (!di->i_xattr_loc)
 986                 return ret;
 987
 988         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 989                                      &blk_bh);
 990         if (ret < 0) {
 991                 mlog_errno(ret);
 992                 return ret;
 993         }
 994
 995         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 996         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 997                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 998                 ret = ocfs2_xattr_list_entries(inode, header,
 999                                                buffer, buffer_size);
1000         } else
1001                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1002                                                    buffer, buffer_size);
1003
1004         brelse(blk_bh);
1005
1006         return ret;
1007 }
1008
1009 ssize_t ocfs2_listxattr(struct dentry *dentry,
1010                         char *buffer,
1011                         size_t size)
1012 {
1013         int ret = 0, i_ret = 0, b_ret = 0;
1014         struct buffer_head *di_bh = NULL;
1015         struct ocfs2_dinode *di = NULL;
1016         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1017
1018         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1019                 return -EOPNOTSUPP;
1020
1021         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1022                 return ret;
1023
1024         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1025         if (ret < 0) {
1026                 mlog_errno(ret);
1027                 return ret;
1028         }
1029
1030         di = (struct ocfs2_dinode *)di_bh->b_data;
1031
1032         down_read(&oi->ip_xattr_sem);
1033         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1034         if (i_ret < 0)
1035                 b_ret = 0;
1036         else {
1037                 if (buffer) {
1038                         buffer += i_ret;
1039                         size -= i_ret;
1040                 }
1041                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1042                                                buffer, size);
1043                 if (b_ret < 0)
1044                         i_ret = 0;
1045         }
1046         up_read(&oi->ip_xattr_sem);
1047         ocfs2_inode_unlock(dentry->d_inode, 0);
1048
1049         brelse(di_bh);
1050
1051         return i_ret + b_ret;
1052 }
1053
1054 static int ocfs2_xattr_find_entry(int name_index,
1055                                   const char *name,
1056                                   struct ocfs2_xattr_search *xs)
1057 {
1058         struct ocfs2_xattr_entry *entry;
1059         size_t name_len;
1060         int i, cmp = 1;
1061
1062         if (name == NULL)
1063                 return -EINVAL;
1064
1065         name_len = strlen(name);
1066         entry = xs->here;
1067         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1068                 cmp = name_index - ocfs2_xattr_get_type(entry);
1069                 if (!cmp)
1070                         cmp = name_len - entry->xe_name_len;
1071                 if (!cmp)
1072                         cmp = memcmp(name, (xs->base +
1073                                      le16_to_cpu(entry->xe_name_offset)),
1074                                      name_len);
1075                 if (cmp == 0)
1076                         break;
1077                 entry += 1;
1078         }
1079         xs->here = entry;
1080
1081         return cmp ? -ENODATA : 0;
1082 }
1083
1084 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1085                                          struct ocfs2_xattr_value_root *xv,
1086                                          void *buffer,
1087                                          size_t len)
1088 {
1089         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1090         u64 blkno;
1091         int i, ret = 0;
1092         size_t cplen, blocksize;
1093         struct buffer_head *bh = NULL;
1094         struct ocfs2_extent_list *el;
1095
1096         el = &xv->xr_list;
1097         clusters = le32_to_cpu(xv->xr_clusters);
1098         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1099         blocksize = inode->i_sb->s_blocksize;
1100
1101         cpos = 0;
1102         while (cpos < clusters) {
1103                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1104                                                &num_clusters, el, NULL);
1105                 if (ret) {
1106                         mlog_errno(ret);
1107                         goto out;
1108                 }
1109
1110                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1111                 /* Copy ocfs2_xattr_value */
1112                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1113                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1114                                                &bh, NULL);
1115                         if (ret) {
1116                                 mlog_errno(ret);
1117                                 goto out;
1118                         }
1119
1120                         cplen = len >= blocksize ? blocksize : len;
1121                         memcpy(buffer, bh->b_data, cplen);
1122                         len -= cplen;
1123                         buffer += cplen;
1124
1125                         brelse(bh);
1126                         bh = NULL;
1127                         if (len == 0)
1128                                 break;
1129                 }
1130                 cpos += num_clusters;
1131         }
1132 out:
1133         return ret;
1134 }
1135
1136 static int ocfs2_xattr_ibody_get(struct inode *inode,
1137                                  int name_index,
1138                                  const char *name,
1139                                  void *buffer,
1140                                  size_t buffer_size,
1141                                  struct ocfs2_xattr_search *xs)
1142 {
1143         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1144         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1145         struct ocfs2_xattr_value_root *xv;
1146         size_t size;
1147         int ret = 0;
1148
1149         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1150                 return -ENODATA;
1151
1152         xs->end = (void *)di + inode->i_sb->s_blocksize;
1153         xs->header = (struct ocfs2_xattr_header *)
1154                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1155         xs->base = (void *)xs->header;
1156         xs->here = xs->header->xh_entries;
1157
1158         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1159         if (ret)
1160                 return ret;
1161         size = le64_to_cpu(xs->here->xe_value_size);
1162         if (buffer) {
1163                 if (size > buffer_size)
1164                         return -ERANGE;
1165                 if (ocfs2_xattr_is_local(xs->here)) {
1166                         memcpy(buffer, (void *)xs->base +
1167                                le16_to_cpu(xs->here->xe_name_offset) +
1168                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1169                 } else {
1170                         xv = (struct ocfs2_xattr_value_root *)
1171                                 (xs->base + le16_to_cpu(
1172                                  xs->here->xe_name_offset) +
1173                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1174                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1175                                                             buffer, size);
1176                         if (ret < 0) {
1177                                 mlog_errno(ret);
1178                                 return ret;
1179                         }
1180                 }
1181         }
1182
1183         return size;
1184 }
1185
1186 static int ocfs2_xattr_block_get(struct inode *inode,
1187                                  int name_index,
1188                                  const char *name,
1189                                  void *buffer,
1190                                  size_t buffer_size,
1191                                  struct ocfs2_xattr_search *xs)
1192 {
1193         struct ocfs2_xattr_block *xb;
1194         struct ocfs2_xattr_value_root *xv;
1195         size_t size;
1196         int ret = -ENODATA, name_offset, name_len, i;
1197         int uninitialized_var(block_off);
1198
1199         xs->bucket = ocfs2_xattr_bucket_new(inode);
1200         if (!xs->bucket) {
1201                 ret = -ENOMEM;
1202                 mlog_errno(ret);
1203                 goto cleanup;
1204         }
1205
1206         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1207         if (ret) {
1208                 mlog_errno(ret);
1209                 goto cleanup;
1210         }
1211
1212         if (xs->not_found) {
1213                 ret = -ENODATA;
1214                 goto cleanup;
1215         }
1216
1217         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1218         size = le64_to_cpu(xs->here->xe_value_size);
1219         if (buffer) {
1220                 ret = -ERANGE;
1221                 if (size > buffer_size)
1222                         goto cleanup;
1223
1224                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1225                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1226                 i = xs->here - xs->header->xh_entries;
1227
1228                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1229                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1230                                                                 bucket_xh(xs->bucket),
1231                                                                 i,
1232                                                                 &block_off,
1233                                                                 &name_offset);
1234                         xs->base = bucket_block(xs->bucket, block_off);
1235                 }
1236                 if (ocfs2_xattr_is_local(xs->here)) {
1237                         memcpy(buffer, (void *)xs->base +
1238                                name_offset + name_len, size);
1239                 } else {
1240                         xv = (struct ocfs2_xattr_value_root *)
1241                                 (xs->base + name_offset + name_len);
1242                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1243                                                             buffer, size);
1244                         if (ret < 0) {
1245                                 mlog_errno(ret);
1246                                 goto cleanup;
1247                         }
1248                 }
1249         }
1250         ret = size;
1251 cleanup:
1252         ocfs2_xattr_bucket_free(xs->bucket);
1253
1254         brelse(xs->xattr_bh);
1255         xs->xattr_bh = NULL;
1256         return ret;
1257 }
1258
1259 int ocfs2_xattr_get_nolock(struct inode *inode,
1260                            struct buffer_head *di_bh,
1261                            int name_index,
1262                            const char *name,
1263                            void *buffer,
1264                            size_t buffer_size)
1265 {
1266         int ret;
1267         struct ocfs2_dinode *di = NULL;
1268         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1269         struct ocfs2_xattr_search xis = {
1270                 .not_found = -ENODATA,
1271         };
1272         struct ocfs2_xattr_search xbs = {
1273                 .not_found = -ENODATA,
1274         };
1275
1276         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1277                 return -EOPNOTSUPP;
1278
1279         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1280                 ret = -ENODATA;
1281
1282         xis.inode_bh = xbs.inode_bh = di_bh;
1283         di = (struct ocfs2_dinode *)di_bh->b_data;
1284
1285         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1286                                     buffer_size, &xis);
1287         if (ret == -ENODATA && di->i_xattr_loc)
1288                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1289                                             buffer_size, &xbs);
1290
1291         return ret;
1292 }
1293
1294 /* ocfs2_xattr_get()
1295  *
1296  * Copy an extended attribute into the buffer provided.
1297  * Buffer is NULL to compute the size of buffer required.
1298  */
1299 static int ocfs2_xattr_get(struct inode *inode,
1300                            int name_index,
1301                            const char *name,
1302                            void *buffer,
1303                            size_t buffer_size)
1304 {
1305         int ret;
1306         struct buffer_head *di_bh = NULL;
1307
1308         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1309         if (ret < 0) {
1310                 mlog_errno(ret);
1311                 return ret;
1312         }
1313         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1314         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1315                                      name, buffer, buffer_size);
1316         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1317
1318         ocfs2_inode_unlock(inode, 0);
1319
1320         brelse(di_bh);
1321
1322         return ret;
1323 }
1324
1325 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1326                                            handle_t *handle,
1327                                            struct ocfs2_xattr_value_buf *vb,
1328                                            const void *value,
1329                                            int value_len)
1330 {
1331         int ret = 0, i, cp_len;
1332         u16 blocksize = inode->i_sb->s_blocksize;
1333         u32 p_cluster, num_clusters;
1334         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1335         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1336         u64 blkno;
1337         struct buffer_head *bh = NULL;
1338         unsigned int ext_flags;
1339         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1340
1341         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1342
1343         while (cpos < clusters) {
1344                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1345                                                &num_clusters, &xv->xr_list,
1346                                                &ext_flags);
1347                 if (ret) {
1348                         mlog_errno(ret);
1349                         goto out;
1350                 }
1351
1352                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1353
1354                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1355
1356                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1357                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1358                                                &bh, NULL);
1359                         if (ret) {
1360                                 mlog_errno(ret);
1361                                 goto out;
1362                         }
1363
1364                         ret = ocfs2_journal_access(handle,
1365                                                    INODE_CACHE(inode),
1366                                                    bh,
1367                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1368                         if (ret < 0) {
1369                                 mlog_errno(ret);
1370                                 goto out;
1371                         }
1372
1373                         cp_len = value_len > blocksize ? blocksize : value_len;
1374                         memcpy(bh->b_data, value, cp_len);
1375                         value_len -= cp_len;
1376                         value += cp_len;
1377                         if (cp_len < blocksize)
1378                                 memset(bh->b_data + cp_len, 0,
1379                                        blocksize - cp_len);
1380
1381                         ocfs2_journal_dirty(handle, bh);
1382                         brelse(bh);
1383                         bh = NULL;
1384
1385                         /*
1386                          * XXX: do we need to empty all the following
1387                          * blocks in this cluster?
1388                          */
1389                         if (!value_len)
1390                                 break;
1391                 }
1392                 cpos += num_clusters;
1393         }
1394 out:
1395         brelse(bh);
1396
1397         return ret;
1398 }
1399
1400 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1401                                        int num_entries)
1402 {
1403         int free_space;
1404
1405         if (!needed_space)
1406                 return 0;
1407
1408         free_space = free_start -
1409                 sizeof(struct ocfs2_xattr_header) -
1410                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1411                 OCFS2_XATTR_HEADER_GAP;
1412         if (free_space < 0)
1413                 return -EIO;
1414         if (free_space < needed_space)
1415                 return -ENOSPC;
1416
1417         return 0;
1418 }
1419
1420 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1421                                    int type)
1422 {
1423         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1424 }
1425
1426 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1427 {
1428         loc->xl_ops->xlo_journal_dirty(handle, loc);
1429 }
1430
1431 /* Give a pointer into the storage for the given offset */
1432 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1433 {
1434         BUG_ON(offset >= loc->xl_size);
1435         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1436 }
1437
1438 /*
1439  * Wipe the name+value pair and allow the storage to reclaim it.  This
1440  * must be followed by either removal of the entry or a call to
1441  * ocfs2_xa_add_namevalue().
1442  */
1443 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1444 {
1445         loc->xl_ops->xlo_wipe_namevalue(loc);
1446 }
1447
1448 /*
1449  * Find lowest offset to a name+value pair.  This is the start of our
1450  * downward-growing free space.
1451  */
1452 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1453 {
1454         return loc->xl_ops->xlo_get_free_start(loc);
1455 }
1456
1457 /* Can we reuse loc->xl_entry for xi? */
1458 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1459                                     struct ocfs2_xattr_info *xi)
1460 {
1461         return loc->xl_ops->xlo_can_reuse(loc, xi);
1462 }
1463
1464 /* How much free space is needed to set the new value */
1465 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1466                                 struct ocfs2_xattr_info *xi)
1467 {
1468         return loc->xl_ops->xlo_check_space(loc, xi);
1469 }
1470
1471 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1472 {
1473         loc->xl_ops->xlo_add_entry(loc, name_hash);
1474         loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1475         /*
1476          * We can't leave the new entry's xe_name_offset at zero or
1477          * add_namevalue() will go nuts.  We set it to the size of our
1478          * storage so that it can never be less than any other entry.
1479          */
1480         loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1481 }
1482
1483 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1484                                    struct ocfs2_xattr_info *xi)
1485 {
1486         int size = namevalue_size_xi(xi);
1487         int nameval_offset;
1488         char *nameval_buf;
1489
1490         loc->xl_ops->xlo_add_namevalue(loc, size);
1491         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1492         loc->xl_entry->xe_name_len = xi->xi_name_len;
1493         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1494         ocfs2_xattr_set_local(loc->xl_entry,
1495                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1496
1497         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1498         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1499         memset(nameval_buf, 0, size);
1500         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1501 }
1502
1503 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1504                                     struct ocfs2_xattr_value_buf *vb)
1505 {
1506         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1507         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1508
1509         /* Value bufs are for value trees */
1510         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1511         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1512                (name_size + OCFS2_XATTR_ROOT_SIZE));
1513
1514         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1515         vb->vb_xv =
1516                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1517                                                         nameval_offset +
1518                                                         name_size);
1519 }
1520
1521 static int ocfs2_xa_block_journal_access(handle_t *handle,
1522                                          struct ocfs2_xa_loc *loc, int type)
1523 {
1524         struct buffer_head *bh = loc->xl_storage;
1525         ocfs2_journal_access_func access;
1526
1527         if (loc->xl_size == (bh->b_size -
1528                              offsetof(struct ocfs2_xattr_block,
1529                                       xb_attrs.xb_header)))
1530                 access = ocfs2_journal_access_xb;
1531         else
1532                 access = ocfs2_journal_access_di;
1533         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1534 }
1535
1536 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1537                                          struct ocfs2_xa_loc *loc)
1538 {
1539         struct buffer_head *bh = loc->xl_storage;
1540
1541         ocfs2_journal_dirty(handle, bh);
1542 }
1543
1544 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1545                                            int offset)
1546 {
1547         return (char *)loc->xl_header + offset;
1548 }
1549
1550 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1551                                     struct ocfs2_xattr_info *xi)
1552 {
1553         /*
1554          * Block storage is strict.  If the sizes aren't exact, we will
1555          * remove the old one and reinsert the new.
1556          */
1557         return namevalue_size_xe(loc->xl_entry) ==
1558                 namevalue_size_xi(xi);
1559 }
1560
1561 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1562 {
1563         struct ocfs2_xattr_header *xh = loc->xl_header;
1564         int i, count = le16_to_cpu(xh->xh_count);
1565         int offset, free_start = loc->xl_size;
1566
1567         for (i = 0; i < count; i++) {
1568                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1569                 if (offset < free_start)
1570                         free_start = offset;
1571         }
1572
1573         return free_start;
1574 }
1575
1576 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1577                                       struct ocfs2_xattr_info *xi)
1578 {
1579         int count = le16_to_cpu(loc->xl_header->xh_count);
1580         int free_start = ocfs2_xa_get_free_start(loc);
1581         int needed_space = ocfs2_xi_entry_usage(xi);
1582
1583         /*
1584          * Block storage will reclaim the original entry before inserting
1585          * the new value, so we only need the difference.  If the new
1586          * entry is smaller than the old one, we don't need anything.
1587          */
1588         if (loc->xl_entry) {
1589                 /* Don't need space if we're reusing! */
1590                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1591                         needed_space = 0;
1592                 else
1593                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1594         }
1595         if (needed_space < 0)
1596                 needed_space = 0;
1597         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1598 }
1599
1600 /*
1601  * Block storage for xattrs keeps the name+value pairs compacted.  When
1602  * we remove one, we have to shift any that preceded it towards the end.
1603  */
1604 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1605 {
1606         int i, offset;
1607         int namevalue_offset, first_namevalue_offset, namevalue_size;
1608         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1609         struct ocfs2_xattr_header *xh = loc->xl_header;
1610         int count = le16_to_cpu(xh->xh_count);
1611
1612         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1613         namevalue_size = namevalue_size_xe(entry);
1614         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1615
1616         /* Shift the name+value pairs */
1617         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1618                 (char *)xh + first_namevalue_offset,
1619                 namevalue_offset - first_namevalue_offset);
1620         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1621
1622         /* Now tell xh->xh_entries about it */
1623         for (i = 0; i < count; i++) {
1624                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1625                 if (offset <= namevalue_offset)
1626                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1627                                      namevalue_size);
1628         }
1629
1630         /*
1631          * Note that we don't update xh_free_start or xh_name_value_len
1632          * because they're not used in block-stored xattrs.
1633          */
1634 }
1635
1636 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1637 {
1638         int count = le16_to_cpu(loc->xl_header->xh_count);
1639         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1640         le16_add_cpu(&loc->xl_header->xh_count, 1);
1641         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1642 }
1643
1644 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1645 {
1646         int free_start = ocfs2_xa_get_free_start(loc);
1647
1648         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1649 }
1650
1651 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1652                                           struct ocfs2_xattr_value_buf *vb)
1653 {
1654         struct buffer_head *bh = loc->xl_storage;
1655
1656         if (loc->xl_size == (bh->b_size -
1657                              offsetof(struct ocfs2_xattr_block,
1658                                       xb_attrs.xb_header)))
1659                 vb->vb_access = ocfs2_journal_access_xb;
1660         else
1661                 vb->vb_access = ocfs2_journal_access_di;
1662         vb->vb_bh = bh;
1663 }
1664
1665 /*
1666  * Operations for xattrs stored in blocks.  This includes inline inode
1667  * storage and unindexed ocfs2_xattr_blocks.
1668  */
1669 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1670         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1671         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1672         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1673         .xlo_check_space        = ocfs2_xa_block_check_space,
1674         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1675         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1676         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1677         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1678         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1679         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1680 };
1681
1682 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1683                                           struct ocfs2_xa_loc *loc, int type)
1684 {
1685         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1686
1687         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1688 }
1689
1690 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1691                                           struct ocfs2_xa_loc *loc)
1692 {
1693         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1694
1695         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1696 }
1697
1698 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1699                                             int offset)
1700 {
1701         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1702         int block, block_offset;
1703
1704         /* The header is at the front of the bucket */
1705         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1706         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1707
1708         return bucket_block(bucket, block) + block_offset;
1709 }
1710
1711 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1712                                      struct ocfs2_xattr_info *xi)
1713 {
1714         return namevalue_size_xe(loc->xl_entry) >=
1715                 namevalue_size_xi(xi);
1716 }
1717
1718 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1719 {
1720         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1721         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1722 }
1723
1724 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1725                                          int free_start, int size)
1726 {
1727         /*
1728          * We need to make sure that the name+value pair fits within
1729          * one block.
1730          */
1731         if (((free_start - size) >> sb->s_blocksize_bits) !=
1732             ((free_start - 1) >> sb->s_blocksize_bits))
1733                 free_start -= free_start % sb->s_blocksize;
1734
1735         return free_start;
1736 }
1737
1738 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1739                                        struct ocfs2_xattr_info *xi)
1740 {
1741         int rc;
1742         int count = le16_to_cpu(loc->xl_header->xh_count);
1743         int free_start = ocfs2_xa_get_free_start(loc);
1744         int needed_space = ocfs2_xi_entry_usage(xi);
1745         int size = namevalue_size_xi(xi);
1746         struct super_block *sb = loc->xl_inode->i_sb;
1747
1748         /*
1749          * Bucket storage does not reclaim name+value pairs it cannot
1750          * reuse.  They live as holes until the bucket fills, and then
1751          * the bucket is defragmented.  However, the bucket can reclaim
1752          * the ocfs2_xattr_entry.
1753          */
1754         if (loc->xl_entry) {
1755                 /* Don't need space if we're reusing! */
1756                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1757                         needed_space = 0;
1758                 else
1759                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1760         }
1761         BUG_ON(needed_space < 0);
1762
1763         if (free_start < size) {
1764                 if (needed_space)
1765                         return -ENOSPC;
1766         } else {
1767                 /*
1768                  * First we check if it would fit in the first place.
1769                  * Below, we align the free start to a block.  This may
1770                  * slide us below the minimum gap.  By checking unaligned
1771                  * first, we avoid that error.
1772                  */
1773                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1774                                                  count);
1775                 if (rc)
1776                         return rc;
1777                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1778                                                            size);
1779         }
1780         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1781 }
1782
1783 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1784 {
1785         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1786                      -namevalue_size_xe(loc->xl_entry));
1787 }
1788
1789 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1790 {
1791         struct ocfs2_xattr_header *xh = loc->xl_header;
1792         int count = le16_to_cpu(xh->xh_count);
1793         int low = 0, high = count - 1, tmp;
1794         struct ocfs2_xattr_entry *tmp_xe;
1795
1796         /*
1797          * We keep buckets sorted by name_hash, so we need to find
1798          * our insert place.
1799          */
1800         while (low <= high && count) {
1801                 tmp = (low + high) / 2;
1802                 tmp_xe = &xh->xh_entries[tmp];
1803
1804                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1805                         low = tmp + 1;
1806                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1807                         high = tmp - 1;
1808                 else {
1809                         low = tmp;
1810                         break;
1811                 }
1812         }
1813
1814         if (low != count)
1815                 memmove(&xh->xh_entries[low + 1],
1816                         &xh->xh_entries[low],
1817                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1818
1819         le16_add_cpu(&xh->xh_count, 1);
1820         loc->xl_entry = &xh->xh_entries[low];
1821         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1822 }
1823
1824 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1825 {
1826         int free_start = ocfs2_xa_get_free_start(loc);
1827         struct ocfs2_xattr_header *xh = loc->xl_header;
1828         struct super_block *sb = loc->xl_inode->i_sb;
1829         int nameval_offset;
1830
1831         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1832         nameval_offset = free_start - size;
1833         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1834         xh->xh_free_start = cpu_to_le16(nameval_offset);
1835         le16_add_cpu(&xh->xh_name_value_len, size);
1836
1837 }
1838
1839 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1840                                            struct ocfs2_xattr_value_buf *vb)
1841 {
1842         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1843         struct super_block *sb = loc->xl_inode->i_sb;
1844         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1845         int size = namevalue_size_xe(loc->xl_entry);
1846         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1847
1848         /* Values are not allowed to straddle block boundaries */
1849         BUG_ON(block_offset !=
1850                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1851         /* We expect the bucket to be filled in */
1852         BUG_ON(!bucket->bu_bhs[block_offset]);
1853
1854         vb->vb_access = ocfs2_journal_access;
1855         vb->vb_bh = bucket->bu_bhs[block_offset];
1856 }
1857
1858 /* Operations for xattrs stored in buckets. */
1859 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1860         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1861         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1862         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1863         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1864         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1865         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1866         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1867         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1868         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1869         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1870 };
1871
1872 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1873 {
1874         struct ocfs2_xattr_value_buf vb;
1875
1876         if (ocfs2_xattr_is_local(loc->xl_entry))
1877                 return 0;
1878
1879         ocfs2_xa_fill_value_buf(loc, &vb);
1880         return le32_to_cpu(vb.vb_xv->xr_clusters);
1881 }
1882
1883 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1884                                    struct ocfs2_xattr_set_ctxt *ctxt)
1885 {
1886         int trunc_rc, access_rc;
1887         struct ocfs2_xattr_value_buf vb;
1888
1889         ocfs2_xa_fill_value_buf(loc, &vb);
1890         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1891                                               ctxt);
1892
1893         /*
1894          * The caller of ocfs2_xa_value_truncate() has already called
1895          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1896          * calls ocfs2_extend_trans().  This may commit the previous
1897          * transaction and open a new one.  If this is a bucket, truncate
1898          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1899          * the caller is expecting to dirty the entire bucket.  So we must
1900          * reset the journal work.  We do this even if truncate has failed,
1901          * as it could have failed after committing the extend.
1902          */
1903         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1904                                             OCFS2_JOURNAL_ACCESS_WRITE);
1905
1906         /* Errors in truncate take precedence */
1907         return trunc_rc ? trunc_rc : access_rc;
1908 }
1909
1910 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1911 {
1912         int index, count;
1913         struct ocfs2_xattr_header *xh = loc->xl_header;
1914         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1915
1916         ocfs2_xa_wipe_namevalue(loc);
1917         loc->xl_entry = NULL;
1918
1919         le16_add_cpu(&xh->xh_count, -1);
1920         count = le16_to_cpu(xh->xh_count);
1921
1922         /*
1923          * Only zero out the entry if there are more remaining.  This is
1924          * important for an empty bucket, as it keeps track of the
1925          * bucket's hash value.  It doesn't hurt empty block storage.
1926          */
1927         if (count) {
1928                 index = ((char *)entry - (char *)&xh->xh_entries) /
1929                         sizeof(struct ocfs2_xattr_entry);
1930                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1931                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1932                 memset(&xh->xh_entries[count], 0,
1933                        sizeof(struct ocfs2_xattr_entry));
1934         }
1935 }
1936
1937 /*
1938  * If we have a problem adjusting the size of an external value during
1939  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1940  * in an intermediate state.  For example, the value may be partially
1941  * truncated.
1942  *
1943  * If the value tree hasn't changed, the extend/truncate went nowhere.
1944  * We have nothing to do.  The caller can treat it as a straight error.
1945  *
1946  * If the value tree got partially truncated, we now have a corrupted
1947  * extended attribute.  We're going to wipe its entry and leak the
1948  * clusters.  Better to leak some storage than leave a corrupt entry.
1949  *
1950  * If the value tree grew, it obviously didn't grow enough for the
1951  * new entry.  We're not going to try and reclaim those clusters either.
1952  * If there was already an external value there (orig_clusters != 0),
1953  * the new clusters are attached safely and we can just leave the old
1954  * value in place.  If there was no external value there, we remove
1955  * the entry.
1956  *
1957  * This way, the xattr block we store in the journal will be consistent.
1958  * If the size change broke because of the journal, no changes will hit
1959  * disk anyway.
1960  */
1961 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1962                                             const char *what,
1963                                             unsigned int orig_clusters)
1964 {
1965         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1966         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1967                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1968
1969         if (new_clusters < orig_clusters) {
1970                 mlog(ML_ERROR,
1971                      "Partial truncate while %s xattr %.*s.  Leaking "
1972                      "%u clusters and removing the entry\n",
1973                      what, loc->xl_entry->xe_name_len, nameval_buf,
1974                      orig_clusters - new_clusters);
1975                 ocfs2_xa_remove_entry(loc);
1976         } else if (!orig_clusters) {
1977                 mlog(ML_ERROR,
1978                      "Unable to allocate an external value for xattr "
1979                      "%.*s safely.  Leaking %u clusters and removing the "
1980                      "entry\n",
1981                      loc->xl_entry->xe_name_len, nameval_buf,
1982                      new_clusters - orig_clusters);
1983                 ocfs2_xa_remove_entry(loc);
1984         } else if (new_clusters > orig_clusters)
1985                 mlog(ML_ERROR,
1986                      "Unable to grow xattr %.*s safely.  %u new clusters "
1987                      "have been added, but the value will not be "
1988                      "modified\n",
1989                      loc->xl_entry->xe_name_len, nameval_buf,
1990                      new_clusters - orig_clusters);
1991 }
1992
1993 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
1994                            struct ocfs2_xattr_set_ctxt *ctxt)
1995 {
1996         int rc = 0;
1997         unsigned int orig_clusters;
1998
1999         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2000                 orig_clusters = ocfs2_xa_value_clusters(loc);
2001                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2002                 if (rc) {
2003                         mlog_errno(rc);
2004                         /*
2005                          * Since this is remove, we can return 0 if
2006                          * ocfs2_xa_cleanup_value_truncate() is going to
2007                          * wipe the entry anyway.  So we check the
2008                          * cluster count as well.
2009                          */
2010                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2011                                 rc = 0;
2012                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2013                                                         orig_clusters);
2014                         if (rc)
2015                                 goto out;
2016                 }
2017         }
2018
2019         ocfs2_xa_remove_entry(loc);
2020
2021 out:
2022         return rc;
2023 }
2024
2025 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2026 {
2027         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2028         char *nameval_buf;
2029
2030         nameval_buf = ocfs2_xa_offset_pointer(loc,
2031                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2032         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2033 }
2034
2035 /*
2036  * Take an existing entry and make it ready for the new value.  This
2037  * won't allocate space, but it may free space.  It should be ready for
2038  * ocfs2_xa_prepare_entry() to finish the work.
2039  */
2040 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2041                                 struct ocfs2_xattr_info *xi,
2042                                 struct ocfs2_xattr_set_ctxt *ctxt)
2043 {
2044         int rc = 0;
2045         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2046         unsigned int orig_clusters;
2047         char *nameval_buf;
2048         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2049         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2050
2051         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2052                name_size);
2053
2054         nameval_buf = ocfs2_xa_offset_pointer(loc,
2055                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2056         if (xe_local) {
2057                 memset(nameval_buf + name_size, 0,
2058                        namevalue_size_xe(loc->xl_entry) - name_size);
2059                 if (!xi_local)
2060                         ocfs2_xa_install_value_root(loc);
2061         } else {
2062                 orig_clusters = ocfs2_xa_value_clusters(loc);
2063                 if (xi_local) {
2064                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2065                         if (rc < 0)
2066                                 mlog_errno(rc);
2067                         else
2068                                 memset(nameval_buf + name_size, 0,
2069                                        namevalue_size_xe(loc->xl_entry) -
2070                                        name_size);
2071                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2072                            xi->xi_value_len) {
2073                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2074                                                      ctxt);
2075                         if (rc < 0)
2076                                 mlog_errno(rc);
2077                 }
2078
2079                 if (rc) {
2080                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2081                                                         orig_clusters);
2082                         goto out;
2083                 }
2084         }
2085
2086         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2087         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2088
2089 out:
2090         return rc;
2091 }
2092
2093 /*
2094  * Prepares loc->xl_entry to receive the new xattr.  This includes
2095  * properly setting up the name+value pair region.  If loc->xl_entry
2096  * already exists, it will take care of modifying it appropriately.
2097  *
2098  * Note that this modifies the data.  You did journal_access already,
2099  * right?
2100  */
2101 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2102                                   struct ocfs2_xattr_info *xi,
2103                                   u32 name_hash,
2104                                   struct ocfs2_xattr_set_ctxt *ctxt)
2105 {
2106         int rc = 0;
2107         unsigned int orig_clusters;
2108         __le64 orig_value_size = 0;
2109
2110         rc = ocfs2_xa_check_space(loc, xi);
2111         if (rc)
2112                 goto out;
2113
2114         if (loc->xl_entry) {
2115                 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2116                         orig_value_size = loc->xl_entry->xe_value_size;
2117                         rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2118                         if (rc)
2119                                 goto out;
2120                         goto alloc_value;
2121                 }
2122
2123                 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2124                         orig_clusters = ocfs2_xa_value_clusters(loc);
2125                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2126                         if (rc) {
2127                                 mlog_errno(rc);
2128                                 ocfs2_xa_cleanup_value_truncate(loc,
2129                                                                 "overwriting",
2130                                                                 orig_clusters);
2131                                 goto out;
2132                         }
2133                 }
2134                 ocfs2_xa_wipe_namevalue(loc);
2135         } else
2136                 ocfs2_xa_add_entry(loc, name_hash);
2137
2138         /*
2139          * If we get here, we have a blank entry.  Fill it.  We grow our
2140          * name+value pair back from the end.
2141          */
2142         ocfs2_xa_add_namevalue(loc, xi);
2143         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2144                 ocfs2_xa_install_value_root(loc);
2145
2146 alloc_value:
2147         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2148                 orig_clusters = ocfs2_xa_value_clusters(loc);
2149                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2150                 if (rc < 0) {
2151                         ctxt->set_abort = 1;
2152                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2153                                                         orig_clusters);
2154                         /*
2155                          * If we were growing an existing value,
2156                          * ocfs2_xa_cleanup_value_truncate() won't remove
2157                          * the entry. We need to restore the original value
2158                          * size.
2159                          */
2160                         if (loc->xl_entry) {
2161                                 BUG_ON(!orig_value_size);
2162                                 loc->xl_entry->xe_value_size = orig_value_size;
2163                         }
2164                         mlog_errno(rc);
2165                 }
2166         }
2167
2168 out:
2169         return rc;
2170 }
2171
2172 /*
2173  * Store the value portion of the name+value pair.  This will skip
2174  * values that are stored externally.  Their tree roots were set up
2175  * by ocfs2_xa_prepare_entry().
2176  */
2177 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2178                                 struct ocfs2_xattr_info *xi,
2179                                 struct ocfs2_xattr_set_ctxt *ctxt)
2180 {
2181         int rc = 0;
2182         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2183         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2184         char *nameval_buf;
2185         struct ocfs2_xattr_value_buf vb;
2186
2187         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2188         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2189                 ocfs2_xa_fill_value_buf(loc, &vb);
2190                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2191                                                      ctxt->handle, &vb,
2192                                                      xi->xi_value,
2193                                                      xi->xi_value_len);
2194         } else
2195                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2196
2197         return rc;
2198 }
2199
2200 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2201                         struct ocfs2_xattr_info *xi,
2202                         struct ocfs2_xattr_set_ctxt *ctxt)
2203 {
2204         int ret;
2205         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2206                                               xi->xi_name_len);
2207
2208         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2209                                       OCFS2_JOURNAL_ACCESS_WRITE);
2210         if (ret) {
2211                 mlog_errno(ret);
2212                 goto out;
2213         }
2214
2215         /*
2216          * From here on out, everything is going to modify the buffer a
2217          * little.  Errors are going to leave the xattr header in a
2218          * sane state.  Thus, even with errors we dirty the sucker.
2219          */
2220
2221         /* Don't worry, we are never called with !xi_value and !xl_entry */
2222         if (!xi->xi_value) {
2223                 ret = ocfs2_xa_remove(loc, ctxt);
2224                 goto out_dirty;
2225         }
2226
2227         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2228         if (ret) {
2229                 if (ret != -ENOSPC)
2230                         mlog_errno(ret);
2231                 goto out_dirty;
2232         }
2233
2234         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2235         if (ret)
2236                 mlog_errno(ret);
2237
2238 out_dirty:
2239         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2240
2241 out:
2242         return ret;
2243 }
2244
2245 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2246                                      struct inode *inode,
2247                                      struct buffer_head *bh,
2248                                      struct ocfs2_xattr_entry *entry)
2249 {
2250         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2251
2252         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2253
2254         loc->xl_inode = inode;
2255         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2256         loc->xl_storage = bh;
2257         loc->xl_entry = entry;
2258         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2259         loc->xl_header =
2260                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2261                                               loc->xl_size);
2262 }
2263
2264 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2265                                           struct inode *inode,
2266                                           struct buffer_head *bh,
2267                                           struct ocfs2_xattr_entry *entry)
2268 {
2269         struct ocfs2_xattr_block *xb =
2270                 (struct ocfs2_xattr_block *)bh->b_data;
2271
2272         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2273
2274         loc->xl_inode = inode;
2275         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2276         loc->xl_storage = bh;
2277         loc->xl_header = &(xb->xb_attrs.xb_header);
2278         loc->xl_entry = entry;
2279         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2280                                              xb_attrs.xb_header);
2281 }
2282
2283 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2284                                            struct ocfs2_xattr_bucket *bucket,
2285                                            struct ocfs2_xattr_entry *entry)
2286 {
2287         loc->xl_inode = bucket->bu_inode;
2288         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2289         loc->xl_storage = bucket;
2290         loc->xl_header = bucket_xh(bucket);
2291         loc->xl_entry = entry;
2292         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2293 }
2294
2295 /*
2296  * In xattr remove, if it is stored outside and refcounted, we may have
2297  * the chance to split the refcount tree. So need the allocators.
2298  */
2299 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2300                                         struct ocfs2_xattr_value_root *xv,
2301                                         struct ocfs2_caching_info *ref_ci,
2302                                         struct buffer_head *ref_root_bh,
2303                                         struct ocfs2_alloc_context **meta_ac,
2304                                         int *ref_credits)
2305 {
2306         int ret, meta_add = 0;
2307         u32 p_cluster, num_clusters;
2308         unsigned int ext_flags;
2309
2310         *ref_credits = 0;
2311         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2312                                        &num_clusters,
2313                                        &xv->xr_list,
2314                                        &ext_flags);
2315         if (ret) {
2316                 mlog_errno(ret);
2317                 goto out;
2318         }
2319
2320         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2321                 goto out;
2322
2323         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2324                                                  ref_root_bh, xv,
2325                                                  &meta_add, ref_credits);
2326         if (ret) {
2327                 mlog_errno(ret);
2328                 goto out;
2329         }
2330
2331         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2332                                                 meta_add, meta_ac);
2333         if (ret)
2334                 mlog_errno(ret);
2335
2336 out:
2337         return ret;
2338 }
2339
2340 static int ocfs2_remove_value_outside(struct inode*inode,
2341                                       struct ocfs2_xattr_value_buf *vb,
2342                                       struct ocfs2_xattr_header *header,
2343                                       struct ocfs2_caching_info *ref_ci,
2344                                       struct buffer_head *ref_root_bh)
2345 {
2346         int ret = 0, i, ref_credits;
2347         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2348         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2349         void *val;
2350
2351         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2352
2353         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2354                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2355
2356                 if (ocfs2_xattr_is_local(entry))
2357                         continue;
2358
2359                 val = (void *)header +
2360                         le16_to_cpu(entry->xe_name_offset);
2361                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2362                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2363
2364                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2365                                                          ref_ci, ref_root_bh,
2366                                                          &ctxt.meta_ac,
2367                                                          &ref_credits);
2368
2369                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2370                                         ocfs2_remove_extent_credits(osb->sb));
2371                 if (IS_ERR(ctxt.handle)) {
2372                         ret = PTR_ERR(ctxt.handle);
2373                         mlog_errno(ret);
2374                         break;
2375                 }
2376
2377                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2378
2379                 ocfs2_commit_trans(osb, ctxt.handle);
2380                 if (ctxt.meta_ac) {
2381                         ocfs2_free_alloc_context(ctxt.meta_ac);
2382                         ctxt.meta_ac = NULL;
2383                 }
2384
2385                 if (ret < 0) {
2386                         mlog_errno(ret);
2387                         break;
2388                 }
2389
2390         }
2391
2392         if (ctxt.meta_ac)
2393                 ocfs2_free_alloc_context(ctxt.meta_ac);
2394         ocfs2_schedule_truncate_log_flush(osb, 1);
2395         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2396         return ret;
2397 }
2398
2399 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2400                                     struct buffer_head *di_bh,
2401                                     struct ocfs2_caching_info *ref_ci,
2402                                     struct buffer_head *ref_root_bh)
2403 {
2404
2405         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2406         struct ocfs2_xattr_header *header;
2407         int ret;
2408         struct ocfs2_xattr_value_buf vb = {
2409                 .vb_bh = di_bh,
2410                 .vb_access = ocfs2_journal_access_di,
2411         };
2412
2413         header = (struct ocfs2_xattr_header *)
2414                  ((void *)di + inode->i_sb->s_blocksize -
2415                  le16_to_cpu(di->i_xattr_inline_size));
2416
2417         ret = ocfs2_remove_value_outside(inode, &vb, header,
2418                                          ref_ci, ref_root_bh);
2419
2420         return ret;
2421 }
2422
2423 struct ocfs2_rm_xattr_bucket_para {
2424         struct ocfs2_caching_info *ref_ci;
2425         struct buffer_head *ref_root_bh;
2426 };
2427
2428 static int ocfs2_xattr_block_remove(struct inode *inode,
2429                                     struct buffer_head *blk_bh,
2430                                     struct ocfs2_caching_info *ref_ci,
2431                                     struct buffer_head *ref_root_bh)
2432 {
2433         struct ocfs2_xattr_block *xb;
2434         int ret = 0;
2435         struct ocfs2_xattr_value_buf vb = {
2436                 .vb_bh = blk_bh,
2437                 .vb_access = ocfs2_journal_access_xb,
2438         };
2439         struct ocfs2_rm_xattr_bucket_para args = {
2440                 .ref_ci = ref_ci,
2441                 .ref_root_bh = ref_root_bh,
2442         };
2443
2444         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2445         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2446                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2447                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2448                                                  ref_ci, ref_root_bh);
2449         } else
2450                 ret = ocfs2_iterate_xattr_index_block(inode,
2451                                                 blk_bh,
2452                                                 ocfs2_rm_xattr_cluster,
2453                                                 &args);
2454
2455         return ret;
2456 }
2457
2458 static int ocfs2_xattr_free_block(struct inode *inode,
2459                                   u64 block,
2460                                   struct ocfs2_caching_info *ref_ci,
2461                                   struct buffer_head *ref_root_bh)
2462 {
2463         struct inode *xb_alloc_inode;
2464         struct buffer_head *xb_alloc_bh = NULL;
2465         struct buffer_head *blk_bh = NULL;
2466         struct ocfs2_xattr_block *xb;
2467         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2468         handle_t *handle;
2469         int ret = 0;
2470         u64 blk, bg_blkno;
2471         u16 bit;
2472
2473         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2474         if (ret < 0) {
2475                 mlog_errno(ret);
2476                 goto out;
2477         }
2478
2479         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2480         if (ret < 0) {
2481                 mlog_errno(ret);
2482                 goto out;
2483         }
2484
2485         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2486         blk = le64_to_cpu(xb->xb_blkno);
2487         bit = le16_to_cpu(xb->xb_suballoc_bit);
2488         if (xb->xb_suballoc_loc)
2489                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2490         else
2491                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2492
2493         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2494                                 EXTENT_ALLOC_SYSTEM_INODE,
2495                                 le16_to_cpu(xb->xb_suballoc_slot));
2496         if (!xb_alloc_inode) {
2497                 ret = -ENOMEM;
2498                 mlog_errno(ret);
2499                 goto out;
2500         }
2501         mutex_lock(&xb_alloc_inode->i_mutex);
2502
2503         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2504         if (ret < 0) {
2505                 mlog_errno(ret);
2506                 goto out_mutex;
2507         }
2508
2509         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2510         if (IS_ERR(handle)) {
2511                 ret = PTR_ERR(handle);
2512                 mlog_errno(ret);
2513                 goto out_unlock;
2514         }
2515
2516         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2517                                        bit, bg_blkno, 1);
2518         if (ret < 0)
2519                 mlog_errno(ret);
2520
2521         ocfs2_commit_trans(osb, handle);
2522 out_unlock:
2523         ocfs2_inode_unlock(xb_alloc_inode, 1);
2524         brelse(xb_alloc_bh);
2525 out_mutex:
2526         mutex_unlock(&xb_alloc_inode->i_mutex);
2527         iput(xb_alloc_inode);
2528 out:
2529         brelse(blk_bh);
2530         return ret;
2531 }
2532
2533 /*
2534  * ocfs2_xattr_remove()
2535  *
2536  * Free extended attribute resources associated with this inode.
2537  */
2538 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2539 {
2540         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2541         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2542         struct ocfs2_refcount_tree *ref_tree = NULL;
2543         struct buffer_head *ref_root_bh = NULL;
2544         struct ocfs2_caching_info *ref_ci = NULL;
2545         handle_t *handle;
2546         int ret;
2547
2548         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2549                 return 0;
2550
2551         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2552                 return 0;
2553
2554         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2555                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2556                                                le64_to_cpu(di->i_refcount_loc),
2557                                                1, &ref_tree, &ref_root_bh);
2558                 if (ret) {
2559                         mlog_errno(ret);
2560                         goto out;
2561                 }
2562                 ref_ci = &ref_tree->rf_ci;
2563
2564         }
2565
2566         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2567                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2568                                                ref_ci, ref_root_bh);
2569                 if (ret < 0) {
2570                         mlog_errno(ret);
2571                         goto out;
2572                 }
2573         }
2574
2575         if (di->i_xattr_loc) {
2576                 ret = ocfs2_xattr_free_block(inode,
2577                                              le64_to_cpu(di->i_xattr_loc),
2578                                              ref_ci, ref_root_bh);
2579                 if (ret < 0) {
2580                         mlog_errno(ret);
2581                         goto out;
2582                 }
2583         }
2584
2585         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2586                                    OCFS2_INODE_UPDATE_CREDITS);
2587         if (IS_ERR(handle)) {
2588                 ret = PTR_ERR(handle);
2589                 mlog_errno(ret);
2590                 goto out;
2591         }
2592         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2593                                       OCFS2_JOURNAL_ACCESS_WRITE);
2594         if (ret) {
2595                 mlog_errno(ret);
2596                 goto out_commit;
2597         }
2598
2599         di->i_xattr_loc = 0;
2600
2601         spin_lock(&oi->ip_lock);
2602         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2603         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2604         spin_unlock(&oi->ip_lock);
2605
2606         ocfs2_journal_dirty(handle, di_bh);
2607 out_commit:
2608         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2609 out:
2610         if (ref_tree)
2611                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2612         brelse(ref_root_bh);
2613         return ret;
2614 }
2615
2616 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2617                                         struct ocfs2_dinode *di)
2618 {
2619         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2620         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2621         int free;
2622
2623         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2624                 return 0;
2625
2626         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2627                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2628                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2629         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2630                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2631                         le64_to_cpu(di->i_size);
2632         } else {
2633                 struct ocfs2_extent_list *el = &di->id2.i_list;
2634                 free = (le16_to_cpu(el->l_count) -
2635                         le16_to_cpu(el->l_next_free_rec)) *
2636                         sizeof(struct ocfs2_extent_rec);
2637         }
2638         if (free >= xattrsize)
2639                 return 1;
2640
2641         return 0;
2642 }
2643
2644 /*
2645  * ocfs2_xattr_ibody_find()
2646  *
2647  * Find extended attribute in inode block and
2648  * fill search info into struct ocfs2_xattr_search.
2649  */
2650 static int ocfs2_xattr_ibody_find(struct inode *inode,
2651                                   int name_index,
2652                                   const char *name,
2653                                   struct ocfs2_xattr_search *xs)
2654 {
2655         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2656         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2657         int ret;
2658         int has_space = 0;
2659
2660         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2661                 return 0;
2662
2663         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2664                 down_read(&oi->ip_alloc_sem);
2665                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2666                 up_read(&oi->ip_alloc_sem);
2667                 if (!has_space)
2668                         return 0;
2669         }
2670
2671         xs->xattr_bh = xs->inode_bh;
2672         xs->end = (void *)di + inode->i_sb->s_blocksize;
2673         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2674                 xs->header = (struct ocfs2_xattr_header *)
2675                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2676         else
2677                 xs->header = (struct ocfs2_xattr_header *)
2678                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2679         xs->base = (void *)xs->header;
2680         xs->here = xs->header->xh_entries;
2681
2682         /* Find the named attribute. */
2683         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2684                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2685                 if (ret && ret != -ENODATA)
2686                         return ret;
2687                 xs->not_found = ret;
2688         }
2689
2690         return 0;
2691 }
2692
2693 static int ocfs2_xattr_ibody_init(struct inode *inode,
2694                                   struct buffer_head *di_bh,
2695                                   struct ocfs2_xattr_set_ctxt *ctxt)
2696 {
2697         int ret;
2698         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2699         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2700         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2701         unsigned int xattrsize = osb->s_xattr_inline_size;
2702
2703         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2704                 ret = -ENOSPC;
2705                 goto out;
2706         }
2707
2708         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2709                                       OCFS2_JOURNAL_ACCESS_WRITE);
2710         if (ret) {
2711                 mlog_errno(ret);
2712                 goto out;
2713         }
2714
2715         /*
2716          * Adjust extent record count or inline data size
2717          * to reserve space for extended attribute.
2718          */
2719         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2720                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2721                 le16_add_cpu(&idata->id_count, -xattrsize);
2722         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2723                 struct ocfs2_extent_list *el = &di->id2.i_list;
2724                 le16_add_cpu(&el->l_count, -(xattrsize /
2725                                              sizeof(struct ocfs2_extent_rec)));
2726         }
2727         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2728
2729         spin_lock(&oi->ip_lock);
2730         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2731         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2732         spin_unlock(&oi->ip_lock);
2733
2734         ocfs2_journal_dirty(ctxt->handle, di_bh);
2735
2736 out:
2737         return ret;
2738 }
2739
2740 /*
2741  * ocfs2_xattr_ibody_set()
2742  *
2743  * Set, replace or remove an extended attribute into inode block.
2744  *
2745  */
2746 static int ocfs2_xattr_ibody_set(struct inode *inode,
2747                                  struct ocfs2_xattr_info *xi,
2748                                  struct ocfs2_xattr_search *xs,
2749                                  struct ocfs2_xattr_set_ctxt *ctxt)
2750 {
2751         int ret;
2752         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2753         struct ocfs2_xa_loc loc;
2754
2755         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2756                 return -ENOSPC;
2757
2758         down_write(&oi->ip_alloc_sem);
2759         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2760                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2761                 if (ret) {
2762                         if (ret != -ENOSPC)
2763                                 mlog_errno(ret);
2764                         goto out;
2765                 }
2766         }
2767
2768         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2769                                  xs->not_found ? NULL : xs->here);
2770         ret = ocfs2_xa_set(&loc, xi, ctxt);
2771         if (ret) {
2772                 if (ret != -ENOSPC)
2773                         mlog_errno(ret);
2774                 goto out;
2775         }
2776         xs->here = loc.xl_entry;
2777
2778 out:
2779         up_write(&oi->ip_alloc_sem);
2780
2781         return ret;
2782 }
2783
2784 /*
2785  * ocfs2_xattr_block_find()
2786  *
2787  * Find extended attribute in external block and
2788  * fill search info into struct ocfs2_xattr_search.
2789  */
2790 static int ocfs2_xattr_block_find(struct inode *inode,
2791                                   int name_index,
2792                                   const char *name,
2793                                   struct ocfs2_xattr_search *xs)
2794 {
2795         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2796         struct buffer_head *blk_bh = NULL;
2797         struct ocfs2_xattr_block *xb;
2798         int ret = 0;
2799
2800         if (!di->i_xattr_loc)
2801                 return ret;
2802
2803         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2804                                      &blk_bh);
2805         if (ret < 0) {
2806                 mlog_errno(ret);
2807                 return ret;
2808         }
2809
2810         xs->xattr_bh = blk_bh;
2811         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2812
2813         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2814                 xs->header = &xb->xb_attrs.xb_header;
2815                 xs->base = (void *)xs->header;
2816                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2817                 xs->here = xs->header->xh_entries;
2818
2819                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2820         } else
2821                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2822                                                    name_index,
2823                                                    name, xs);
2824
2825         if (ret && ret != -ENODATA) {
2826                 xs->xattr_bh = NULL;
2827                 goto cleanup;
2828         }
2829         xs->not_found = ret;
2830         return 0;
2831 cleanup:
2832         brelse(blk_bh);
2833
2834         return ret;
2835 }
2836
2837 static int ocfs2_create_xattr_block(struct inode *inode,
2838                                     struct buffer_head *inode_bh,
2839                                     struct ocfs2_xattr_set_ctxt *ctxt,
2840                                     int indexed,
2841                                     struct buffer_head **ret_bh)
2842 {
2843         int ret;
2844         u16 suballoc_bit_start;
2845         u32 num_got;
2846         u64 suballoc_loc, first_blkno;
2847         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2848         struct buffer_head *new_bh = NULL;
2849         struct ocfs2_xattr_block *xblk;
2850
2851         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2852                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2853         if (ret < 0) {
2854                 mlog_errno(ret);
2855                 goto end;
2856         }
2857
2858         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2859                                    &suballoc_loc, &suballoc_bit_start,
2860                                    &num_got, &first_blkno);
2861         if (ret < 0) {
2862                 mlog_errno(ret);
2863                 goto end;
2864         }
2865
2866         new_bh = sb_getblk(inode->i_sb, first_blkno);
2867         if (!new_bh) {
2868                 ret = -ENOMEM;
2869                 mlog_errno(ret);
2870                 goto end;
2871         }
2872
2873         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2874
2875         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2876                                       new_bh,
2877                                       OCFS2_JOURNAL_ACCESS_CREATE);
2878         if (ret < 0) {
2879                 mlog_errno(ret);
2880                 goto end;
2881         }
2882
2883         /* Initialize ocfs2_xattr_block */
2884         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2885         memset(xblk, 0, inode->i_sb->s_blocksize);
2886         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2887         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2888         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2889         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2890         xblk->xb_fs_generation =
2891                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2892         xblk->xb_blkno = cpu_to_le64(first_blkno);
2893         if (indexed) {
2894                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2895                 xr->xt_clusters = cpu_to_le32(1);
2896                 xr->xt_last_eb_blk = 0;
2897                 xr->xt_list.l_tree_depth = 0;
2898                 xr->xt_list.l_count = cpu_to_le16(
2899                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2900                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2901                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2902         }
2903         ocfs2_journal_dirty(ctxt->handle, new_bh);
2904
2905         /* Add it to the inode */
2906         di->i_xattr_loc = cpu_to_le64(first_blkno);
2907
2908         spin_lock(&OCFS2_I(inode)->ip_lock);
2909         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2910         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2911         spin_unlock(&OCFS2_I(inode)->ip_lock);
2912
2913         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2914
2915         *ret_bh = new_bh;
2916         new_bh = NULL;
2917
2918 end:
2919         brelse(new_bh);
2920         return ret;
2921 }
2922
2923 /*
2924  * ocfs2_xattr_block_set()
2925  *
2926  * Set, replace or remove an extended attribute into external block.
2927  *
2928  */
2929 static int ocfs2_xattr_block_set(struct inode *inode,
2930                                  struct ocfs2_xattr_info *xi,
2931                                  struct ocfs2_xattr_search *xs,
2932                                  struct ocfs2_xattr_set_ctxt *ctxt)
2933 {
2934         struct buffer_head *new_bh = NULL;
2935         struct ocfs2_xattr_block *xblk = NULL;
2936         int ret;
2937         struct ocfs2_xa_loc loc;
2938
2939         if (!xs->xattr_bh) {
2940                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2941                                                0, &new_bh);
2942                 if (ret) {
2943                         mlog_errno(ret);
2944                         goto end;
2945                 }
2946
2947                 xs->xattr_bh = new_bh;
2948                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2949                 xs->header = &xblk->xb_attrs.xb_header;
2950                 xs->base = (void *)xs->header;
2951                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2952                 xs->here = xs->header->xh_entries;
2953         } else
2954                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2955
2956         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2957                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2958                                               xs->not_found ? NULL : xs->here);
2959
2960                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2961                 if (!ret)
2962                         xs->here = loc.xl_entry;
2963                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2964                         goto end;
2965                 else {
2966                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2967                         if (ret)
2968                                 goto end;
2969                 }
2970         }
2971
2972         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2973                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2974
2975 end:
2976         return ret;
2977 }
2978
2979 /* Check whether the new xattr can be inserted into the inode. */
2980 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2981                                        struct ocfs2_xattr_info *xi,
2982                                        struct ocfs2_xattr_search *xs)
2983 {
2984         struct ocfs2_xattr_entry *last;
2985         int free, i;
2986         size_t min_offs = xs->end - xs->base;
2987
2988         if (!xs->header)
2989                 return 0;
2990
2991         last = xs->header->xh_entries;
2992
2993         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2994                 size_t offs = le16_to_cpu(last->xe_name_offset);
2995                 if (offs < min_offs)
2996                         min_offs = offs;
2997                 last += 1;
2998         }
2999
3000         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3001         if (free < 0)
3002                 return 0;
3003
3004         BUG_ON(!xs->not_found);
3005
3006         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3007                 return 1;
3008
3009         return 0;
3010 }
3011
3012 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3013                                      struct ocfs2_dinode *di,
3014                                      struct ocfs2_xattr_info *xi,
3015                                      struct ocfs2_xattr_search *xis,
3016                                      struct ocfs2_xattr_search *xbs,
3017                                      int *clusters_need,
3018                                      int *meta_need,
3019                                      int *credits_need)
3020 {
3021         int ret = 0, old_in_xb = 0;
3022         int clusters_add = 0, meta_add = 0, credits = 0;
3023         struct buffer_head *bh = NULL;
3024         struct ocfs2_xattr_block *xb = NULL;
3025         struct ocfs2_xattr_entry *xe = NULL;
3026         struct ocfs2_xattr_value_root *xv = NULL;
3027         char *base = NULL;
3028         int name_offset, name_len = 0;
3029         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3030                                                     xi->xi_value_len);
3031         u64 value_size;
3032
3033         /*
3034          * Calculate the clusters we need to write.
3035          * No matter whether we replace an old one or add a new one,
3036          * we need this for writing.
3037          */
3038         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3039                 credits += new_clusters *
3040                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3041
3042         if (xis->not_found && xbs->not_found) {
3043                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3044
3045                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3046                         clusters_add += new_clusters;
3047                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3048                                                         &def_xv.xv.xr_list);
3049                 }
3050
3051                 goto meta_guess;
3052         }
3053
3054         if (!xis->not_found) {
3055                 xe = xis->here;
3056                 name_offset = le16_to_cpu(xe->xe_name_offset);
3057                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3058                 base = xis->base;
3059                 credits += OCFS2_INODE_UPDATE_CREDITS;
3060         } else {
3061                 int i, block_off = 0;
3062                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3063                 xe = xbs->here;
3064                 name_offset = le16_to_cpu(xe->xe_name_offset);
3065                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3066                 i = xbs->here - xbs->header->xh_entries;
3067                 old_in_xb = 1;
3068
3069                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3070                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3071                                                         bucket_xh(xbs->bucket),
3072                                                         i, &block_off,
3073                                                         &name_offset);
3074                         base = bucket_block(xbs->bucket, block_off);
3075                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3076                 } else {
3077                         base = xbs->base;
3078                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3079                 }
3080         }
3081
3082         /*
3083          * delete a xattr doesn't need metadata and cluster allocation.
3084          * so just calculate the credits and return.
3085          *
3086          * The credits for removing the value tree will be extended
3087          * by ocfs2_remove_extent itself.
3088          */
3089         if (!xi->xi_value) {
3090                 if (!ocfs2_xattr_is_local(xe))
3091                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3092
3093                 goto out;
3094         }
3095
3096         /* do cluster allocation guess first. */
3097         value_size = le64_to_cpu(xe->xe_value_size);
3098
3099         if (old_in_xb) {
3100                 /*
3101                  * In xattr set, we always try to set the xe in inode first,
3102                  * so if it can be inserted into inode successfully, the old
3103                  * one will be removed from the xattr block, and this xattr
3104                  * will be inserted into inode as a new xattr in inode.
3105                  */
3106                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3107                         clusters_add += new_clusters;
3108                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3109                                     OCFS2_INODE_UPDATE_CREDITS;
3110                         if (!ocfs2_xattr_is_local(xe))
3111                                 credits += ocfs2_calc_extend_credits(
3112                                                         inode->i_sb,
3113                                                         &def_xv.xv.xr_list);
3114                         goto out;
3115                 }
3116         }
3117
3118         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3119                 /* the new values will be stored outside. */
3120                 u32 old_clusters = 0;
3121
3122                 if (!ocfs2_xattr_is_local(xe)) {
3123                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3124                                                                  value_size);
3125                         xv = (struct ocfs2_xattr_value_root *)
3126                              (base + name_offset + name_len);
3127                         value_size = OCFS2_XATTR_ROOT_SIZE;
3128                 } else
3129                         xv = &def_xv.xv;
3130
3131                 if (old_clusters >= new_clusters) {
3132                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3133                         goto out;
3134                 } else {
3135                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3136                         clusters_add += new_clusters - old_clusters;
3137                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3138                                                              &xv->xr_list);
3139                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3140                                 goto out;
3141                 }
3142         } else {
3143                 /*
3144                  * Now the new value will be stored inside. So if the new
3145                  * value is smaller than the size of value root or the old
3146                  * value, we don't need any allocation, otherwise we have
3147                  * to guess metadata allocation.
3148                  */
3149                 if ((ocfs2_xattr_is_local(xe) &&
3150                      (value_size >= xi->xi_value_len)) ||
3151                     (!ocfs2_xattr_is_local(xe) &&
3152                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3153                         goto out;
3154         }
3155
3156 meta_guess:
3157         /* calculate metadata allocation. */
3158         if (di->i_xattr_loc) {
3159                 if (!xbs->xattr_bh) {
3160                         ret = ocfs2_read_xattr_block(inode,
3161                                                      le64_to_cpu(di->i_xattr_loc),
3162                                                      &bh);
3163                         if (ret) {
3164                                 mlog_errno(ret);
3165                                 goto out;
3166                         }
3167
3168                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3169                 } else
3170                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3171
3172                 /*
3173                  * If there is already an xattr tree, good, we can calculate
3174                  * like other b-trees. Otherwise we may have the chance of
3175                  * create a tree, the credit calculation is borrowed from
3176                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3177                  * new tree will be cluster based, so no meta is needed.
3178                  */
3179                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3180                         struct ocfs2_extent_list *el =
3181                                  &xb->xb_attrs.xb_root.xt_list;
3182                         meta_add += ocfs2_extend_meta_needed(el);
3183                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3184                                                              el);
3185                 } else
3186                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3187
3188                 /*
3189                  * This cluster will be used either for new bucket or for
3190                  * new xattr block.
3191                  * If the cluster size is the same as the bucket size, one
3192                  * more is needed since we may need to extend the bucket
3193                  * also.
3194                  */
3195                 clusters_add += 1;
3196                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3197                 if (OCFS2_XATTR_BUCKET_SIZE ==
3198                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3199                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3200                         clusters_add += 1;
3201                 }
3202         } else {
3203                 meta_add += 1;
3204                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3205         }
3206 out:
3207         if (clusters_need)
3208                 *clusters_need = clusters_add;
3209         if (meta_need)
3210                 *meta_need = meta_add;
3211         if (credits_need)
3212                 *credits_need = credits;
3213         brelse(bh);
3214         return ret;
3215 }
3216
3217 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3218                                      struct ocfs2_dinode *di,
3219                                      struct ocfs2_xattr_info *xi,
3220                                      struct ocfs2_xattr_search *xis,
3221                                      struct ocfs2_xattr_search *xbs,
3222                                      struct ocfs2_xattr_set_ctxt *ctxt,
3223                                      int extra_meta,
3224                                      int *credits)
3225 {
3226         int clusters_add, meta_add, ret;
3227         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3228
3229         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3230
3231         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3232
3233         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3234                                         &clusters_add, &meta_add, credits);
3235         if (ret) {
3236                 mlog_errno(ret);
3237                 return ret;
3238         }
3239
3240         meta_add += extra_meta;
3241         trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3242                                         clusters_add, *credits);
3243
3244         if (meta_add) {
3245                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3246                                                         &ctxt->meta_ac);
3247                 if (ret) {
3248                         mlog_errno(ret);
3249                         goto out;
3250                 }
3251         }
3252
3253         if (clusters_add) {
3254                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3255                 if (ret)
3256                         mlog_errno(ret);
3257         }
3258 out:
3259         if (ret) {
3260                 if (ctxt->meta_ac) {
3261                         ocfs2_free_alloc_context(ctxt->meta_ac);
3262                         ctxt->meta_ac = NULL;
3263                 }
3264
3265                 /*
3266                  * We cannot have an error and a non null ctxt->data_ac.
3267                  */
3268         }
3269
3270         return ret;
3271 }
3272
3273 static int __ocfs2_xattr_set_handle(struct inode *inode,
3274                                     struct ocfs2_dinode *di,
3275                                     struct ocfs2_xattr_info *xi,
3276                                     struct ocfs2_xattr_search *xis,
3277                                     struct ocfs2_xattr_search *xbs,
3278                                     struct ocfs2_xattr_set_ctxt *ctxt)
3279 {
3280         int ret = 0, credits, old_found;
3281
3282         if (!xi->xi_value) {
3283                 /* Remove existing extended attribute */
3284                 if (!xis->not_found)
3285                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3286                 else if (!xbs->not_found)
3287                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3288         } else {
3289                 /* We always try to set extended attribute into inode first*/
3290                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3291                 if (!ret && !xbs->not_found) {
3292                         /*
3293                          * If succeed and that extended attribute existing in
3294                          * external block, then we will remove it.
3295                          */
3296                         xi->xi_value = NULL;
3297                         xi->xi_value_len = 0;
3298
3299                         old_found = xis->not_found;
3300                         xis->not_found = -ENODATA;
3301                         ret = ocfs2_calc_xattr_set_need(inode,
3302                                                         di,
3303                                                         xi,
3304                                                         xis,
3305                                                         xbs,
3306                                                         NULL,
3307                                                         NULL,
3308                                                         &credits);
3309                         xis->not_found = old_found;
3310                         if (ret) {
3311                                 mlog_errno(ret);
3312                                 goto out;
3313                         }
3314
3315                         ret = ocfs2_extend_trans(ctxt->handle, credits);
3316                         if (ret) {
3317                                 mlog_errno(ret);
3318                                 goto out;
3319                         }
3320                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3321                 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3322                         if (di->i_xattr_loc && !xbs->xattr_bh) {
3323                                 ret = ocfs2_xattr_block_find(inode,
3324                                                              xi->xi_name_index,
3325                                                              xi->xi_name, xbs);
3326                                 if (ret)
3327                                         goto out;
3328
3329                                 old_found = xis->not_found;
3330                                 xis->not_found = -ENODATA;
3331                                 ret = ocfs2_calc_xattr_set_need(inode,
3332                                                                 di,
3333                                                                 xi,
3334                                                                 xis,
3335                                                                 xbs,
3336                                                                 NULL,
3337                                                                 NULL,
3338                                                                 &credits);
3339                                 xis->not_found = old_found;
3340                                 if (ret) {
3341                                         mlog_errno(ret);
3342                                         goto out;
3343                                 }
3344
3345                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3346                                 if (ret) {
3347                                         mlog_errno(ret);
3348                                         goto out;
3349                                 }
3350                         }
3351                         /*
3352                          * If no space in inode, we will set extended attribute
3353                          * into external block.
3354                          */
3355                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3356                         if (ret)
3357                                 goto out;
3358                         if (!xis->not_found) {
3359                                 /*
3360                                  * If succeed and that extended attribute
3361                                  * existing in inode, we will remove it.
3362                                  */
3363                                 xi->xi_value = NULL;
3364                                 xi->xi_value_len = 0;
3365                                 xbs->not_found = -ENODATA;
3366                                 ret = ocfs2_calc_xattr_set_need(inode,
3367                                                                 di,
3368                                                                 xi,
3369                                                                 xis,
3370                                                                 xbs,
3371                                                                 NULL,
3372                                                                 NULL,
3373                                                                 &credits);
3374                                 if (ret) {
3375                                         mlog_errno(ret);
3376                                         goto out;
3377                                 }
3378
3379                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3380                                 if (ret) {
3381                                         mlog_errno(ret);
3382                                         goto out;
3383                                 }
3384                                 ret = ocfs2_xattr_ibody_set(inode, xi,
3385                                                             xis, ctxt);
3386                         }
3387                 }
3388         }
3389
3390         if (!ret) {
3391                 /* Update inode ctime. */
3392                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3393                                               xis->inode_bh,
3394                                               OCFS2_JOURNAL_ACCESS_WRITE);
3395                 if (ret) {
3396                         mlog_errno(ret);
3397                         goto out;
3398                 }
3399
3400                 inode->i_ctime = CURRENT_TIME;
3401                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3402                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3403                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3404         }
3405 out:
3406         return ret;
3407 }
3408
3409 /*
3410  * This function only called duing creating inode
3411  * for init security/acl xattrs of the new inode.
3412  * All transanction credits have been reserved in mknod.
3413  */
3414 int ocfs2_xattr_set_handle(handle_t *handle,
3415                            struct inode *inode,
3416                            struct buffer_head *di_bh,
3417                            int name_index,
3418                            const char *name,
3419                            const void *value,
3420                            size_t value_len,
3421                            int flags,
3422                            struct ocfs2_alloc_context *meta_ac,
3423                            struct ocfs2_alloc_context *data_ac)
3424 {
3425         struct ocfs2_dinode *di;
3426         int ret;
3427
3428         struct ocfs2_xattr_info xi = {
3429                 .xi_name_index = name_index,
3430                 .xi_name = name,
3431                 .xi_name_len = strlen(name),
3432                 .xi_value = value,
3433                 .xi_value_len = value_len,
3434         };
3435
3436         struct ocfs2_xattr_search xis = {
3437                 .not_found = -ENODATA,
3438         };
3439
3440         struct ocfs2_xattr_search xbs = {
3441                 .not_found = -ENODATA,
3442         };
3443
3444         struct ocfs2_xattr_set_ctxt ctxt = {
3445                 .handle = handle,
3446                 .meta_ac = meta_ac,
3447                 .data_ac = data_ac,
3448         };
3449
3450         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3451                 return -EOPNOTSUPP;
3452
3453         /*
3454          * In extreme situation, may need xattr bucket when
3455          * block size is too small. And we have already reserved
3456          * the credits for bucket in mknod.
3457          */
3458         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3459                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3460                 if (!xbs.bucket) {
3461                         mlog_errno(-ENOMEM);
3462                         return -ENOMEM;
3463                 }
3464         }
3465
3466         xis.inode_bh = xbs.inode_bh = di_bh;
3467         di = (struct ocfs2_dinode *)di_bh->b_data;
3468
3469         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3470
3471         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3472         if (ret)
3473                 goto cleanup;
3474         if (xis.not_found) {
3475                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3476                 if (ret)
3477                         goto cleanup;
3478         }
3479
3480         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3481
3482 cleanup:
3483         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3484         brelse(xbs.xattr_bh);
3485         ocfs2_xattr_bucket_free(xbs.bucket);
3486
3487         return ret;
3488 }
3489
3490 /*
3491  * ocfs2_xattr_set()
3492  *
3493  * Set, replace or remove an extended attribute for this inode.
3494  * value is NULL to remove an existing extended attribute, else either
3495  * create or replace an extended attribute.
3496  */
3497 int ocfs2_xattr_set(struct inode *inode,
3498                     int name_index,
3499                     const char *name,
3500                     const void *value,
3501                     size_t value_len,
3502                     int flags)
3503 {
3504         struct buffer_head *di_bh = NULL;
3505         struct ocfs2_dinode *di;
3506         int ret, credits, ref_meta = 0, ref_credits = 0;
3507         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3508         struct inode *tl_inode = osb->osb_tl_inode;
3509         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3510         struct ocfs2_refcount_tree *ref_tree = NULL;
3511
3512         struct ocfs2_xattr_info xi = {
3513                 .xi_name_index = name_index,
3514                 .xi_name = name,
3515                 .xi_name_len = strlen(name),
3516                 .xi_value = value,
3517                 .xi_value_len = value_len,
3518         };
3519
3520         struct ocfs2_xattr_search xis = {
3521                 .not_found = -ENODATA,
3522         };
3523
3524         struct ocfs2_xattr_search xbs = {
3525                 .not_found = -ENODATA,
3526         };
3527
3528         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3529                 return -EOPNOTSUPP;
3530
3531         /*
3532          * Only xbs will be used on indexed trees.  xis doesn't need a
3533          * bucket.
3534          */
3535         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3536         if (!xbs.bucket) {
3537                 mlog_errno(-ENOMEM);
3538                 return -ENOMEM;
3539         }
3540
3541         ret = ocfs2_inode_lock(inode, &di_bh, 1);
3542         if (ret < 0) {
3543                 mlog_errno(ret);
3544                 goto cleanup_nolock;
3545         }
3546         xis.inode_bh = xbs.inode_bh = di_bh;
3547         di = (struct ocfs2_dinode *)di_bh->b_data;
3548
3549         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3550         /*
3551          * Scan inode and external block to find the same name
3552          * extended attribute and collect search information.
3553          */
3554         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3555         if (ret)
3556                 goto cleanup;
3557         if (xis.not_found) {
3558                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3559                 if (ret)
3560                         goto cleanup;
3561         }
3562
3563         if (xis.not_found && xbs.not_found) {
3564                 ret = -ENODATA;
3565                 if (flags & XATTR_REPLACE)
3566                         goto cleanup;
3567                 ret = 0;
3568                 if (!value)
3569                         goto cleanup;
3570         } else {
3571                 ret = -EEXIST;
3572                 if (flags & XATTR_CREATE)
3573                         goto cleanup;
3574         }
3575
3576         /* Check whether the value is refcounted and do some preparation. */
3577         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3578             (!xis.not_found || !xbs.not_found)) {
3579                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3580                                                    &xis, &xbs, &ref_tree,
3581                                                    &ref_meta, &ref_credits);
3582                 if (ret) {
3583                         mlog_errno(ret);
3584                         goto cleanup;
3585                 }
3586         }
3587
3588         mutex_lock(&tl_inode->i_mutex);
3589
3590         if (ocfs2_truncate_log_needs_flush(osb)) {
3591                 ret = __ocfs2_flush_truncate_log(osb);
3592                 if (ret < 0) {
3593                         mutex_unlock(&tl_inode->i_mutex);
3594                         mlog_errno(ret);
3595                         goto cleanup;
3596                 }
3597         }
3598         mutex_unlock(&tl_inode->i_mutex);
3599
3600         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3601                                         &xbs, &ctxt, ref_meta, &credits);
3602         if (ret) {
3603                 mlog_errno(ret);
3604                 goto cleanup;
3605         }
3606
3607         /* we need to update inode's ctime field, so add credit for it. */
3608         credits += OCFS2_INODE_UPDATE_CREDITS;
3609         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3610         if (IS_ERR(ctxt.handle)) {
3611                 ret = PTR_ERR(ctxt.handle);
3612                 mlog_errno(ret);
3613                 goto out_free_ac;
3614         }
3615
3616         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3617
3618         ocfs2_commit_trans(osb, ctxt.handle);
3619
3620 out_free_ac:
3621         if (ctxt.data_ac)
3622                 ocfs2_free_alloc_context(ctxt.data_ac);
3623         if (ctxt.meta_ac)
3624                 ocfs2_free_alloc_context(ctxt.meta_ac);
3625         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3626                 ocfs2_schedule_truncate_log_flush(osb, 1);
3627         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3628
3629 cleanup:
3630         if (ref_tree)
3631                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3632         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3633         if (!value && !ret) {
3634                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3635                 if (ret)
3636                         mlog_errno(ret);
3637         }
3638         ocfs2_inode_unlock(inode, 1);
3639 cleanup_nolock:
3640         brelse(di_bh);
3641         brelse(xbs.xattr_bh);
3642         ocfs2_xattr_bucket_free(xbs.bucket);
3643
3644         return ret;
3645 }
3646
3647 /*
3648  * Find the xattr extent rec which may contains name_hash.
3649  * e_cpos will be the first name hash of the xattr rec.
3650  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3651  */
3652 static int ocfs2_xattr_get_rec(struct inode *inode,
3653                                u32 name_hash,
3654                                u64 *p_blkno,
3655                                u32 *e_cpos,
3656                                u32 *num_clusters,
3657                                struct ocfs2_extent_list *el)
3658 {
3659         int ret = 0, i;
3660         struct buffer_head *eb_bh = NULL;
3661         struct ocfs2_extent_block *eb;
3662         struct ocfs2_extent_rec *rec = NULL;
3663         u64 e_blkno = 0;
3664
3665         if (el->l_tree_depth) {
3666                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3667                                       &eb_bh);
3668                 if (ret) {
3669                         mlog_errno(ret);
3670                         goto out;
3671                 }
3672
3673                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3674                 el = &eb->h_list;
3675
3676                 if (el->l_tree_depth) {
3677                         ocfs2_error(inode->i_sb,
3678                                     "Inode %lu has non zero tree depth in "
3679                                     "xattr tree block %llu\n", inode->i_ino,
3680                                     (unsigned long long)eb_bh->b_blocknr);
3681                         ret = -EROFS;
3682                         goto out;
3683                 }
3684         }
3685
3686         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3687                 rec = &el->l_recs[i];
3688
3689                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3690                         e_blkno = le64_to_cpu(rec->e_blkno);
3691                         break;
3692                 }
3693         }
3694
3695         if (!e_blkno) {
3696                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3697                             "record (%u, %u, 0) in xattr", inode->i_ino,
3698                             le32_to_cpu(rec->e_cpos),
3699                             ocfs2_rec_clusters(el, rec));
3700                 ret = -EROFS;
3701                 goto out;
3702         }
3703
3704         *p_blkno = le64_to_cpu(rec->e_blkno);
3705         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3706         if (e_cpos)
3707                 *e_cpos = le32_to_cpu(rec->e_cpos);
3708 out:
3709         brelse(eb_bh);
3710         return ret;
3711 }
3712
3713 typedef int (xattr_bucket_func)(struct inode *inode,
3714                                 struct ocfs2_xattr_bucket *bucket,
3715                                 void *para);
3716
3717 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3718                                    struct ocfs2_xattr_bucket *bucket,
3719                                    int name_index,
3720                                    const char *name,
3721                                    u32 name_hash,
3722                                    u16 *xe_index,
3723                                    int *found)
3724 {
3725         int i, ret = 0, cmp = 1, block_off, new_offset;
3726         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3727         size_t name_len = strlen(name);
3728         struct ocfs2_xattr_entry *xe = NULL;
3729         char *xe_name;
3730
3731         /*
3732          * We don't use binary search in the bucket because there
3733          * may be multiple entries with the same name hash.
3734          */
3735         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3736                 xe = &xh->xh_entries[i];
3737
3738                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3739                         continue;
3740                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3741                         break;
3742
3743                 cmp = name_index - ocfs2_xattr_get_type(xe);
3744                 if (!cmp)
3745                         cmp = name_len - xe->xe_name_len;
3746                 if (cmp)
3747                         continue;
3748
3749                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3750                                                         xh,
3751                                                         i,
3752                                                         &block_off,
3753                                                         &new_offset);
3754                 if (ret) {
3755                         mlog_errno(ret);
3756                         break;
3757                 }
3758
3759
3760                 xe_name = bucket_block(bucket, block_off) + new_offset;
3761                 if (!memcmp(name, xe_name, name_len)) {
3762                         *xe_index = i;
3763                         *found = 1;
3764                         ret = 0;
3765                         break;
3766                 }
3767         }
3768
3769         return ret;
3770 }
3771
3772 /*
3773  * Find the specified xattr entry in a series of buckets.
3774  * This series start from p_blkno and last for num_clusters.
3775  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3776  * the num of the valid buckets.
3777  *
3778  * Return the buffer_head this xattr should reside in. And if the xattr's
3779  * hash is in the gap of 2 buckets, return the lower bucket.
3780  */
3781 static int ocfs2_xattr_bucket_find(struct inode *inode,
3782                                    int name_index,
3783                                    const char *name,
3784                                    u32 name_hash,
3785                                    u64 p_blkno,
3786                                    u32 first_hash,
3787                                    u32 num_clusters,
3788                                    struct ocfs2_xattr_search *xs)
3789 {
3790         int ret, found = 0;
3791         struct ocfs2_xattr_header *xh = NULL;
3792         struct ocfs2_xattr_entry *xe = NULL;
3793         u16 index = 0;
3794         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3795         int low_bucket = 0, bucket, high_bucket;
3796         struct ocfs2_xattr_bucket *search;
3797         u32 last_hash;
3798         u64 blkno, lower_blkno = 0;
3799
3800         search = ocfs2_xattr_bucket_new(inode);
3801         if (!search) {
3802                 ret = -ENOMEM;
3803                 mlog_errno(ret);
3804                 goto out;
3805         }
3806
3807         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3808         if (ret) {
3809                 mlog_errno(ret);
3810                 goto out;
3811         }
3812
3813         xh = bucket_xh(search);
3814         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3815         while (low_bucket <= high_bucket) {
3816                 ocfs2_xattr_bucket_relse(search);
3817
3818                 bucket = (low_bucket + high_bucket) / 2;
3819                 blkno = p_blkno + bucket * blk_per_bucket;
3820                 ret = ocfs2_read_xattr_bucket(search, blkno);
3821                 if (ret) {
3822                         mlog_errno(ret);
3823                         goto out;
3824                 }
3825
3826                 xh = bucket_xh(search);
3827                 xe = &xh->xh_entries[0];
3828                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3829                         high_bucket = bucket - 1;
3830                         continue;
3831                 }
3832
3833                 /*
3834                  * Check whether the hash of the last entry in our
3835                  * bucket is larger than the search one. for an empty
3836                  * bucket, the last one is also the first one.
3837                  */
3838                 if (xh->xh_count)
3839                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3840
3841                 last_hash = le32_to_cpu(xe->xe_name_hash);
3842
3843                 /* record lower_blkno which may be the insert place. */
3844                 lower_blkno = blkno;
3845
3846                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3847                         low_bucket = bucket + 1;
3848                         continue;
3849                 }
3850
3851                 /* the searched xattr should reside in this bucket if exists. */
3852                 ret = ocfs2_find_xe_in_bucket(inode, search,
3853                                               name_index, name, name_hash,
3854                                               &index, &found);
3855                 if (ret) {
3856                         mlog_errno(ret);
3857                         goto out;
3858                 }
3859                 break;
3860         }
3861
3862         /*
3863          * Record the bucket we have found.
3864          * When the xattr's hash value is in the gap of 2 buckets, we will
3865          * always set it to the previous bucket.
3866          */
3867         if (!lower_blkno)
3868                 lower_blkno = p_blkno;
3869
3870         /* This should be in cache - we just read it during the search */
3871         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3872         if (ret) {
3873                 mlog_errno(ret);
3874                 goto out;
3875         }
3876
3877         xs->header = bucket_xh(xs->bucket);
3878         xs->base = bucket_block(xs->bucket, 0);
3879         xs->end = xs->base + inode->i_sb->s_blocksize;
3880
3881         if (found) {
3882                 xs->here = &xs->header->xh_entries[index];
3883                 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3884                         name, name_index, name_hash,
3885                         (unsigned long long)bucket_blkno(xs->bucket),
3886                         index);
3887         } else
3888                 ret = -ENODATA;
3889
3890 out:
3891         ocfs2_xattr_bucket_free(search);
3892         return ret;
3893 }
3894
3895 static int ocfs2_xattr_index_block_find(struct inode *inode,
3896                                         struct buffer_head *root_bh,
3897                                         int name_index,
3898                                         const char *name,
3899                                         struct ocfs2_xattr_search *xs)
3900 {
3901         int ret;
3902         struct ocfs2_xattr_block *xb =
3903                         (struct ocfs2_xattr_block *)root_bh->b_data;
3904         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3905         struct ocfs2_extent_list *el = &xb_root->xt_list;
3906         u64 p_blkno = 0;
3907         u32 first_hash, num_clusters = 0;
3908         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3909
3910         if (le16_to_cpu(el->l_next_free_rec) == 0)
3911                 return -ENODATA;
3912
3913         trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3914                                         name, name_index, name_hash,
3915                                         (unsigned long long)root_bh->b_blocknr,
3916                                         -1);
3917
3918         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3919                                   &num_clusters, el);
3920         if (ret) {
3921                 mlog_errno(ret);
3922                 goto out;
3923         }
3924
3925         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3926
3927         trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3928                                         name, name_index, first_hash,
3929                                         (unsigned long long)p_blkno,
3930                                         num_clusters);
3931
3932         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3933                                       p_blkno, first_hash, num_clusters, xs);
3934
3935 out:
3936         return ret;
3937 }
3938
3939 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3940                                        u64 blkno,
3941                                        u32 clusters,
3942                                        xattr_bucket_func *func,
3943                                        void *para)
3944 {
3945         int i, ret = 0;
3946         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3947         u32 num_buckets = clusters * bpc;
3948         struct ocfs2_xattr_bucket *bucket;
3949
3950         bucket = ocfs2_xattr_bucket_new(inode);
3951         if (!bucket) {
3952                 mlog_errno(-ENOMEM);
3953                 return -ENOMEM;
3954         }
3955
3956         trace_ocfs2_iterate_xattr_buckets(
3957                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3958                 (unsigned long long)blkno, clusters);
3959
3960         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3961                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3962                 if (ret) {
3963                         mlog_errno(ret);
3964                         break;
3965                 }
3966
3967                 /*
3968                  * The real bucket num in this series of blocks is stored
3969                  * in the 1st bucket.
3970                  */
3971                 if (i == 0)
3972                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3973
3974                 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
3975                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3976                 if (func) {
3977                         ret = func(inode, bucket, para);
3978                         if (ret && ret != -ERANGE)
3979                                 mlog_errno(ret);
3980                         /* Fall through to bucket_relse() */
3981                 }
3982
3983                 ocfs2_xattr_bucket_relse(bucket);
3984                 if (ret)
3985                         break;
3986         }
3987
3988         ocfs2_xattr_bucket_free(bucket);
3989         return ret;
3990 }
3991
3992 struct ocfs2_xattr_tree_list {
3993         char *buffer;
3994         size_t buffer_size;
3995         size_t result;
3996 };
3997
3998 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3999                                              struct ocfs2_xattr_header *xh,
4000                                              int index,
4001                                              int *block_off,
4002                                              int *new_offset)
4003 {
4004         u16 name_offset;
4005
4006         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4007                 return -EINVAL;
4008
4009         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4010
4011         *block_off = name_offset >> sb->s_blocksize_bits;
4012         *new_offset = name_offset % sb->s_blocksize;
4013
4014         return 0;
4015 }
4016
4017 static int ocfs2_list_xattr_bucket(struct inode *inode,
4018                                    struct ocfs2_xattr_bucket *bucket,
4019                                    void *para)
4020 {
4021         int ret = 0, type;
4022         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4023         int i, block_off, new_offset;
4024         const char *prefix, *name;
4025
4026         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4027                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4028                 type = ocfs2_xattr_get_type(entry);
4029                 prefix = ocfs2_xattr_prefix(type);
4030
4031                 if (prefix) {
4032                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4033                                                                 bucket_xh(bucket),
4034                                                                 i,
4035                                                                 &block_off,
4036                                                                 &new_offset);
4037                         if (ret)
4038                                 break;
4039
4040                         name = (const char *)bucket_block(bucket, block_off) +
4041                                 new_offset;
4042                         ret = ocfs2_xattr_list_entry(xl->buffer,
4043                                                      xl->buffer_size,
4044                                                      &xl->result,
4045                                                      prefix, name,
4046                                                      entry->xe_name_len);
4047                         if (ret)
4048                                 break;
4049                 }
4050         }
4051
4052         return ret;
4053 }
4054
4055 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4056                                            struct buffer_head *blk_bh,
4057                                            xattr_tree_rec_func *rec_func,
4058                                            void *para)
4059 {
4060         struct ocfs2_xattr_block *xb =
4061                         (struct ocfs2_xattr_block *)blk_bh->b_data;
4062         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4063         int ret = 0;
4064         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4065         u64 p_blkno = 0;
4066
4067         if (!el->l_next_free_rec || !rec_func)
4068                 return 0;
4069
4070         while (name_hash > 0) {
4071                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4072                                           &e_cpos, &num_clusters, el);
4073                 if (ret) {
4074                         mlog_errno(ret);
4075                         break;
4076                 }
4077
4078                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4079                                num_clusters, para);
4080                 if (ret) {
4081                         if (ret != -ERANGE)
4082                                 mlog_errno(ret);
4083                         break;
4084                 }
4085
4086                 if (e_cpos == 0)
4087                         break;
4088
4089                 name_hash = e_cpos - 1;
4090         }
4091
4092         return ret;
4093
4094 }
4095
4096 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4097                                      struct buffer_head *root_bh,
4098                                      u64 blkno, u32 cpos, u32 len, void *para)
4099 {
4100         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4101                                            ocfs2_list_xattr_bucket, para);
4102 }
4103
4104 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4105                                              struct buffer_head *blk_bh,
4106                                              char *buffer,
4107                                              size_t buffer_size)
4108 {
4109         int ret;
4110         struct ocfs2_xattr_tree_list xl = {
4111                 .buffer = buffer,
4112                 .buffer_size = buffer_size,
4113                 .result = 0,
4114         };
4115
4116         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4117                                               ocfs2_list_xattr_tree_rec, &xl);
4118         if (ret) {
4119                 mlog_errno(ret);
4120                 goto out;
4121         }
4122
4123         ret = xl.result;
4124 out:
4125         return ret;
4126 }
4127
4128 static int cmp_xe(const void *a, const void *b)
4129 {
4130         const struct ocfs2_xattr_entry *l = a, *r = b;
4131         u32 l_hash = le32_to_cpu(l->xe_name_hash);
4132         u32 r_hash = le32_to_cpu(r->xe_name_hash);
4133
4134         if (l_hash > r_hash)
4135                 return 1;
4136         if (l_hash < r_hash)
4137                 return -1;
4138         return 0;
4139 }
4140
4141 static void swap_xe(void *a, void *b, int size)
4142 {
4143         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4144
4145         tmp = *l;
4146         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4147         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4148 }
4149
4150 /*
4151  * When the ocfs2_xattr_block is filled up, new bucket will be created
4152  * and all the xattr entries will be moved to the new bucket.
4153  * The header goes at the start of the bucket, and the names+values are
4154  * filled from the end.  This is why *target starts as the last buffer.
4155  * Note: we need to sort the entries since they are not saved in order
4156  * in the ocfs2_xattr_block.
4157  */
4158 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4159                                            struct buffer_head *xb_bh,
4160                                            struct ocfs2_xattr_bucket *bucket)
4161 {
4162         int i, blocksize = inode->i_sb->s_blocksize;
4163         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4164         u16 offset, size, off_change;
4165         struct ocfs2_xattr_entry *xe;
4166         struct ocfs2_xattr_block *xb =
4167                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
4168         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4169         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4170         u16 count = le16_to_cpu(xb_xh->xh_count);
4171         char *src = xb_bh->b_data;
4172         char *target = bucket_block(bucket, blks - 1);
4173
4174         trace_ocfs2_cp_xattr_block_to_bucket_begin(
4175                                 (unsigned long long)xb_bh->b_blocknr,
4176                                 (unsigned long long)bucket_blkno(bucket));
4177
4178         for (i = 0; i < blks; i++)
4179                 memset(bucket_block(bucket, i), 0, blocksize);
4180
4181         /*
4182          * Since the xe_name_offset is based on ocfs2_xattr_header,
4183          * there is a offset change corresponding to the change of
4184          * ocfs2_xattr_header's position.
4185          */
4186         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4187         xe = &xb_xh->xh_entries[count - 1];
4188         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4189         size = blocksize - offset;
4190
4191         /* copy all the names and values. */
4192         memcpy(target + offset, src + offset, size);
4193
4194         /* Init new header now. */
4195         xh->xh_count = xb_xh->xh_count;
4196         xh->xh_num_buckets = cpu_to_le16(1);
4197         xh->xh_name_value_len = cpu_to_le16(size);
4198         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4199
4200         /* copy all the entries. */
4201         target = bucket_block(bucket, 0);
4202         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4203         size = count * sizeof(struct ocfs2_xattr_entry);
4204         memcpy(target + offset, (char *)xb_xh + offset, size);
4205
4206         /* Change the xe offset for all the xe because of the move. */
4207         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4208                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4209         for (i = 0; i < count; i++)
4210                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4211
4212         trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4213
4214         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4215              cmp_xe, swap_xe);
4216 }
4217
4218 /*
4219  * After we move xattr from block to index btree, we have to
4220  * update ocfs2_xattr_search to the new xe and base.
4221  *
4222  * When the entry is in xattr block, xattr_bh indicates the storage place.
4223  * While if the entry is in index b-tree, "bucket" indicates the
4224  * real place of the xattr.
4225  */
4226 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4227                                             struct ocfs2_xattr_search *xs,
4228                                             struct buffer_head *old_bh)
4229 {
4230         char *buf = old_bh->b_data;
4231         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4232         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4233         int i;
4234
4235         xs->header = bucket_xh(xs->bucket);
4236         xs->base = bucket_block(xs->bucket, 0);
4237         xs->end = xs->base + inode->i_sb->s_blocksize;
4238
4239         if (xs->not_found)
4240                 return;
4241
4242         i = xs->here - old_xh->xh_entries;
4243         xs->here = &xs->header->xh_entries[i];
4244 }
4245
4246 static int ocfs2_xattr_create_index_block(struct inode *inode,
4247                                           struct ocfs2_xattr_search *xs,
4248                                           struct ocfs2_xattr_set_ctxt *ctxt)
4249 {
4250         int ret;
4251         u32 bit_off, len;
4252         u64 blkno;
4253         handle_t *handle = ctxt->handle;
4254         struct ocfs2_inode_info *oi = OCFS2_I(inode);
4255         struct buffer_head *xb_bh = xs->xattr_bh;
4256         struct ocfs2_xattr_block *xb =
4257                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4258         struct ocfs2_xattr_tree_root *xr;
4259         u16 xb_flags = le16_to_cpu(xb->xb_flags);
4260
4261         trace_ocfs2_xattr_create_index_block_begin(
4262                                 (unsigned long long)xb_bh->b_blocknr);
4263
4264         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4265         BUG_ON(!xs->bucket);
4266
4267         /*
4268          * XXX:
4269          * We can use this lock for now, and maybe move to a dedicated mutex
4270          * if performance becomes a problem later.
4271          */
4272         down_write(&oi->ip_alloc_sem);
4273
4274         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4275                                       OCFS2_JOURNAL_ACCESS_WRITE);
4276         if (ret) {
4277                 mlog_errno(ret);
4278                 goto out;
4279         }
4280
4281         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4282                                      1, 1, &bit_off, &len);
4283         if (ret) {
4284                 mlog_errno(ret);
4285                 goto out;
4286         }
4287
4288         /*
4289          * The bucket may spread in many blocks, and
4290          * we will only touch the 1st block and the last block
4291          * in the whole bucket(one for entry and one for data).
4292          */
4293         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4294
4295         trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4296
4297         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
4298         if (ret) {
4299                 mlog_errno(ret);
4300                 goto out;
4301         }
4302
4303         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4304                                                 OCFS2_JOURNAL_ACCESS_CREATE);
4305         if (ret) {
4306                 mlog_errno(ret);
4307                 goto out;
4308         }
4309
4310         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4311         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4312
4313         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4314
4315         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4316         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4317                offsetof(struct ocfs2_xattr_block, xb_attrs));
4318
4319         xr = &xb->xb_attrs.xb_root;
4320         xr->xt_clusters = cpu_to_le32(1);
4321         xr->xt_last_eb_blk = 0;
4322         xr->xt_list.l_tree_depth = 0;
4323         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4324         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4325
4326         xr->xt_list.l_recs[0].e_cpos = 0;
4327         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4328         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4329
4330         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4331
4332         ocfs2_journal_dirty(handle, xb_bh);
4333
4334 out:
4335         up_write(&oi->ip_alloc_sem);
4336
4337         return ret;
4338 }
4339
4340 static int cmp_xe_offset(const void *a, const void *b)
4341 {
4342         const struct ocfs2_xattr_entry *l = a, *r = b;
4343         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4344         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4345
4346         if (l_name_offset < r_name_offset)
4347                 return 1;
4348         if (l_name_offset > r_name_offset)
4349                 return -1;
4350         return 0;
4351 }
4352
4353 /*
4354  * defrag a xattr bucket if we find that the bucket has some
4355  * holes beteen name/value pairs.
4356  * We will move all the name/value pairs to the end of the bucket
4357  * so that we can spare some space for insertion.
4358  */
4359 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4360                                      handle_t *handle,
4361                                      struct ocfs2_xattr_bucket *bucket)
4362 {
4363         int ret, i;
4364         size_t end, offset, len;
4365         struct ocfs2_xattr_header *xh;
4366         char *entries, *buf, *bucket_buf = NULL;
4367         u64 blkno = bucket_blkno(bucket);
4368         u16 xh_free_start;
4369         size_t blocksize = inode->i_sb->s_blocksize;
4370         struct ocfs2_xattr_entry *xe;
4371
4372         /*
4373          * In order to make the operation more efficient and generic,
4374          * we copy all the blocks into a contiguous memory and do the
4375          * defragment there, so if anything is error, we will not touch
4376          * the real block.
4377          */
4378         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4379         if (!bucket_buf) {
4380                 ret = -EIO;
4381                 goto out;
4382         }
4383
4384         buf = bucket_buf;
4385         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4386                 memcpy(buf, bucket_block(bucket, i), blocksize);
4387
4388         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4389                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4390         if (ret < 0) {
4391                 mlog_errno(ret);
4392                 goto out;
4393         }
4394
4395         xh = (struct ocfs2_xattr_header *)bucket_buf;
4396         entries = (char *)xh->xh_entries;
4397         xh_free_start = le16_to_cpu(xh->xh_free_start);
4398
4399         trace_ocfs2_defrag_xattr_bucket(
4400              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4401              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4402
4403         /*
4404          * sort all the entries by their offset.
4405          * the largest will be the first, so that we can
4406          * move them to the end one by one.
4407          */
4408         sort(entries, le16_to_cpu(xh->xh_count),
4409              sizeof(struct ocfs2_xattr_entry),
4410              cmp_xe_offset, swap_xe);
4411
4412         /* Move all name/values to the end of the bucket. */
4413         xe = xh->xh_entries;
4414         end = OCFS2_XATTR_BUCKET_SIZE;
4415         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4416                 offset = le16_to_cpu(xe->xe_name_offset);
4417                 len = namevalue_size_xe(xe);
4418
4419                 /*
4420                  * We must make sure that the name/value pair
4421                  * exist in the same block. So adjust end to
4422                  * the previous block end if needed.
4423                  */
4424                 if (((end - len) / blocksize !=
4425                         (end - 1) / blocksize))
4426                         end = end - end % blocksize;
4427
4428                 if (end > offset + len) {
4429                         memmove(bucket_buf + end - len,
4430                                 bucket_buf + offset, len);
4431                         xe->xe_name_offset = cpu_to_le16(end - len);
4432                 }
4433
4434                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4435                                 "bucket %llu\n", (unsigned long long)blkno);
4436
4437                 end -= len;
4438         }
4439
4440         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4441                         "bucket %llu\n", (unsigned long long)blkno);
4442
4443         if (xh_free_start == end)
4444                 goto out;
4445
4446         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4447         xh->xh_free_start = cpu_to_le16(end);
4448
4449         /* sort the entries by their name_hash. */
4450         sort(entries, le16_to_cpu(xh->xh_count),
4451              sizeof(struct ocfs2_xattr_entry),
4452              cmp_xe, swap_xe);
4453
4454         buf = bucket_buf;
4455         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4456                 memcpy(bucket_block(bucket, i), buf, blocksize);
4457         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4458
4459 out:
4460         kfree(bucket_buf);
4461         return ret;
4462 }
4463
4464 /*
4465  * prev_blkno points to the start of an existing extent.  new_blkno
4466  * points to a newly allocated extent.  Because we know each of our
4467  * clusters contains more than bucket, we can easily split one cluster
4468  * at a bucket boundary.  So we take the last cluster of the existing
4469  * extent and split it down the middle.  We move the last half of the
4470  * buckets in the last cluster of the existing extent over to the new
4471  * extent.
4472  *
4473  * first_bh is the buffer at prev_blkno so we can update the existing
4474  * extent's bucket count.  header_bh is the bucket were we were hoping
4475  * to insert our xattr.  If the bucket move places the target in the new
4476  * extent, we'll update first_bh and header_bh after modifying the old
4477  * extent.
4478  *
4479  * first_hash will be set as the 1st xe's name_hash in the new extent.
4480  */
4481 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4482                                                handle_t *handle,
4483                                                struct ocfs2_xattr_bucket *first,
4484                                                struct ocfs2_xattr_bucket *target,
4485                                                u64 new_blkno,
4486                                                u32 num_clusters,
4487                                                u32 *first_hash)
4488 {
4489         int ret;
4490         struct super_block *sb = inode->i_sb;
4491         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4492         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4493         int to_move = num_buckets / 2;
4494         u64 src_blkno;
4495         u64 last_cluster_blkno = bucket_blkno(first) +
4496                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4497
4498         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4499         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4500
4501         trace_ocfs2_mv_xattr_bucket_cross_cluster(
4502                                 (unsigned long long)last_cluster_blkno,
4503                                 (unsigned long long)new_blkno);
4504
4505         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4506                                      last_cluster_blkno, new_blkno,
4507                                      to_move, first_hash);
4508         if (ret) {
4509                 mlog_errno(ret);
4510                 goto out;
4511         }
4512
4513         /* This is the first bucket that got moved */
4514         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4515
4516         /*
4517          * If the target bucket was part of the moved buckets, we need to
4518          * update first and target.
4519          */
4520         if (bucket_blkno(target) >= src_blkno) {
4521                 /* Find the block for the new target bucket */
4522                 src_blkno = new_blkno +
4523                         (bucket_blkno(target) - src_blkno);
4524
4525                 ocfs2_xattr_bucket_relse(first);
4526                 ocfs2_xattr_bucket_relse(target);
4527
4528                 /*
4529                  * These shouldn't fail - the buffers are in the
4530                  * journal from ocfs2_cp_xattr_bucket().
4531                  */
4532                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4533                 if (ret) {
4534                         mlog_errno(ret);
4535                         goto out;
4536                 }
4537                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4538                 if (ret)
4539                         mlog_errno(ret);
4540
4541         }
4542
4543 out:
4544         return ret;
4545 }
4546
4547 /*
4548  * Find the suitable pos when we divide a bucket into 2.
4549  * We have to make sure the xattrs with the same hash value exist
4550  * in the same bucket.
4551  *
4552  * If this ocfs2_xattr_header covers more than one hash value, find a
4553  * place where the hash value changes.  Try to find the most even split.
4554  * The most common case is that all entries have different hash values,
4555  * and the first check we make will find a place to split.
4556  */
4557 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4558 {
4559         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4560         int count = le16_to_cpu(xh->xh_count);
4561         int delta, middle = count / 2;
4562
4563         /*
4564          * We start at the middle.  Each step gets farther away in both
4565          * directions.  We therefore hit the change in hash value
4566          * nearest to the middle.  Note that this loop does not execute for
4567          * count < 2.
4568          */
4569         for (delta = 0; delta < middle; delta++) {
4570                 /* Let's check delta earlier than middle */
4571                 if (cmp_xe(&entries[middle - delta - 1],
4572                            &entries[middle - delta]))
4573                         return middle - delta;
4574
4575                 /* For even counts, don't walk off the end */
4576                 if ((middle + delta + 1) == count)
4577                         continue;
4578
4579                 /* Now try delta past middle */
4580                 if (cmp_xe(&entries[middle + delta],
4581                            &entries[middle + delta + 1]))
4582                         return middle + delta + 1;
4583         }
4584
4585         /* Every entry had the same hash */
4586         return count;
4587 }
4588
4589 /*
4590  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4591  * first_hash will record the 1st hash of the new bucket.
4592  *
4593  * Normally half of the xattrs will be moved.  But we have to make
4594  * sure that the xattrs with the same hash value are stored in the
4595  * same bucket. If all the xattrs in this bucket have the same hash
4596  * value, the new bucket will be initialized as an empty one and the
4597  * first_hash will be initialized as (hash_value+1).
4598  */
4599 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4600                                     handle_t *handle,
4601                                     u64 blk,
4602                                     u64 new_blk,
4603                                     u32 *first_hash,
4604                                     int new_bucket_head)
4605 {
4606         int ret, i;
4607         int count, start, len, name_value_len = 0, name_offset = 0;
4608         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4609         struct ocfs2_xattr_header *xh;
4610         struct ocfs2_xattr_entry *xe;
4611         int blocksize = inode->i_sb->s_blocksize;
4612
4613         trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4614                                               (unsigned long long)new_blk);
4615
4616         s_bucket = ocfs2_xattr_bucket_new(inode);
4617         t_bucket = ocfs2_xattr_bucket_new(inode);
4618         if (!s_bucket || !t_bucket) {
4619                 ret = -ENOMEM;
4620                 mlog_errno(ret);
4621                 goto out;
4622         }
4623
4624         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4625         if (ret) {
4626                 mlog_errno(ret);
4627                 goto out;
4628         }
4629
4630         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4631                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4632         if (ret) {
4633                 mlog_errno(ret);
4634                 goto out;
4635         }
4636
4637         /*
4638          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4639          * there's no need to read it.
4640          */
4641         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4642         if (ret) {
4643                 mlog_errno(ret);
4644                 goto out;
4645         }
4646
4647         /*
4648          * Hey, if we're overwriting t_bucket, what difference does
4649          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4650          * same part of ocfs2_cp_xattr_bucket().
4651          */
4652         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4653                                                 new_bucket_head ?
4654                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4655                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4656         if (ret) {
4657                 mlog_errno(ret);
4658                 goto out;
4659         }
4660
4661         xh = bucket_xh(s_bucket);
4662         count = le16_to_cpu(xh->xh_count);
4663         start = ocfs2_xattr_find_divide_pos(xh);
4664
4665         if (start == count) {
4666                 xe = &xh->xh_entries[start-1];
4667
4668                 /*
4669                  * initialized a new empty bucket here.
4670                  * The hash value is set as one larger than
4671                  * that of the last entry in the previous bucket.
4672                  */
4673                 for (i = 0; i < t_bucket->bu_blocks; i++)
4674                         memset(bucket_block(t_bucket, i), 0, blocksize);
4675
4676                 xh = bucket_xh(t_bucket);
4677                 xh->xh_free_start = cpu_to_le16(blocksize);
4678                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4679                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4680
4681                 goto set_num_buckets;
4682         }
4683
4684         /* copy the whole bucket to the new first. */
4685         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4686
4687         /* update the new bucket. */
4688         xh = bucket_xh(t_bucket);
4689
4690         /*
4691          * Calculate the total name/value len and xh_free_start for
4692          * the old bucket first.
4693          */
4694         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4695         name_value_len = 0;
4696         for (i = 0; i < start; i++) {
4697                 xe = &xh->xh_entries[i];
4698                 name_value_len += namevalue_size_xe(xe);
4699                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4700                         name_offset = le16_to_cpu(xe->xe_name_offset);
4701         }
4702
4703         /*
4704          * Now begin the modification to the new bucket.
4705          *
4706          * In the new bucket, We just move the xattr entry to the beginning
4707          * and don't touch the name/value. So there will be some holes in the
4708          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4709          * called.
4710          */
4711         xe = &xh->xh_entries[start];
4712         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4713         trace_ocfs2_divide_xattr_bucket_move(len,
4714                         (int)((char *)xe - (char *)xh),
4715                         (int)((char *)xh->xh_entries - (char *)xh));
4716         memmove((char *)xh->xh_entries, (char *)xe, len);
4717         xe = &xh->xh_entries[count - start];
4718         len = sizeof(struct ocfs2_xattr_entry) * start;
4719         memset((char *)xe, 0, len);
4720
4721         le16_add_cpu(&xh->xh_count, -start);
4722         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4723
4724         /* Calculate xh_free_start for the new bucket. */
4725         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4726         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4727                 xe = &xh->xh_entries[i];
4728                 if (le16_to_cpu(xe->xe_name_offset) <
4729                     le16_to_cpu(xh->xh_free_start))
4730                         xh->xh_free_start = xe->xe_name_offset;
4731         }
4732
4733 set_num_buckets:
4734         /* set xh->xh_num_buckets for the new xh. */
4735         if (new_bucket_head)
4736                 xh->xh_num_buckets = cpu_to_le16(1);
4737         else
4738                 xh->xh_num_buckets = 0;
4739
4740         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4741
4742         /* store the first_hash of the new bucket. */
4743         if (first_hash)
4744                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4745
4746         /*
4747          * Now only update the 1st block of the old bucket.  If we
4748          * just added a new empty bucket, there is no need to modify
4749          * it.
4750          */
4751         if (start == count)
4752                 goto out;
4753
4754         xh = bucket_xh(s_bucket);
4755         memset(&xh->xh_entries[start], 0,
4756                sizeof(struct ocfs2_xattr_entry) * (count - start));
4757         xh->xh_count = cpu_to_le16(start);
4758         xh->xh_free_start = cpu_to_le16(name_offset);
4759         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4760
4761         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4762
4763 out:
4764         ocfs2_xattr_bucket_free(s_bucket);
4765         ocfs2_xattr_bucket_free(t_bucket);
4766
4767         return ret;
4768 }
4769
4770 /*
4771  * Copy xattr from one bucket to another bucket.
4772  *
4773  * The caller must make sure that the journal transaction
4774  * has enough space for journaling.
4775  */
4776 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4777                                  handle_t *handle,
4778                                  u64 s_blkno,
4779                                  u64 t_blkno,
4780                                  int t_is_new)
4781 {
4782         int ret;
4783         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4784
4785         BUG_ON(s_blkno == t_blkno);
4786
4787         trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4788                                     (unsigned long long)t_blkno,
4789                                     t_is_new);
4790
4791         s_bucket = ocfs2_xattr_bucket_new(inode);
4792         t_bucket = ocfs2_xattr_bucket_new(inode);
4793         if (!s_bucket || !t_bucket) {
4794                 ret = -ENOMEM;
4795                 mlog_errno(ret);
4796                 goto out;
4797         }
4798
4799         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4800         if (ret)
4801                 goto out;
4802
4803         /*
4804          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4805          * there's no need to read it.
4806          */
4807         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4808         if (ret)
4809                 goto out;
4810
4811         /*
4812          * Hey, if we're overwriting t_bucket, what difference does
4813          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4814          * cluster to fill, we came here from
4815          * ocfs2_mv_xattr_buckets(), and it is really new -
4816          * ACCESS_CREATE is required.  But we also might have moved data
4817          * out of t_bucket before extending back into it.
4818          * ocfs2_add_new_xattr_bucket() can do this - its call to
4819          * ocfs2_add_new_xattr_cluster() may have created a new extent
4820          * and copied out the end of the old extent.  Then it re-extends
4821          * the old extent back to create space for new xattrs.  That's
4822          * how we get here, and the bucket isn't really new.
4823          */
4824         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4825                                                 t_is_new ?
4826                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4827                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4828         if (ret)
4829                 goto out;
4830
4831         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4832         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4833
4834 out:
4835         ocfs2_xattr_bucket_free(t_bucket);
4836         ocfs2_xattr_bucket_free(s_bucket);
4837
4838         return ret;
4839 }
4840
4841 /*
4842  * src_blk points to the start of an existing extent.  last_blk points to
4843  * last cluster in that extent.  to_blk points to a newly allocated
4844  * extent.  We copy the buckets from the cluster at last_blk to the new
4845  * extent.  If start_bucket is non-zero, we skip that many buckets before
4846  * we start copying.  The new extent's xh_num_buckets gets set to the
4847  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4848  * by the same amount.
4849  */
4850 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4851                                   u64 src_blk, u64 last_blk, u64 to_blk,
4852                                   unsigned int start_bucket,
4853                                   u32 *first_hash)
4854 {
4855         int i, ret, credits;
4856         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4857         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4858         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4859         struct ocfs2_xattr_bucket *old_first, *new_first;
4860
4861         trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4862                                      (unsigned long long)to_blk);
4863
4864         BUG_ON(start_bucket >= num_buckets);
4865         if (start_bucket) {
4866                 num_buckets -= start_bucket;
4867                 last_blk += (start_bucket * blks_per_bucket);
4868         }
4869
4870         /* The first bucket of the original extent */
4871         old_first = ocfs2_xattr_bucket_new(inode);
4872         /* The first bucket of the new extent */
4873         new_first = ocfs2_xattr_bucket_new(inode);
4874         if (!old_first || !new_first) {
4875                 ret = -ENOMEM;
4876                 mlog_errno(ret);
4877                 goto out;
4878         }
4879
4880         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4881         if (ret) {
4882                 mlog_errno(ret);
4883                 goto out;
4884         }
4885
4886         /*
4887          * We need to update the first bucket of the old extent and all
4888          * the buckets going to the new extent.
4889          */
4890         credits = ((num_buckets + 1) * blks_per_bucket);
4891         ret = ocfs2_extend_trans(handle, credits);
4892         if (ret) {
4893                 mlog_errno(ret);
4894                 goto out;
4895         }
4896
4897         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4898                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4899         if (ret) {
4900                 mlog_errno(ret);
4901                 goto out;
4902         }
4903
4904         for (i = 0; i < num_buckets; i++) {
4905                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4906                                             last_blk + (i * blks_per_bucket),
4907                                             to_blk + (i * blks_per_bucket),
4908                                             1);
4909                 if (ret) {
4910                         mlog_errno(ret);
4911                         goto out;
4912                 }
4913         }
4914
4915         /*
4916          * Get the new bucket ready before we dirty anything
4917          * (This actually shouldn't fail, because we already dirtied
4918          * it once in ocfs2_cp_xattr_bucket()).
4919          */
4920         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4921         if (ret) {
4922                 mlog_errno(ret);
4923                 goto out;
4924         }
4925         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4926                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4927         if (ret) {
4928                 mlog_errno(ret);
4929                 goto out;
4930         }
4931
4932         /* Now update the headers */
4933         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4934         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4935
4936         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4937         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4938
4939         if (first_hash)
4940                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4941
4942 out:
4943         ocfs2_xattr_bucket_free(new_first);
4944         ocfs2_xattr_bucket_free(old_first);
4945         return ret;
4946 }
4947
4948 /*
4949  * Move some xattrs in this cluster to the new cluster.
4950  * This function should only be called when bucket size == cluster size.
4951  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4952  */
4953 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4954                                       handle_t *handle,
4955                                       u64 prev_blk,
4956                                       u64 new_blk,
4957                                       u32 *first_hash)
4958 {
4959         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4960         int ret, credits = 2 * blk_per_bucket;
4961
4962         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4963
4964         ret = ocfs2_extend_trans(handle, credits);
4965         if (ret) {
4966                 mlog_errno(ret);
4967                 return ret;
4968         }
4969
4970         /* Move half of the xattr in start_blk to the next bucket. */
4971         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4972                                           new_blk, first_hash, 1);
4973 }
4974
4975 /*
4976  * Move some xattrs from the old cluster to the new one since they are not
4977  * contiguous in ocfs2 xattr tree.
4978  *
4979  * new_blk starts a new separate cluster, and we will move some xattrs from
4980  * prev_blk to it. v_start will be set as the first name hash value in this
4981  * new cluster so that it can be used as e_cpos during tree insertion and
4982  * don't collide with our original b-tree operations. first_bh and header_bh
4983  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4984  * to extend the insert bucket.
4985  *
4986  * The problem is how much xattr should we move to the new one and when should
4987  * we update first_bh and header_bh?
4988  * 1. If cluster size > bucket size, that means the previous cluster has more
4989  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4990  *    update the first_bh and header_bh if the insert bucket has been moved
4991  *    to the new cluster.
4992  * 2. If cluster_size == bucket_size:
4993  *    a) If the previous extent rec has more than one cluster and the insert
4994  *       place isn't in the last cluster, copy the entire last cluster to the
4995  *       new one. This time, we don't need to upate the first_bh and header_bh
4996  *       since they will not be moved into the new cluster.
4997  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4998  *       the new one. And we set the extend flag to zero if the insert place is
4999  *       moved into the new allocated cluster since no extend is needed.
5000  */
5001 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5002                                             handle_t *handle,
5003                                             struct ocfs2_xattr_bucket *first,
5004                                             struct ocfs2_xattr_bucket *target,
5005                                             u64 new_blk,
5006                                             u32 prev_clusters,
5007                                             u32 *v_start,
5008                                             int *extend)
5009 {
5010         int ret;
5011
5012         trace_ocfs2_adjust_xattr_cross_cluster(
5013                         (unsigned long long)bucket_blkno(first),
5014                         (unsigned long long)new_blk, prev_clusters);
5015
5016         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5017                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5018                                                           handle,
5019                                                           first, target,
5020                                                           new_blk,
5021                                                           prev_clusters,
5022                                                           v_start);
5023                 if (ret)
5024                         mlog_errno(ret);
5025         } else {
5026                 /* The start of the last cluster in the first extent */
5027                 u64 last_blk = bucket_blkno(first) +
5028                         ((prev_clusters - 1) *
5029                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
5030
5031                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5032                         ret = ocfs2_mv_xattr_buckets(inode, handle,
5033                                                      bucket_blkno(first),
5034                                                      last_blk, new_blk, 0,
5035                                                      v_start);
5036                         if (ret)
5037                                 mlog_errno(ret);
5038                 } else {
5039                         ret = ocfs2_divide_xattr_cluster(inode, handle,
5040                                                          last_blk, new_blk,
5041                                                          v_start);
5042                         if (ret)
5043                                 mlog_errno(ret);
5044
5045                         if ((bucket_blkno(target) == last_blk) && extend)
5046                                 *extend = 0;
5047                 }
5048         }
5049
5050         return ret;
5051 }
5052
5053 /*
5054  * Add a new cluster for xattr storage.
5055  *
5056  * If the new cluster is contiguous with the previous one, it will be
5057  * appended to the same extent record, and num_clusters will be updated.
5058  * If not, we will insert a new extent for it and move some xattrs in
5059  * the last cluster into the new allocated one.
5060  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5061  * lose the benefits of hashing because we'll have to search large leaves.
5062  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5063  * if it's bigger).
5064  *
5065  * first_bh is the first block of the previous extent rec and header_bh
5066  * indicates the bucket we will insert the new xattrs. They will be updated
5067  * when the header_bh is moved into the new cluster.
5068  */
5069 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5070                                        struct buffer_head *root_bh,
5071                                        struct ocfs2_xattr_bucket *first,
5072                                        struct ocfs2_xattr_bucket *target,
5073                                        u32 *num_clusters,
5074                                        u32 prev_cpos,
5075                                        int *extend,
5076                                        struct ocfs2_xattr_set_ctxt *ctxt)
5077 {
5078         int ret;
5079         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5080         u32 prev_clusters = *num_clusters;
5081         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5082         u64 block;
5083         handle_t *handle = ctxt->handle;
5084         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5085         struct ocfs2_extent_tree et;
5086
5087         trace_ocfs2_add_new_xattr_cluster_begin(
5088                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5089                 (unsigned long long)bucket_blkno(first),
5090                 prev_cpos, prev_clusters);
5091
5092         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5093
5094         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5095                                       OCFS2_JOURNAL_ACCESS_WRITE);
5096         if (ret < 0) {
5097                 mlog_errno(ret);
5098                 goto leave;
5099         }
5100
5101         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5102                                      clusters_to_add, &bit_off, &num_bits);
5103         if (ret < 0) {
5104                 if (ret != -ENOSPC)
5105                         mlog_errno(ret);
5106                 goto leave;
5107         }
5108
5109         BUG_ON(num_bits > clusters_to_add);
5110
5111         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5112         trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5113
5114         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5115             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5116              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5117                 /*
5118                  * If this cluster is contiguous with the old one and
5119                  * adding this new cluster, we don't surpass the limit of
5120                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5121                  * initialized and used like other buckets in the previous
5122                  * cluster.
5123                  * So add it as a contiguous one. The caller will handle
5124                  * its init process.
5125                  */
5126                 v_start = prev_cpos + prev_clusters;
5127                 *num_clusters = prev_clusters + num_bits;
5128         } else {
5129                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5130                                                        handle,
5131                                                        first,
5132                                                        target,
5133                                                        block,
5134                                                        prev_clusters,
5135                                                        &v_start,
5136                                                        extend);
5137                 if (ret) {
5138                         mlog_errno(ret);
5139                         goto leave;
5140                 }
5141         }
5142
5143         trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5144                                                  v_start, num_bits);
5145         ret = ocfs2_insert_extent(handle, &et, v_start, block,
5146                                   num_bits, 0, ctxt->meta_ac);
5147         if (ret < 0) {
5148                 mlog_errno(ret);
5149                 goto leave;
5150         }
5151
5152         ocfs2_journal_dirty(handle, root_bh);
5153
5154 leave:
5155         return ret;
5156 }
5157
5158 /*
5159  * We are given an extent.  'first' is the bucket at the very front of
5160  * the extent.  The extent has space for an additional bucket past
5161  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5162  * of the target bucket.  We wish to shift every bucket past the target
5163  * down one, filling in that additional space.  When we get back to the
5164  * target, we split the target between itself and the now-empty bucket
5165  * at target+1 (aka, target_blkno + blks_per_bucket).
5166  */
5167 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5168                                      handle_t *handle,
5169                                      struct ocfs2_xattr_bucket *first,
5170                                      u64 target_blk,
5171                                      u32 num_clusters)
5172 {
5173         int ret, credits;
5174         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5175         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5176         u64 end_blk;
5177         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5178
5179         trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5180                                         (unsigned long long)bucket_blkno(first),
5181                                         num_clusters, new_bucket);
5182
5183         /* The extent must have room for an additional bucket */
5184         BUG_ON(new_bucket >=
5185                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5186
5187         /* end_blk points to the last existing bucket */
5188         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5189
5190         /*
5191          * end_blk is the start of the last existing bucket.
5192          * Thus, (end_blk - target_blk) covers the target bucket and
5193          * every bucket after it up to, but not including, the last
5194          * existing bucket.  Then we add the last existing bucket, the
5195          * new bucket, and the first bucket (3 * blk_per_bucket).
5196          */
5197         credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5198         ret = ocfs2_extend_trans(handle, credits);
5199         if (ret) {
5200                 mlog_errno(ret);
5201                 goto out;
5202         }
5203
5204         ret = ocfs2_xattr_bucket_journal_access(handle, first,
5205                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5206         if (ret) {
5207                 mlog_errno(ret);
5208                 goto out;
5209         }
5210
5211         while (end_blk != target_blk) {
5212                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5213                                             end_blk + blk_per_bucket, 0);
5214                 if (ret)
5215                         goto out;
5216                 end_blk -= blk_per_bucket;
5217         }
5218
5219         /* Move half of the xattr in target_blkno to the next bucket. */
5220         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5221                                         target_blk + blk_per_bucket, NULL, 0);
5222
5223         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5224         ocfs2_xattr_bucket_journal_dirty(handle, first);
5225
5226 out:
5227         return ret;
5228 }
5229
5230 /*
5231  * Add new xattr bucket in an extent record and adjust the buckets
5232  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5233  * bucket we want to insert into.
5234  *
5235  * In the easy case, we will move all the buckets after target down by
5236  * one. Half of target's xattrs will be moved to the next bucket.
5237  *
5238  * If current cluster is full, we'll allocate a new one.  This may not
5239  * be contiguous.  The underlying calls will make sure that there is
5240  * space for the insert, shifting buckets around if necessary.
5241  * 'target' may be moved by those calls.
5242  */
5243 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5244                                       struct buffer_head *xb_bh,
5245                                       struct ocfs2_xattr_bucket *target,
5246                                       struct ocfs2_xattr_set_ctxt *ctxt)
5247 {
5248         struct ocfs2_xattr_block *xb =
5249                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5250         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5251         struct ocfs2_extent_list *el = &xb_root->xt_list;
5252         u32 name_hash =
5253                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5254         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5255         int ret, num_buckets, extend = 1;
5256         u64 p_blkno;
5257         u32 e_cpos, num_clusters;
5258         /* The bucket at the front of the extent */
5259         struct ocfs2_xattr_bucket *first;
5260
5261         trace_ocfs2_add_new_xattr_bucket(
5262                                 (unsigned long long)bucket_blkno(target));
5263
5264         /* The first bucket of the original extent */
5265         first = ocfs2_xattr_bucket_new(inode);
5266         if (!first) {
5267                 ret = -ENOMEM;
5268                 mlog_errno(ret);
5269                 goto out;
5270         }
5271
5272         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5273                                   &num_clusters, el);
5274         if (ret) {
5275                 mlog_errno(ret);
5276                 goto out;
5277         }
5278
5279         ret = ocfs2_read_xattr_bucket(first, p_blkno);
5280         if (ret) {
5281                 mlog_errno(ret);
5282                 goto out;
5283         }
5284
5285         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5286         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5287                 /*
5288                  * This can move first+target if the target bucket moves
5289                  * to the new extent.
5290                  */
5291                 ret = ocfs2_add_new_xattr_cluster(inode,
5292                                                   xb_bh,
5293                                                   first,
5294                                                   target,
5295                                                   &num_clusters,
5296                                                   e_cpos,
5297                                                   &extend,
5298                                                   ctxt);
5299                 if (ret) {
5300                         mlog_errno(ret);
5301                         goto out;
5302                 }
5303         }
5304
5305         if (extend) {
5306                 ret = ocfs2_extend_xattr_bucket(inode,
5307                                                 ctxt->handle,
5308                                                 first,
5309                                                 bucket_blkno(target),
5310                                                 num_clusters);
5311                 if (ret)
5312                         mlog_errno(ret);
5313         }
5314
5315 out:
5316         ocfs2_xattr_bucket_free(first);
5317
5318         return ret;
5319 }
5320
5321 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
5322                                         struct ocfs2_xattr_bucket *bucket,
5323                                         int offs)
5324 {
5325         int block_off = offs >> inode->i_sb->s_blocksize_bits;
5326
5327         offs = offs % inode->i_sb->s_blocksize;
5328         return bucket_block(bucket, block_off) + offs;
5329 }
5330
5331 /*
5332  * Truncate the specified xe_off entry in xattr bucket.
5333  * bucket is indicated by header_bh and len is the new length.
5334  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5335  *
5336  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5337  */
5338 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5339                                              struct ocfs2_xattr_bucket *bucket,
5340                                              int xe_off,
5341                                              int len,
5342                                              struct ocfs2_xattr_set_ctxt *ctxt)
5343 {
5344         int ret, offset;
5345         u64 value_blk;
5346         struct ocfs2_xattr_entry *xe;
5347         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5348         size_t blocksize = inode->i_sb->s_blocksize;
5349         struct ocfs2_xattr_value_buf vb = {
5350                 .vb_access = ocfs2_journal_access,
5351         };
5352
5353         xe = &xh->xh_entries[xe_off];
5354
5355         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5356
5357         offset = le16_to_cpu(xe->xe_name_offset) +
5358                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5359
5360         value_blk = offset / blocksize;
5361
5362         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5363         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5364
5365         vb.vb_bh = bucket->bu_bhs[value_blk];
5366         BUG_ON(!vb.vb_bh);
5367
5368         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5369                 (vb.vb_bh->b_data + offset % blocksize);
5370
5371         /*
5372          * From here on out we have to dirty the bucket.  The generic
5373          * value calls only modify one of the bucket's bhs, but we need
5374          * to send the bucket at once.  So if they error, they *could* have
5375          * modified something.  We have to assume they did, and dirty
5376          * the whole bucket.  This leaves us in a consistent state.
5377          */
5378         trace_ocfs2_xattr_bucket_value_truncate(
5379                         (unsigned long long)bucket_blkno(bucket), xe_off, len);
5380         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5381         if (ret) {
5382                 mlog_errno(ret);
5383                 goto out;
5384         }
5385
5386         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5387                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5388         if (ret) {
5389                 mlog_errno(ret);
5390                 goto out;
5391         }
5392
5393         xe->xe_value_size = cpu_to_le64(len);
5394
5395         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5396
5397 out:
5398         return ret;
5399 }
5400
5401 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5402                                   struct buffer_head *root_bh,
5403                                   u64 blkno,
5404                                   u32 cpos,
5405                                   u32 len,
5406                                   void *para)
5407 {
5408         int ret;
5409         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5410         struct inode *tl_inode = osb->osb_tl_inode;
5411         handle_t *handle;
5412         struct ocfs2_xattr_block *xb =
5413                         (struct ocfs2_xattr_block *)root_bh->b_data;
5414         struct ocfs2_alloc_context *meta_ac = NULL;
5415         struct ocfs2_cached_dealloc_ctxt dealloc;
5416         struct ocfs2_extent_tree et;
5417
5418         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5419                                           ocfs2_delete_xattr_in_bucket, para);
5420         if (ret) {
5421                 mlog_errno(ret);
5422                 return ret;
5423         }
5424
5425         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5426
5427         ocfs2_init_dealloc_ctxt(&dealloc);
5428
5429         trace_ocfs2_rm_xattr_cluster(
5430                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
5431                         (unsigned long long)blkno, cpos, len);
5432
5433         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5434                                                len);
5435
5436         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5437         if (ret) {
5438                 mlog_errno(ret);
5439                 return ret;
5440         }
5441
5442         mutex_lock(&tl_inode->i_mutex);
5443
5444         if (ocfs2_truncate_log_needs_flush(osb)) {
5445                 ret = __ocfs2_flush_truncate_log(osb);
5446                 if (ret < 0) {
5447                         mlog_errno(ret);
5448                         goto out;
5449                 }
5450         }
5451
5452         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5453         if (IS_ERR(handle)) {
5454                 ret = -ENOMEM;
5455                 mlog_errno(ret);
5456                 goto out;
5457         }
5458
5459         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5460                                       OCFS2_JOURNAL_ACCESS_WRITE);
5461         if (ret) {
5462                 mlog_errno(ret);
5463                 goto out_commit;
5464         }
5465
5466         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5467                                   &dealloc);
5468         if (ret) {
5469                 mlog_errno(ret);
5470                 goto out_commit;
5471         }
5472
5473         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5474         ocfs2_journal_dirty(handle, root_bh);
5475
5476         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5477         if (ret)
5478                 mlog_errno(ret);
5479
5480 out_commit:
5481         ocfs2_commit_trans(osb, handle);
5482 out:
5483         ocfs2_schedule_truncate_log_flush(osb, 1);
5484
5485         mutex_unlock(&tl_inode->i_mutex);
5486
5487         if (meta_ac)
5488                 ocfs2_free_alloc_context(meta_ac);
5489
5490         ocfs2_run_deallocs(osb, &dealloc);
5491
5492         return ret;
5493 }
5494
5495 /*
5496  * check whether the xattr bucket is filled up with the same hash value.
5497  * If we want to insert the xattr with the same hash, return -ENOSPC.
5498  * If we want to insert a xattr with different hash value, go ahead
5499  * and ocfs2_divide_xattr_bucket will handle this.
5500  */
5501 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5502                                               struct ocfs2_xattr_bucket *bucket,
5503                                               const char *name)
5504 {
5505         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5506         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5507
5508         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5509                 return 0;
5510
5511         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5512             xh->xh_entries[0].xe_name_hash) {
5513                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5514                      "hash = %u\n",
5515                      (unsigned long long)bucket_blkno(bucket),
5516                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5517                 return -ENOSPC;
5518         }
5519
5520         return 0;
5521 }
5522
5523 /*
5524  * Try to set the entry in the current bucket.  If we fail, the caller
5525  * will handle getting us another bucket.
5526  */
5527 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5528                                         struct ocfs2_xattr_info *xi,
5529                                         struct ocfs2_xattr_search *xs,
5530                                         struct ocfs2_xattr_set_ctxt *ctxt)
5531 {
5532         int ret;
5533         struct ocfs2_xa_loc loc;
5534
5535         trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5536
5537         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5538                                        xs->not_found ? NULL : xs->here);
5539         ret = ocfs2_xa_set(&loc, xi, ctxt);
5540         if (!ret) {
5541                 xs->here = loc.xl_entry;
5542                 goto out;
5543         }
5544         if (ret != -ENOSPC) {
5545                 mlog_errno(ret);
5546                 goto out;
5547         }
5548
5549         /* Ok, we need space.  Let's try defragmenting the bucket. */
5550         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5551                                         xs->bucket);
5552         if (ret) {
5553                 mlog_errno(ret);
5554                 goto out;
5555         }
5556
5557         ret = ocfs2_xa_set(&loc, xi, ctxt);
5558         if (!ret) {
5559                 xs->here = loc.xl_entry;
5560                 goto out;
5561         }
5562         if (ret != -ENOSPC)
5563                 mlog_errno(ret);
5564
5565
5566 out:
5567         return ret;
5568 }
5569
5570 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5571                                              struct ocfs2_xattr_info *xi,
5572                                              struct ocfs2_xattr_search *xs,
5573                                              struct ocfs2_xattr_set_ctxt *ctxt)
5574 {
5575         int ret;
5576
5577         trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5578
5579         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5580         if (!ret)
5581                 goto out;
5582         if (ret != -ENOSPC) {
5583                 mlog_errno(ret);
5584                 goto out;
5585         }
5586
5587         /* Ack, need more space.  Let's try to get another bucket! */
5588
5589         /*
5590          * We do not allow for overlapping ranges between buckets. And
5591          * the maximum number of collisions we will allow for then is
5592          * one bucket's worth, so check it here whether we need to
5593          * add a new bucket for the insert.
5594          */
5595         ret = ocfs2_check_xattr_bucket_collision(inode,
5596                                                  xs->bucket,
5597                                                  xi->xi_name);
5598         if (ret) {
5599                 mlog_errno(ret);
5600                 goto out;
5601         }
5602
5603         ret = ocfs2_add_new_xattr_bucket(inode,
5604                                          xs->xattr_bh,
5605                                          xs->bucket,
5606                                          ctxt);
5607         if (ret) {
5608                 mlog_errno(ret);
5609                 goto out;
5610         }
5611
5612         /*
5613          * ocfs2_add_new_xattr_bucket() will have updated
5614          * xs->bucket if it moved, but it will not have updated
5615          * any of the other search fields.  Thus, we drop it and
5616          * re-search.  Everything should be cached, so it'll be
5617          * quick.
5618          */
5619         ocfs2_xattr_bucket_relse(xs->bucket);
5620         ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5621                                            xi->xi_name_index,
5622                                            xi->xi_name, xs);
5623         if (ret && ret != -ENODATA)
5624                 goto out;
5625         xs->not_found = ret;
5626
5627         /* Ok, we have a new bucket, let's try again */
5628         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5629         if (ret && (ret != -ENOSPC))
5630                 mlog_errno(ret);
5631
5632 out:
5633         return ret;
5634 }
5635
5636 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5637                                         struct ocfs2_xattr_bucket *bucket,
5638                                         void *para)
5639 {
5640         int ret = 0, ref_credits;
5641         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5642         u16 i;
5643         struct ocfs2_xattr_entry *xe;
5644         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5645         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5646         int credits = ocfs2_remove_extent_credits(osb->sb) +
5647                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5648         struct ocfs2_xattr_value_root *xv;
5649         struct ocfs2_rm_xattr_bucket_para *args =
5650                         (struct ocfs2_rm_xattr_bucket_para *)para;
5651
5652         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5653
5654         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5655                 xe = &xh->xh_entries[i];
5656                 if (ocfs2_xattr_is_local(xe))
5657                         continue;
5658
5659                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5660                                                       i, &xv, NULL);
5661
5662                 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5663                                                          args->ref_ci,
5664                                                          args->ref_root_bh,
5665                                                          &ctxt.meta_ac,
5666                                                          &ref_credits);
5667
5668                 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5669                 if (IS_ERR(ctxt.handle)) {
5670                         ret = PTR_ERR(ctxt.handle);
5671                         mlog_errno(ret);
5672                         break;
5673                 }
5674
5675                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5676                                                         i, 0, &ctxt);
5677
5678                 ocfs2_commit_trans(osb, ctxt.handle);
5679                 if (ctxt.meta_ac) {
5680                         ocfs2_free_alloc_context(ctxt.meta_ac);
5681                         ctxt.meta_ac = NULL;
5682                 }
5683                 if (ret) {
5684                         mlog_errno(ret);
5685                         break;
5686                 }
5687         }
5688
5689         if (ctxt.meta_ac)
5690                 ocfs2_free_alloc_context(ctxt.meta_ac);
5691         ocfs2_schedule_truncate_log_flush(osb, 1);
5692         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5693         return ret;
5694 }
5695
5696 /*
5697  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5698  * or change the extent record flag), we need to recalculate
5699  * the metaecc for the whole bucket. So it is done here.
5700  *
5701  * Note:
5702  * We have to give the extra credits for the caller.
5703  */
5704 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5705                                             handle_t *handle,
5706                                             void *para)
5707 {
5708         int ret;
5709         struct ocfs2_xattr_bucket *bucket =
5710                         (struct ocfs2_xattr_bucket *)para;
5711
5712         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5713                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5714         if (ret) {
5715                 mlog_errno(ret);
5716                 return ret;
5717         }
5718
5719         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5720
5721         return 0;
5722 }
5723
5724 /*
5725  * Special action we need if the xattr value is refcounted.
5726  *
5727  * 1. If the xattr is refcounted, lock the tree.
5728  * 2. CoW the xattr if we are setting the new value and the value
5729  *    will be stored outside.
5730  * 3. In other case, decrease_refcount will work for us, so just
5731  *    lock the refcount tree, calculate the meta and credits is OK.
5732  *
5733  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5734  * currently CoW is a completed transaction, while this function
5735  * will also lock the allocators and let us deadlock. So we will
5736  * CoW the whole xattr value.
5737  */
5738 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5739                                         struct ocfs2_dinode *di,
5740                                         struct ocfs2_xattr_info *xi,
5741                                         struct ocfs2_xattr_search *xis,
5742                                         struct ocfs2_xattr_search *xbs,
5743                                         struct ocfs2_refcount_tree **ref_tree,
5744                                         int *meta_add,
5745                                         int *credits)
5746 {
5747         int ret = 0;
5748         struct ocfs2_xattr_block *xb;
5749         struct ocfs2_xattr_entry *xe;
5750         char *base;
5751         u32 p_cluster, num_clusters;
5752         unsigned int ext_flags;
5753         int name_offset, name_len;
5754         struct ocfs2_xattr_value_buf vb;
5755         struct ocfs2_xattr_bucket *bucket = NULL;
5756         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5757         struct ocfs2_post_refcount refcount;
5758         struct ocfs2_post_refcount *p = NULL;
5759         struct buffer_head *ref_root_bh = NULL;
5760
5761         if (!xis->not_found) {
5762                 xe = xis->here;
5763                 name_offset = le16_to_cpu(xe->xe_name_offset);
5764                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5765                 base = xis->base;
5766                 vb.vb_bh = xis->inode_bh;
5767                 vb.vb_access = ocfs2_journal_access_di;
5768         } else {
5769                 int i, block_off = 0;
5770                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5771                 xe = xbs->here;
5772                 name_offset = le16_to_cpu(xe->xe_name_offset);
5773                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5774                 i = xbs->here - xbs->header->xh_entries;
5775
5776                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5777                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5778                                                         bucket_xh(xbs->bucket),
5779                                                         i, &block_off,
5780                                                         &name_offset);
5781                         if (ret) {
5782                                 mlog_errno(ret);
5783                                 goto out;
5784                         }
5785                         base = bucket_block(xbs->bucket, block_off);
5786                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5787                         vb.vb_access = ocfs2_journal_access;
5788
5789                         if (ocfs2_meta_ecc(osb)) {
5790                                 /*create parameters for ocfs2_post_refcount. */
5791                                 bucket = xbs->bucket;
5792                                 refcount.credits = bucket->bu_blocks;
5793                                 refcount.para = bucket;
5794                                 refcount.func =
5795                                         ocfs2_xattr_bucket_post_refcount;
5796                                 p = &refcount;
5797                         }
5798                 } else {
5799                         base = xbs->base;
5800                         vb.vb_bh = xbs->xattr_bh;
5801                         vb.vb_access = ocfs2_journal_access_xb;
5802                 }
5803         }
5804
5805         if (ocfs2_xattr_is_local(xe))
5806                 goto out;
5807
5808         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5809                                 (base + name_offset + name_len);
5810
5811         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5812                                        &num_clusters, &vb.vb_xv->xr_list,
5813                                        &ext_flags);
5814         if (ret) {
5815                 mlog_errno(ret);
5816                 goto out;
5817         }
5818
5819         /*
5820          * We just need to check the 1st extent record, since we always
5821          * CoW the whole xattr. So there shouldn't be a xattr with
5822          * some REFCOUNT extent recs after the 1st one.
5823          */
5824         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5825                 goto out;
5826
5827         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5828                                        1, ref_tree, &ref_root_bh);
5829         if (ret) {
5830                 mlog_errno(ret);
5831                 goto out;
5832         }
5833
5834         /*
5835          * If we are deleting the xattr or the new size will be stored inside,
5836          * cool, leave it there, the xattr truncate process will remove them
5837          * for us(it still needs the refcount tree lock and the meta, credits).
5838          * And the worse case is that every cluster truncate will split the
5839          * refcount tree, and make the original extent become 3. So we will need
5840          * 2 * cluster more extent recs at most.
5841          */
5842         if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5843
5844                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5845                                                          &(*ref_tree)->rf_ci,
5846                                                          ref_root_bh, vb.vb_xv,
5847                                                          meta_add, credits);
5848                 if (ret)
5849                         mlog_errno(ret);
5850                 goto out;
5851         }
5852
5853         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5854                                        *ref_tree, ref_root_bh, 0,
5855                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5856         if (ret)
5857                 mlog_errno(ret);
5858
5859 out:
5860         brelse(ref_root_bh);
5861         return ret;
5862 }
5863
5864 /*
5865  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5866  * The physical clusters will be added to refcount tree.
5867  */
5868 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5869                                 struct ocfs2_xattr_value_root *xv,
5870                                 struct ocfs2_extent_tree *value_et,
5871                                 struct ocfs2_caching_info *ref_ci,
5872                                 struct buffer_head *ref_root_bh,
5873                                 struct ocfs2_cached_dealloc_ctxt *dealloc,
5874                                 struct ocfs2_post_refcount *refcount)
5875 {
5876         int ret = 0;
5877         u32 clusters = le32_to_cpu(xv->xr_clusters);
5878         u32 cpos, p_cluster, num_clusters;
5879         struct ocfs2_extent_list *el = &xv->xr_list;
5880         unsigned int ext_flags;
5881
5882         cpos = 0;
5883         while (cpos < clusters) {
5884                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5885                                                &num_clusters, el, &ext_flags);
5886                 if (ret) {
5887                         mlog_errno(ret);
5888                         break;
5889                 }
5890
5891                 cpos += num_clusters;
5892                 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5893                         continue;
5894
5895                 BUG_ON(!p_cluster);
5896
5897                 ret = ocfs2_add_refcount_flag(inode, value_et,
5898                                               ref_ci, ref_root_bh,
5899                                               cpos - num_clusters,
5900                                               p_cluster, num_clusters,
5901                                               dealloc, refcount);
5902                 if (ret) {
5903                         mlog_errno(ret);
5904                         break;
5905                 }
5906         }
5907
5908         return ret;
5909 }
5910
5911 /*
5912  * Given a normal ocfs2_xattr_header, refcount all the entries which
5913  * have value stored outside.
5914  * Used for xattrs stored in inode and ocfs2_xattr_block.
5915  */
5916 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5917                                 struct ocfs2_xattr_value_buf *vb,
5918                                 struct ocfs2_xattr_header *header,
5919                                 struct ocfs2_caching_info *ref_ci,
5920                                 struct buffer_head *ref_root_bh,
5921                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5922 {
5923
5924         struct ocfs2_xattr_entry *xe;
5925         struct ocfs2_xattr_value_root *xv;
5926         struct ocfs2_extent_tree et;
5927         int i, ret = 0;
5928
5929         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5930                 xe = &header->xh_entries[i];
5931
5932                 if (ocfs2_xattr_is_local(xe))
5933                         continue;
5934
5935                 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5936                         le16_to_cpu(xe->xe_name_offset) +
5937                         OCFS2_XATTR_SIZE(xe->xe_name_len));
5938
5939                 vb->vb_xv = xv;
5940                 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5941
5942                 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5943                                                         ref_ci, ref_root_bh,
5944                                                         dealloc, NULL);
5945                 if (ret) {
5946                         mlog_errno(ret);
5947                         break;
5948                 }
5949         }
5950
5951         return ret;
5952 }
5953
5954 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5955                                 struct buffer_head *fe_bh,
5956                                 struct ocfs2_caching_info *ref_ci,
5957                                 struct buffer_head *ref_root_bh,
5958                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5959 {
5960         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5961         struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5962                                 (fe_bh->b_data + inode->i_sb->s_blocksize -
5963                                 le16_to_cpu(di->i_xattr_inline_size));
5964         struct ocfs2_xattr_value_buf vb = {
5965                 .vb_bh = fe_bh,
5966                 .vb_access = ocfs2_journal_access_di,
5967         };
5968
5969         return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5970                                                   ref_ci, ref_root_bh, dealloc);
5971 }
5972
5973 struct ocfs2_xattr_tree_value_refcount_para {
5974         struct ocfs2_caching_info *ref_ci;
5975         struct buffer_head *ref_root_bh;
5976         struct ocfs2_cached_dealloc_ctxt *dealloc;
5977 };
5978
5979 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5980                                            struct ocfs2_xattr_bucket *bucket,
5981                                            int offset,
5982                                            struct ocfs2_xattr_value_root **xv,
5983                                            struct buffer_head **bh)
5984 {
5985         int ret, block_off, name_offset;
5986         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5987         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
5988         void *base;
5989
5990         ret = ocfs2_xattr_bucket_get_name_value(sb,
5991                                                 bucket_xh(bucket),
5992                                                 offset,
5993                                                 &block_off,
5994                                                 &name_offset);
5995         if (ret) {
5996                 mlog_errno(ret);
5997                 goto out;
5998         }
5999
6000         base = bucket_block(bucket, block_off);
6001
6002         *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6003                          OCFS2_XATTR_SIZE(xe->xe_name_len));
6004
6005         if (bh)
6006                 *bh = bucket->bu_bhs[block_off];
6007 out:
6008         return ret;
6009 }
6010
6011 /*
6012  * For a given xattr bucket, refcount all the entries which
6013  * have value stored outside.
6014  */
6015 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6016                                              struct ocfs2_xattr_bucket *bucket,
6017                                              void *para)
6018 {
6019         int i, ret = 0;
6020         struct ocfs2_extent_tree et;
6021         struct ocfs2_xattr_tree_value_refcount_para *ref =
6022                         (struct ocfs2_xattr_tree_value_refcount_para *)para;
6023         struct ocfs2_xattr_header *xh =
6024                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6025         struct ocfs2_xattr_entry *xe;
6026         struct ocfs2_xattr_value_buf vb = {
6027                 .vb_access = ocfs2_journal_access,
6028         };
6029         struct ocfs2_post_refcount refcount = {
6030                 .credits = bucket->bu_blocks,
6031                 .para = bucket,
6032                 .func = ocfs2_xattr_bucket_post_refcount,
6033         };
6034         struct ocfs2_post_refcount *p = NULL;
6035
6036         /* We only need post_refcount if we support metaecc. */
6037         if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6038                 p = &refcount;
6039
6040         trace_ocfs2_xattr_bucket_value_refcount(
6041                                 (unsigned long long)bucket_blkno(bucket),
6042                                 le16_to_cpu(xh->xh_count));
6043         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6044                 xe = &xh->xh_entries[i];
6045
6046                 if (ocfs2_xattr_is_local(xe))
6047                         continue;
6048
6049                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6050                                                       &vb.vb_xv, &vb.vb_bh);
6051                 if (ret) {
6052                         mlog_errno(ret);
6053                         break;
6054                 }
6055
6056                 ocfs2_init_xattr_value_extent_tree(&et,
6057                                                    INODE_CACHE(inode), &vb);
6058
6059                 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6060                                                         &et, ref->ref_ci,
6061                                                         ref->ref_root_bh,
6062                                                         ref->dealloc, p);
6063                 if (ret) {
6064                         mlog_errno(ret);
6065                         break;
6066                 }
6067         }
6068
6069         return ret;
6070
6071 }
6072
6073 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6074                                      struct buffer_head *root_bh,
6075                                      u64 blkno, u32 cpos, u32 len, void *para)
6076 {
6077         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6078                                            ocfs2_xattr_bucket_value_refcount,
6079                                            para);
6080 }
6081
6082 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6083                                 struct buffer_head *blk_bh,
6084                                 struct ocfs2_caching_info *ref_ci,
6085                                 struct buffer_head *ref_root_bh,
6086                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
6087 {
6088         int ret = 0;
6089         struct ocfs2_xattr_block *xb =
6090                                 (struct ocfs2_xattr_block *)blk_bh->b_data;
6091
6092         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6093                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6094                 struct ocfs2_xattr_value_buf vb = {
6095                         .vb_bh = blk_bh,
6096                         .vb_access = ocfs2_journal_access_xb,
6097                 };
6098
6099                 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6100                                                          ref_ci, ref_root_bh,
6101                                                          dealloc);
6102         } else {
6103                 struct ocfs2_xattr_tree_value_refcount_para para = {
6104                         .ref_ci = ref_ci,
6105                         .ref_root_bh = ref_root_bh,
6106                         .dealloc = dealloc,
6107                 };
6108
6109                 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6110                                                 ocfs2_refcount_xattr_tree_rec,
6111                                                 &para);
6112         }
6113
6114         return ret;
6115 }
6116
6117 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6118                                      struct buffer_head *fe_bh,
6119                                      struct ocfs2_caching_info *ref_ci,
6120                                      struct buffer_head *ref_root_bh,
6121                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
6122 {
6123         int ret = 0;
6124         struct ocfs2_inode_info *oi = OCFS2_I(inode);
6125         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6126         struct buffer_head *blk_bh = NULL;
6127
6128         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6129                 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6130                                                          ref_ci, ref_root_bh,
6131                                                          dealloc);
6132                 if (ret) {
6133                         mlog_errno(ret);
6134                         goto out;
6135                 }
6136         }
6137
6138         if (!di->i_xattr_loc)
6139                 goto out;
6140
6141         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6142                                      &blk_bh);
6143         if (ret < 0) {
6144                 mlog_errno(ret);
6145                 goto out;
6146         }
6147
6148         ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6149                                                 ref_root_bh, dealloc);
6150         if (ret)
6151                 mlog_errno(ret);
6152
6153         brelse(blk_bh);
6154 out:
6155
6156         return ret;
6157 }
6158
6159 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6160 /*
6161  * Store the information we need in xattr reflink.
6162  * old_bh and new_bh are inode bh for the old and new inode.
6163  */
6164 struct ocfs2_xattr_reflink {
6165         struct inode *old_inode;
6166         struct inode *new_inode;
6167         struct buffer_head *old_bh;
6168         struct buffer_head *new_bh;
6169         struct ocfs2_caching_info *ref_ci;
6170         struct buffer_head *ref_root_bh;
6171         struct ocfs2_cached_dealloc_ctxt *dealloc;
6172         should_xattr_reflinked *xattr_reflinked;
6173 };
6174
6175 /*
6176  * Given a xattr header and xe offset,
6177  * return the proper xv and the corresponding bh.
6178  * xattr in inode, block and xattr tree have different implementaions.
6179  */
6180 typedef int (get_xattr_value_root)(struct super_block *sb,
6181                                    struct buffer_head *bh,
6182                                    struct ocfs2_xattr_header *xh,
6183                                    int offset,
6184                                    struct ocfs2_xattr_value_root **xv,
6185                                    struct buffer_head **ret_bh,
6186                                    void *para);
6187
6188 /*
6189  * Calculate all the xattr value root metadata stored in this xattr header and
6190  * credits we need if we create them from the scratch.
6191  * We use get_xattr_value_root so that all types of xattr container can use it.
6192  */
6193 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6194                                              struct buffer_head *bh,
6195                                              struct ocfs2_xattr_header *xh,
6196                                              int *metas, int *credits,
6197                                              int *num_recs,
6198                                              get_xattr_value_root *func,
6199                                              void *para)
6200 {
6201         int i, ret = 0;
6202         struct ocfs2_xattr_value_root *xv;
6203         struct ocfs2_xattr_entry *xe;
6204
6205         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6206                 xe = &xh->xh_entries[i];
6207                 if (ocfs2_xattr_is_local(xe))
6208                         continue;
6209
6210                 ret = func(sb, bh, xh, i, &xv, NULL, para);
6211                 if (ret) {
6212                         mlog_errno(ret);
6213                         break;
6214                 }
6215
6216                 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6217                           le16_to_cpu(xv->xr_list.l_next_free_rec);
6218
6219                 *credits += ocfs2_calc_extend_credits(sb,
6220                                                 &def_xv.xv.xr_list);
6221
6222                 /*
6223                  * If the value is a tree with depth > 1, We don't go deep
6224                  * to the extent block, so just calculate a maximum record num.
6225                  */
6226                 if (!xv->xr_list.l_tree_depth)
6227                         *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6228                 else
6229                         *num_recs += ocfs2_clusters_for_bytes(sb,
6230                                                               XATTR_SIZE_MAX);
6231         }
6232
6233         return ret;
6234 }
6235
6236 /* Used by xattr inode and block to return the right xv and buffer_head. */
6237 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6238                                       struct buffer_head *bh,
6239                                       struct ocfs2_xattr_header *xh,
6240                                       int offset,
6241                                       struct ocfs2_xattr_value_root **xv,
6242                                       struct buffer_head **ret_bh,
6243                                       void *para)
6244 {
6245         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6246
6247         *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6248                 le16_to_cpu(xe->xe_name_offset) +
6249                 OCFS2_XATTR_SIZE(xe->xe_name_len));
6250
6251         if (ret_bh)
6252                 *ret_bh = bh;
6253
6254         return 0;
6255 }
6256
6257 /*
6258  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6259  * It is only used for inline xattr and xattr block.
6260  */
6261 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6262                                         struct ocfs2_xattr_header *xh,
6263                                         struct buffer_head *ref_root_bh,
6264                                         int *credits,
6265                                         struct ocfs2_alloc_context **meta_ac)
6266 {
6267         int ret, meta_add = 0, num_recs = 0;
6268         struct ocfs2_refcount_block *rb =
6269                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6270
6271         *credits = 0;
6272
6273         ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6274                                                 &meta_add, credits, &num_recs,
6275                                                 ocfs2_get_xattr_value_root,
6276                                                 NULL);
6277         if (ret) {
6278                 mlog_errno(ret);
6279                 goto out;
6280         }
6281
6282         /*
6283          * We need to add/modify num_recs in refcount tree, so just calculate
6284          * an approximate number we need for refcount tree change.
6285          * Sometimes we need to split the tree, and after split,  half recs
6286          * will be moved to the new block, and a new block can only provide
6287          * half number of recs. So we multiple new blocks by 2.
6288          */
6289         num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6290         meta_add += num_recs;
6291         *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6292         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6293                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6294                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6295         else
6296                 *credits += 1;
6297
6298         ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6299         if (ret)
6300                 mlog_errno(ret);
6301
6302 out:
6303         return ret;
6304 }
6305
6306 /*
6307  * Given a xattr header, reflink all the xattrs in this container.
6308  * It can be used for inode, block and bucket.
6309  *
6310  * NOTE:
6311  * Before we call this function, the caller has memcpy the xattr in
6312  * old_xh to the new_xh.
6313  *
6314  * If args.xattr_reflinked is set, call it to decide whether the xe should
6315  * be reflinked or not. If not, remove it from the new xattr header.
6316  */
6317 static int ocfs2_reflink_xattr_header(handle_t *handle,
6318                                       struct ocfs2_xattr_reflink *args,
6319                                       struct buffer_head *old_bh,
6320                                       struct ocfs2_xattr_header *xh,
6321                                       struct buffer_head *new_bh,
6322                                       struct ocfs2_xattr_header *new_xh,
6323                                       struct ocfs2_xattr_value_buf *vb,
6324                                       struct ocfs2_alloc_context *meta_ac,
6325                                       get_xattr_value_root *func,
6326                                       void *para)
6327 {
6328         int ret = 0, i, j;
6329         struct super_block *sb = args->old_inode->i_sb;
6330         struct buffer_head *value_bh;
6331         struct ocfs2_xattr_entry *xe, *last;
6332         struct ocfs2_xattr_value_root *xv, *new_xv;
6333         struct ocfs2_extent_tree data_et;
6334         u32 clusters, cpos, p_cluster, num_clusters;
6335         unsigned int ext_flags = 0;
6336
6337         trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6338                                          le16_to_cpu(xh->xh_count));
6339
6340         last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6341         for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6342                 xe = &xh->xh_entries[i];
6343
6344                 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6345                         xe = &new_xh->xh_entries[j];
6346
6347                         le16_add_cpu(&new_xh->xh_count, -1);
6348                         if (new_xh->xh_count) {
6349                                 memmove(xe, xe + 1,
6350                                         (void *)last - (void *)xe);
6351                                 memset(last, 0,
6352                                        sizeof(struct ocfs2_xattr_entry));
6353                         }
6354
6355                         /*
6356                          * We don't want j to increase in the next round since
6357                          * it is already moved ahead.
6358                          */
6359                         j--;
6360                         continue;
6361                 }
6362
6363                 if (ocfs2_xattr_is_local(xe))
6364                         continue;
6365
6366                 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6367                 if (ret) {
6368                         mlog_errno(ret);
6369                         break;
6370                 }
6371
6372                 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6373                 if (ret) {
6374                         mlog_errno(ret);
6375                         break;
6376                 }
6377
6378                 /*
6379                  * For the xattr which has l_tree_depth = 0, all the extent
6380                  * recs have already be copied to the new xh with the
6381                  * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6382                  * increase the refount count int the refcount tree.
6383                  *
6384                  * For the xattr which has l_tree_depth > 0, we need
6385                  * to initialize it to the empty default value root,
6386                  * and then insert the extents one by one.
6387                  */
6388                 if (xv->xr_list.l_tree_depth) {
6389                         memcpy(new_xv, &def_xv, sizeof(def_xv));
6390                         vb->vb_xv = new_xv;
6391                         vb->vb_bh = value_bh;
6392                         ocfs2_init_xattr_value_extent_tree(&data_et,
6393                                         INODE_CACHE(args->new_inode), vb);
6394                 }
6395
6396                 clusters = le32_to_cpu(xv->xr_clusters);
6397                 cpos = 0;
6398                 while (cpos < clusters) {
6399                         ret = ocfs2_xattr_get_clusters(args->old_inode,
6400                                                        cpos,
6401                                                        &p_cluster,
6402                                                        &num_clusters,
6403                                                        &xv->xr_list,
6404                                                        &ext_flags);
6405                         if (ret) {
6406                                 mlog_errno(ret);
6407                                 goto out;
6408                         }
6409
6410                         BUG_ON(!p_cluster);
6411
6412                         if (xv->xr_list.l_tree_depth) {
6413                                 ret = ocfs2_insert_extent(handle,
6414                                                 &data_et, cpos,
6415                                                 ocfs2_clusters_to_blocks(
6416                                                         args->old_inode->i_sb,
6417                                                         p_cluster),
6418                                                 num_clusters, ext_flags,
6419                                                 meta_ac);
6420                                 if (ret) {
6421                                         mlog_errno(ret);
6422                                         goto out;
6423                                 }
6424                         }
6425
6426                         ret = ocfs2_increase_refcount(handle, args->ref_ci,
6427                                                       args->ref_root_bh,
6428                                                       p_cluster, num_clusters,
6429                                                       meta_ac, args->dealloc);
6430                         if (ret) {
6431                                 mlog_errno(ret);
6432                                 goto out;
6433                         }
6434
6435                         cpos += num_clusters;
6436                 }
6437         }
6438
6439 out:
6440         return ret;
6441 }
6442
6443 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6444 {
6445         int ret = 0, credits = 0;
6446         handle_t *handle;
6447         struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6448         struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6449         int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6450         int header_off = osb->sb->s_blocksize - inline_size;
6451         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6452                                         (args->old_bh->b_data + header_off);
6453         struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6454                                         (args->new_bh->b_data + header_off);
6455         struct ocfs2_alloc_context *meta_ac = NULL;
6456         struct ocfs2_inode_info *new_oi;
6457         struct ocfs2_dinode *new_di;
6458         struct ocfs2_xattr_value_buf vb = {
6459                 .vb_bh = args->new_bh,
6460                 .vb_access = ocfs2_journal_access_di,
6461         };
6462
6463         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6464                                                   &credits, &meta_ac);
6465         if (ret) {
6466                 mlog_errno(ret);
6467                 goto out;
6468         }
6469
6470         handle = ocfs2_start_trans(osb, credits);
6471         if (IS_ERR(handle)) {
6472                 ret = PTR_ERR(handle);
6473                 mlog_errno(ret);
6474                 goto out;
6475         }
6476
6477         ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6478                                       args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6479         if (ret) {
6480                 mlog_errno(ret);
6481                 goto out_commit;
6482         }
6483
6484         memcpy(args->new_bh->b_data + header_off,
6485                args->old_bh->b_data + header_off, inline_size);
6486
6487         new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6488         new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6489
6490         ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6491                                          args->new_bh, new_xh, &vb, meta_ac,
6492                                          ocfs2_get_xattr_value_root, NULL);
6493         if (ret) {
6494                 mlog_errno(ret);
6495                 goto out_commit;
6496         }
6497
6498         new_oi = OCFS2_I(args->new_inode);
6499         /*
6500          * Adjust extent record count to reserve space for extended attribute.
6501          * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6502          */
6503         if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6504             !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6505                 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6506                 le16_add_cpu(&el->l_count, -(inline_size /
6507                                         sizeof(struct ocfs2_extent_rec)));
6508         }
6509         spin_lock(&new_oi->ip_lock);
6510         new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6511         new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6512         spin_unlock(&new_oi->ip_lock);
6513
6514         ocfs2_journal_dirty(handle, args->new_bh);
6515
6516 out_commit:
6517         ocfs2_commit_trans(osb, handle);
6518
6519 out:
6520         if (meta_ac)
6521                 ocfs2_free_alloc_context(meta_ac);
6522         return ret;
6523 }
6524
6525 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6526                                           struct buffer_head *fe_bh,
6527                                           struct buffer_head **ret_bh,
6528                                           int indexed)
6529 {
6530         int ret;
6531         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6532         struct ocfs2_xattr_set_ctxt ctxt;
6533
6534         memset(&ctxt, 0, sizeof(ctxt));
6535         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6536         if (ret < 0) {
6537                 mlog_errno(ret);
6538                 return ret;
6539         }
6540
6541         ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6542         if (IS_ERR(ctxt.handle)) {
6543                 ret = PTR_ERR(ctxt.handle);
6544                 mlog_errno(ret);
6545                 goto out;
6546         }
6547
6548         trace_ocfs2_create_empty_xattr_block(
6549                                 (unsigned long long)fe_bh->b_blocknr, indexed);
6550         ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6551                                        ret_bh);
6552         if (ret)
6553                 mlog_errno(ret);
6554
6555         ocfs2_commit_trans(osb, ctxt.handle);
6556 out:
6557         ocfs2_free_alloc_context(ctxt.meta_ac);
6558         return ret;
6559 }
6560
6561 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6562                                      struct buffer_head *blk_bh,
6563                                      struct buffer_head *new_blk_bh)
6564 {
6565         int ret = 0, credits = 0;
6566         handle_t *handle;
6567         struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6568         struct ocfs2_dinode *new_di;
6569         struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6570         int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6571         struct ocfs2_xattr_block *xb =
6572                         (struct ocfs2_xattr_block *)blk_bh->b_data;
6573         struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6574         struct ocfs2_xattr_block *new_xb =
6575                         (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6576         struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6577         struct ocfs2_alloc_context *meta_ac;
6578         struct ocfs2_xattr_value_buf vb = {
6579                 .vb_bh = new_blk_bh,
6580                 .vb_access = ocfs2_journal_access_xb,
6581         };
6582
6583         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6584                                                   &credits, &meta_ac);
6585         if (ret) {
6586                 mlog_errno(ret);
6587                 return ret;
6588         }
6589
6590         /* One more credits in case we need to add xattr flags in new inode. */
6591         handle = ocfs2_start_trans(osb, credits + 1);
6592         if (IS_ERR(handle)) {
6593                 ret = PTR_ERR(handle);
6594                 mlog_errno(ret);
6595                 goto out;
6596         }
6597
6598         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6599                 ret = ocfs2_journal_access_di(handle,
6600                                               INODE_CACHE(args->new_inode),
6601                                               args->new_bh,
6602                                               OCFS2_JOURNAL_ACCESS_WRITE);
6603                 if (ret) {
6604                         mlog_errno(ret);
6605                         goto out_commit;
6606                 }
6607         }
6608
6609         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6610                                       new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6611         if (ret) {
6612                 mlog_errno(ret);
6613                 goto out_commit;
6614         }
6615
6616         memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6617                osb->sb->s_blocksize - header_off);
6618
6619         ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6620                                          new_blk_bh, new_xh, &vb, meta_ac,
6621                                          ocfs2_get_xattr_value_root, NULL);
6622         if (ret) {
6623                 mlog_errno(ret);
6624                 goto out_commit;
6625         }
6626
6627         ocfs2_journal_dirty(handle, new_blk_bh);
6628
6629         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6630                 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6631                 spin_lock(&new_oi->ip_lock);
6632                 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6633                 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6634                 spin_unlock(&new_oi->ip_lock);
6635
6636                 ocfs2_journal_dirty(handle, args->new_bh);
6637         }
6638
6639 out_commit:
6640         ocfs2_commit_trans(osb, handle);
6641
6642 out:
6643         ocfs2_free_alloc_context(meta_ac);
6644         return ret;
6645 }
6646
6647 struct ocfs2_reflink_xattr_tree_args {
6648         struct ocfs2_xattr_reflink *reflink;
6649         struct buffer_head *old_blk_bh;
6650         struct buffer_head *new_blk_bh;
6651         struct ocfs2_xattr_bucket *old_bucket;
6652         struct ocfs2_xattr_bucket *new_bucket;
6653 };
6654
6655 /*
6656  * NOTE:
6657  * We have to handle the case that both old bucket and new bucket
6658  * will call this function to get the right ret_bh.
6659  * So The caller must give us the right bh.
6660  */
6661 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6662                                         struct buffer_head *bh,
6663                                         struct ocfs2_xattr_header *xh,
6664                                         int offset,
6665                                         struct ocfs2_xattr_value_root **xv,
6666                                         struct buffer_head **ret_bh,
6667                                         void *para)
6668 {
6669         struct ocfs2_reflink_xattr_tree_args *args =
6670                         (struct ocfs2_reflink_xattr_tree_args *)para;
6671         struct ocfs2_xattr_bucket *bucket;
6672
6673         if (bh == args->old_bucket->bu_bhs[0])
6674                 bucket = args->old_bucket;
6675         else
6676                 bucket = args->new_bucket;
6677
6678         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6679                                                xv, ret_bh);
6680 }
6681
6682 struct ocfs2_value_tree_metas {
6683         int num_metas;
6684         int credits;
6685         int num_recs;
6686 };
6687
6688 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6689                                         struct buffer_head *bh,
6690                                         struct ocfs2_xattr_header *xh,
6691                                         int offset,
6692                                         struct ocfs2_xattr_value_root **xv,
6693                                         struct buffer_head **ret_bh,
6694                                         void *para)
6695 {
6696         struct ocfs2_xattr_bucket *bucket =
6697                                 (struct ocfs2_xattr_bucket *)para;
6698
6699         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6700                                                xv, ret_bh);
6701 }
6702
6703 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6704                                       struct ocfs2_xattr_bucket *bucket,
6705                                       void *para)
6706 {
6707         struct ocfs2_value_tree_metas *metas =
6708                         (struct ocfs2_value_tree_metas *)para;
6709         struct ocfs2_xattr_header *xh =
6710                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6711
6712         /* Add the credits for this bucket first. */
6713         metas->credits += bucket->bu_blocks;
6714         return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6715                                         xh, &metas->num_metas,
6716                                         &metas->credits, &metas->num_recs,
6717                                         ocfs2_value_tree_metas_in_bucket,
6718                                         bucket);
6719 }
6720
6721 /*
6722  * Given a xattr extent rec starting from blkno and having len clusters,
6723  * iterate all the buckets calculate how much metadata we need for reflinking
6724  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6725  */
6726 static int ocfs2_lock_reflink_xattr_rec_allocators(
6727                                 struct ocfs2_reflink_xattr_tree_args *args,
6728                                 struct ocfs2_extent_tree *xt_et,
6729                                 u64 blkno, u32 len, int *credits,
6730                                 struct ocfs2_alloc_context **meta_ac,
6731                                 struct ocfs2_alloc_context **data_ac)
6732 {
6733         int ret, num_free_extents;
6734         struct ocfs2_value_tree_metas metas;
6735         struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6736         struct ocfs2_refcount_block *rb;
6737
6738         memset(&metas, 0, sizeof(metas));
6739
6740         ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6741                                           ocfs2_calc_value_tree_metas, &metas);
6742         if (ret) {
6743                 mlog_errno(ret);
6744                 goto out;
6745         }
6746
6747         *credits = metas.credits;
6748
6749         /*
6750          * Calculate we need for refcount tree change.
6751          *
6752          * We need to add/modify num_recs in refcount tree, so just calculate
6753          * an approximate number we need for refcount tree change.
6754          * Sometimes we need to split the tree, and after split,  half recs
6755          * will be moved to the new block, and a new block can only provide
6756          * half number of recs. So we multiple new blocks by 2.
6757          * In the end, we have to add credits for modifying the already
6758          * existed refcount block.
6759          */
6760         rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6761         metas.num_recs =
6762                 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6763                  ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6764         metas.num_metas += metas.num_recs;
6765         *credits += metas.num_recs +
6766                     metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6767         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6768                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6769                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6770         else
6771                 *credits += 1;
6772
6773         /* count in the xattr tree change. */
6774         num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6775         if (num_free_extents < 0) {
6776                 ret = num_free_extents;
6777                 mlog_errno(ret);
6778                 goto out;
6779         }
6780
6781         if (num_free_extents < len)
6782                 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6783
6784         *credits += ocfs2_calc_extend_credits(osb->sb,
6785                                               xt_et->et_root_el);
6786
6787         if (metas.num_metas) {
6788                 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6789                                                         meta_ac);
6790                 if (ret) {
6791                         mlog_errno(ret);
6792                         goto out;
6793                 }
6794         }
6795
6796         if (len) {
6797                 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6798                 if (ret)
6799                         mlog_errno(ret);
6800         }
6801 out:
6802         if (ret) {
6803                 if (*meta_ac) {
6804                         ocfs2_free_alloc_context(*meta_ac);
6805                         *meta_ac = NULL;
6806                 }
6807         }
6808
6809         return ret;
6810 }
6811
6812 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6813                                 u64 blkno, u64 new_blkno, u32 clusters,
6814                                 u32 *cpos, int num_buckets,
6815                                 struct ocfs2_alloc_context *meta_ac,
6816                                 struct ocfs2_alloc_context *data_ac,
6817                                 struct ocfs2_reflink_xattr_tree_args *args)
6818 {
6819         int i, j, ret = 0;
6820         struct super_block *sb = args->reflink->old_inode->i_sb;
6821         int bpb = args->old_bucket->bu_blocks;
6822         struct ocfs2_xattr_value_buf vb = {
6823                 .vb_access = ocfs2_journal_access,
6824         };
6825
6826         for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6827                 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6828                 if (ret) {
6829                         mlog_errno(ret);
6830                         break;
6831                 }
6832
6833                 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno);
6834                 if (ret) {
6835                         mlog_errno(ret);
6836                         break;
6837                 }
6838
6839                 ret = ocfs2_xattr_bucket_journal_access(handle,
6840                                                 args->new_bucket,
6841                                                 OCFS2_JOURNAL_ACCESS_CREATE);
6842                 if (ret) {
6843                         mlog_errno(ret);
6844                         break;
6845                 }
6846
6847                 for (j = 0; j < bpb; j++)
6848                         memcpy(bucket_block(args->new_bucket, j),
6849                                bucket_block(args->old_bucket, j),
6850                                sb->s_blocksize);
6851
6852                 /*
6853                  * Record the start cpos so that we can use it to initialize
6854                  * our xattr tree we also set the xh_num_bucket for the new
6855                  * bucket.
6856                  */
6857                 if (i == 0) {
6858                         *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6859                                             xh_entries[0].xe_name_hash);
6860                         bucket_xh(args->new_bucket)->xh_num_buckets =
6861                                 cpu_to_le16(num_buckets);
6862                 }
6863
6864                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6865
6866                 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6867                                         args->old_bucket->bu_bhs[0],
6868                                         bucket_xh(args->old_bucket),
6869                                         args->new_bucket->bu_bhs[0],
6870                                         bucket_xh(args->new_bucket),
6871                                         &vb, meta_ac,
6872                                         ocfs2_get_reflink_xattr_value_root,
6873                                         args);
6874                 if (ret) {
6875                         mlog_errno(ret);
6876                         break;
6877                 }
6878
6879                 /*
6880                  * Re-access and dirty the bucket to calculate metaecc.
6881                  * Because we may extend the transaction in reflink_xattr_header
6882                  * which will let the already accessed block gone.
6883                  */
6884                 ret = ocfs2_xattr_bucket_journal_access(handle,
6885                                                 args->new_bucket,
6886                                                 OCFS2_JOURNAL_ACCESS_WRITE);
6887                 if (ret) {
6888                         mlog_errno(ret);
6889                         break;
6890                 }
6891
6892                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6893
6894                 ocfs2_xattr_bucket_relse(args->old_bucket);
6895                 ocfs2_xattr_bucket_relse(args->new_bucket);
6896         }
6897
6898         ocfs2_xattr_bucket_relse(args->old_bucket);
6899         ocfs2_xattr_bucket_relse(args->new_bucket);
6900         return ret;
6901 }
6902
6903 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6904                                 struct inode *inode,
6905                                 struct ocfs2_reflink_xattr_tree_args *args,
6906                                 struct ocfs2_extent_tree *et,
6907                                 struct ocfs2_alloc_context *meta_ac,
6908                                 struct ocfs2_alloc_context *data_ac,
6909                                 u64 blkno, u32 cpos, u32 len)
6910 {
6911         int ret, first_inserted = 0;
6912         u32 p_cluster, num_clusters, reflink_cpos = 0;
6913         u64 new_blkno;
6914         unsigned int num_buckets, reflink_buckets;
6915         unsigned int bpc =
6916                 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6917
6918         ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6919         if (ret) {
6920                 mlog_errno(ret);
6921                 goto out;
6922         }
6923         num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6924         ocfs2_xattr_bucket_relse(args->old_bucket);
6925
6926         while (len && num_buckets) {
6927                 ret = ocfs2_claim_clusters(handle, data_ac,
6928                                            1, &p_cluster, &num_clusters);
6929                 if (ret) {
6930                         mlog_errno(ret);
6931                         goto out;
6932                 }
6933
6934                 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6935                 reflink_buckets = min(num_buckets, bpc * num_clusters);
6936
6937                 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6938                                                  new_blkno, num_clusters,
6939                                                  &reflink_cpos, reflink_buckets,
6940                                                  meta_ac, data_ac, args);
6941                 if (ret) {
6942                         mlog_errno(ret);
6943                         goto out;
6944                 }
6945
6946                 /*
6947                  * For the 1st allocated cluster, we make it use the same cpos
6948                  * so that the xattr tree looks the same as the original one
6949                  * in the most case.
6950                  */
6951                 if (!first_inserted) {
6952                         reflink_cpos = cpos;
6953                         first_inserted = 1;
6954                 }
6955                 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6956                                           num_clusters, 0, meta_ac);
6957                 if (ret)
6958                         mlog_errno(ret);
6959
6960                 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6961                                                   num_clusters, reflink_cpos);
6962
6963                 len -= num_clusters;
6964                 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6965                 num_buckets -= reflink_buckets;
6966         }
6967 out:
6968         return ret;
6969 }
6970
6971 /*
6972  * Create the same xattr extent record in the new inode's xattr tree.
6973  */
6974 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6975                                    struct buffer_head *root_bh,
6976                                    u64 blkno,
6977                                    u32 cpos,
6978                                    u32 len,
6979                                    void *para)
6980 {
6981         int ret, credits = 0;
6982         handle_t *handle;
6983         struct ocfs2_reflink_xattr_tree_args *args =
6984                         (struct ocfs2_reflink_xattr_tree_args *)para;
6985         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6986         struct ocfs2_alloc_context *meta_ac = NULL;
6987         struct ocfs2_alloc_context *data_ac = NULL;
6988         struct ocfs2_extent_tree et;
6989
6990         trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
6991
6992         ocfs2_init_xattr_tree_extent_tree(&et,
6993                                           INODE_CACHE(args->reflink->new_inode),
6994                                           args->new_blk_bh);
6995
6996         ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
6997                                                       len, &credits,
6998                                                       &meta_ac, &data_ac);
6999         if (ret) {
7000                 mlog_errno(ret);
7001                 goto out;
7002         }
7003
7004         handle = ocfs2_start_trans(osb, credits);
7005         if (IS_ERR(handle)) {
7006                 ret = PTR_ERR(handle);
7007                 mlog_errno(ret);
7008                 goto out;
7009         }
7010
7011         ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7012                                           meta_ac, data_ac,
7013                                           blkno, cpos, len);
7014         if (ret)
7015                 mlog_errno(ret);
7016
7017         ocfs2_commit_trans(osb, handle);
7018
7019 out:
7020         if (meta_ac)
7021                 ocfs2_free_alloc_context(meta_ac);
7022         if (data_ac)
7023                 ocfs2_free_alloc_context(data_ac);
7024         return ret;
7025 }
7026
7027 /*
7028  * Create reflinked xattr buckets.
7029  * We will add bucket one by one, and refcount all the xattrs in the bucket
7030  * if they are stored outside.
7031  */
7032 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7033                                     struct buffer_head *blk_bh,
7034                                     struct buffer_head *new_blk_bh)
7035 {
7036         int ret;
7037         struct ocfs2_reflink_xattr_tree_args para;
7038
7039         memset(&para, 0, sizeof(para));
7040         para.reflink = args;
7041         para.old_blk_bh = blk_bh;
7042         para.new_blk_bh = new_blk_bh;
7043
7044         para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7045         if (!para.old_bucket) {
7046                 mlog_errno(-ENOMEM);
7047                 return -ENOMEM;
7048         }
7049
7050         para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7051         if (!para.new_bucket) {
7052                 ret = -ENOMEM;
7053                 mlog_errno(ret);
7054                 goto out;
7055         }
7056
7057         ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7058                                               ocfs2_reflink_xattr_rec,
7059                                               &para);
7060         if (ret)
7061                 mlog_errno(ret);
7062
7063 out:
7064         ocfs2_xattr_bucket_free(para.old_bucket);
7065         ocfs2_xattr_bucket_free(para.new_bucket);
7066         return ret;
7067 }
7068
7069 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7070                                         struct buffer_head *blk_bh)
7071 {
7072         int ret, indexed = 0;
7073         struct buffer_head *new_blk_bh = NULL;
7074         struct ocfs2_xattr_block *xb =
7075                         (struct ocfs2_xattr_block *)blk_bh->b_data;
7076
7077
7078         if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7079                 indexed = 1;
7080
7081         ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7082                                              &new_blk_bh, indexed);
7083         if (ret) {
7084                 mlog_errno(ret);
7085                 goto out;
7086         }
7087
7088         if (!indexed)
7089                 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7090         else
7091                 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7092         if (ret)
7093                 mlog_errno(ret);
7094
7095 out:
7096         brelse(new_blk_bh);
7097         return ret;
7098 }
7099
7100 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7101 {
7102         int type = ocfs2_xattr_get_type(xe);
7103
7104         return type != OCFS2_XATTR_INDEX_SECURITY &&
7105                type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7106                type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7107 }
7108
7109 int ocfs2_reflink_xattrs(struct inode *old_inode,
7110                          struct buffer_head *old_bh,
7111                          struct inode *new_inode,
7112                          struct buffer_head *new_bh,
7113                          bool preserve_security)
7114 {
7115         int ret;
7116         struct ocfs2_xattr_reflink args;
7117         struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7118         struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7119         struct buffer_head *blk_bh = NULL;
7120         struct ocfs2_cached_dealloc_ctxt dealloc;
7121         struct ocfs2_refcount_tree *ref_tree;
7122         struct buffer_head *ref_root_bh = NULL;
7123
7124         ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7125                                        le64_to_cpu(di->i_refcount_loc),
7126                                        1, &ref_tree, &ref_root_bh);
7127         if (ret) {
7128                 mlog_errno(ret);
7129                 goto out;
7130         }
7131
7132         ocfs2_init_dealloc_ctxt(&dealloc);
7133
7134         args.old_inode = old_inode;
7135         args.new_inode = new_inode;
7136         args.old_bh = old_bh;
7137         args.new_bh = new_bh;
7138         args.ref_ci = &ref_tree->rf_ci;
7139         args.ref_root_bh = ref_root_bh;
7140         args.dealloc = &dealloc;
7141         if (preserve_security)
7142                 args.xattr_reflinked = NULL;
7143         else
7144                 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7145
7146         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7147                 ret = ocfs2_reflink_xattr_inline(&args);
7148                 if (ret) {
7149                         mlog_errno(ret);
7150                         goto out_unlock;
7151                 }
7152         }
7153
7154         if (!di->i_xattr_loc)
7155                 goto out_unlock;
7156
7157         ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7158                                      &blk_bh);
7159         if (ret < 0) {
7160                 mlog_errno(ret);
7161                 goto out_unlock;
7162         }
7163
7164         ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7165         if (ret)
7166                 mlog_errno(ret);
7167
7168         brelse(blk_bh);
7169
7170 out_unlock:
7171         ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7172                                    ref_tree, 1);
7173         brelse(ref_root_bh);
7174
7175         if (ocfs2_dealloc_has_cluster(&dealloc)) {
7176                 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7177                 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7178         }
7179
7180 out:
7181         return ret;
7182 }
7183
7184 /*
7185  * Initialize security and acl for a already created inode.
7186  * Used for reflink a non-preserve-security file.
7187  *
7188  * It uses common api like ocfs2_xattr_set, so the caller
7189  * must not hold any lock expect i_mutex.
7190  */
7191 int ocfs2_init_security_and_acl(struct inode *dir,
7192                                 struct inode *inode,
7193                                 const struct qstr *qstr,
7194                                 struct posix_acl *default_acl,
7195                                 struct posix_acl *acl)
7196 {
7197         struct buffer_head *dir_bh = NULL;
7198         int ret = 0;
7199
7200         ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7201         if (ret) {
7202                 mlog_errno(ret);
7203                 goto leave;
7204         }
7205
7206         ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7207         if (ret) {
7208                 mlog_errno(ret);
7209                 goto leave;
7210         }
7211
7212         if (!ret && default_acl)
7213                 ret = ocfs2_iop_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
7214         if (!ret && acl)
7215                 ret = ocfs2_iop_set_acl(inode, acl, ACL_TYPE_ACCESS);
7216
7217         ocfs2_inode_unlock(dir, 0);
7218         brelse(dir_bh);
7219 leave:
7220         return ret;
7221 }
7222 /*
7223  * 'security' attributes support
7224  */
7225 static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
7226                                         size_t list_size, const char *name,
7227                                         size_t name_len, int type)
7228 {
7229         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7230         const size_t total_len = prefix_len + name_len + 1;
7231
7232         if (list && total_len <= list_size) {
7233                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7234                 memcpy(list + prefix_len, name, name_len);
7235                 list[prefix_len + name_len] = '\0';
7236         }
7237         return total_len;
7238 }
7239
7240 static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7241                                     void *buffer, size_t size, int type)
7242 {
7243         if (strcmp(name, "") == 0)
7244                 return -EINVAL;
7245         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7246                                name, buffer, size);
7247 }
7248
7249 static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7250                 const void *value, size_t size, int flags, int type)
7251 {
7252         if (strcmp(name, "") == 0)
7253                 return -EINVAL;
7254
7255         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7256                                name, value, size, flags);
7257 }
7258
7259 int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7260                      void *fs_info)
7261 {
7262         const struct xattr *xattr;
7263         int err = 0;
7264
7265         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7266                 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7267                                       xattr->name, xattr->value,
7268                                       xattr->value_len, XATTR_CREATE);
7269                 if (err)
7270                         break;
7271         }
7272         return err;
7273 }
7274
7275 int ocfs2_init_security_get(struct inode *inode,
7276                             struct inode *dir,
7277                             const struct qstr *qstr,
7278                             struct ocfs2_security_xattr_info *si)
7279 {
7280         /* check whether ocfs2 support feature xattr */
7281         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7282                 return -EOPNOTSUPP;
7283         if (si)
7284                 return security_old_inode_init_security(inode, dir, qstr,
7285                                                         &si->name, &si->value,
7286                                                         &si->value_len);
7287
7288         return security_inode_init_security(inode, dir, qstr,
7289                                             &ocfs2_initxattrs, NULL);
7290 }
7291
7292 int ocfs2_init_security_set(handle_t *handle,
7293                             struct inode *inode,
7294                             struct buffer_head *di_bh,
7295                             struct ocfs2_security_xattr_info *si,
7296                             struct ocfs2_alloc_context *xattr_ac,
7297                             struct ocfs2_alloc_context *data_ac)
7298 {
7299         return ocfs2_xattr_set_handle(handle, inode, di_bh,
7300                                      OCFS2_XATTR_INDEX_SECURITY,
7301                                      si->name, si->value, si->value_len, 0,
7302                                      xattr_ac, data_ac);
7303 }
7304
7305 const struct xattr_handler ocfs2_xattr_security_handler = {
7306         .prefix = XATTR_SECURITY_PREFIX,
7307         .list   = ocfs2_xattr_security_list,
7308         .get    = ocfs2_xattr_security_get,
7309         .set    = ocfs2_xattr_security_set,
7310 };
7311
7312 /*
7313  * 'trusted' attributes support
7314  */
7315 static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
7316                                        size_t list_size, const char *name,
7317                                        size_t name_len, int type)
7318 {
7319         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7320         const size_t total_len = prefix_len + name_len + 1;
7321
7322         if (list && total_len <= list_size) {
7323                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7324                 memcpy(list + prefix_len, name, name_len);
7325                 list[prefix_len + name_len] = '\0';
7326         }
7327         return total_len;
7328 }
7329
7330 static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7331                 void *buffer, size_t size, int type)
7332 {
7333         if (strcmp(name, "") == 0)
7334                 return -EINVAL;
7335         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7336                                name, buffer, size);
7337 }
7338
7339 static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7340                 const void *value, size_t size, int flags, int type)
7341 {
7342         if (strcmp(name, "") == 0)
7343                 return -EINVAL;
7344
7345         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7346                                name, value, size, flags);
7347 }
7348
7349 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7350         .prefix = XATTR_TRUSTED_PREFIX,
7351         .list   = ocfs2_xattr_trusted_list,
7352         .get    = ocfs2_xattr_trusted_get,
7353         .set    = ocfs2_xattr_trusted_set,
7354 };
7355
7356 /*
7357  * 'user' attributes support
7358  */
7359 static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
7360                                     size_t list_size, const char *name,
7361                                     size_t name_len, int type)
7362 {
7363         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7364         const size_t total_len = prefix_len + name_len + 1;
7365         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7366
7367         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7368                 return 0;
7369
7370         if (list && total_len <= list_size) {
7371                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7372                 memcpy(list + prefix_len, name, name_len);
7373                 list[prefix_len + name_len] = '\0';
7374         }
7375         return total_len;
7376 }
7377
7378 static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7379                 void *buffer, size_t size, int type)
7380 {
7381         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7382
7383         if (strcmp(name, "") == 0)
7384                 return -EINVAL;
7385         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7386                 return -EOPNOTSUPP;
7387         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
7388                                buffer, size);
7389 }
7390
7391 static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7392                 const void *value, size_t size, int flags, int type)
7393 {
7394         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7395
7396         if (strcmp(name, "") == 0)
7397                 return -EINVAL;
7398         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7399                 return -EOPNOTSUPP;
7400
7401         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7402                                name, value, size, flags);
7403 }
7404
7405 const struct xattr_handler ocfs2_xattr_user_handler = {
7406         .prefix = XATTR_USER_PREFIX,
7407         .list   = ocfs2_xattr_user_list,
7408         .get    = ocfs2_xattr_user_get,
7409         .set    = ocfs2_xattr_user_set,
7410 };