fs/ocfs2/xattr.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * xattr.c
   4  *
   5  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   6  *
   7  * CREDITS:
   8  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
   9  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  10  */
  11
  12 #include <linux/capability.h>
  13 #include <linux/fs.h>
  14 #include <linux/types.h>
  15 #include <linux/slab.h>
  16 #include <linux/highmem.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/uio.h>
  19 #include <linux/sched.h>
  20 #include <linux/splice.h>
  21 #include <linux/mount.h>
  22 #include <linux/writeback.h>
  23 #include <linux/falloc.h>
  24 #include <linux/sort.h>
  25 #include <linux/init.h>
  26 #include <linux/module.h>
  27 #include <linux/string.h>
  28 #include <linux/security.h>
  29
  30 #include <cluster/masklog.h>
  31
  32 #include "ocfs2.h"
  33 #include "alloc.h"
  34 #include "blockcheck.h"
  35 #include "dlmglue.h"
  36 #include "file.h"
  37 #include "symlink.h"
  38 #include "sysfile.h"
  39 #include "inode.h"
  40 #include "journal.h"
  41 #include "ocfs2_fs.h"
  42 #include "suballoc.h"
  43 #include "uptodate.h"
  44 #include "buffer_head_io.h"
  45 #include "super.h"
  46 #include "xattr.h"
  47 #include "refcounttree.h"
  48 #include "acl.h"
  49 #include "ocfs2_trace.h"
  50
  51 struct ocfs2_xattr_def_value_root {
  52         struct ocfs2_xattr_value_root   xv;
  53         struct ocfs2_extent_rec         er;
  54 };
  55
  56 struct ocfs2_xattr_bucket {
  57         /* The inode these xattrs are associated with */
  58         struct inode *bu_inode;
  59
  60         /* The actual buffers that make up the bucket */
  61         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  62
  63         /* How many blocks make up one bucket for this filesystem */
  64         int bu_blocks;
  65 };
  66
  67 struct ocfs2_xattr_set_ctxt {
  68         handle_t *handle;
  69         struct ocfs2_alloc_context *meta_ac;
  70         struct ocfs2_alloc_context *data_ac;
  71         struct ocfs2_cached_dealloc_ctxt dealloc;
  72         int set_abort;
  73 };
  74
  75 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  76 #define OCFS2_XATTR_INLINE_SIZE 80
  77 #define OCFS2_XATTR_HEADER_GAP  4
  78 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  79                                          - sizeof(struct ocfs2_xattr_header) \
  80                                          - OCFS2_XATTR_HEADER_GAP)
  81 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  82                                          - sizeof(struct ocfs2_xattr_block) \
  83                                          - sizeof(struct ocfs2_xattr_header) \
  84                                          - OCFS2_XATTR_HEADER_GAP)
  85
  86 static struct ocfs2_xattr_def_value_root def_xv = {
  87         .xv.xr_list.l_count = cpu_to_le16(1),
  88 };
  89
  90 const struct xattr_handler *ocfs2_xattr_handlers[] = {
  91         &ocfs2_xattr_user_handler,
  92         &posix_acl_access_xattr_handler,
  93         &posix_acl_default_xattr_handler,
  94         &ocfs2_xattr_trusted_handler,
  95         &ocfs2_xattr_security_handler,
  96         NULL
  97 };
  98
  99 static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 100         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 101         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 102                                         = &posix_acl_access_xattr_handler,
 103         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 104                                         = &posix_acl_default_xattr_handler,
 105         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 106         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 107 };
 108
 109 struct ocfs2_xattr_info {
 110         int             xi_name_index;
 111         const char      *xi_name;
 112         int             xi_name_len;
 113         const void      *xi_value;
 114         size_t          xi_value_len;
 115 };
 116
 117 struct ocfs2_xattr_search {
 118         struct buffer_head *inode_bh;
 119         /*
 120          * xattr_bh point to the block buffer head which has extended attribute
 121          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 122          */
 123         struct buffer_head *xattr_bh;
 124         struct ocfs2_xattr_header *header;
 125         struct ocfs2_xattr_bucket *bucket;
 126         void *base;
 127         void *end;
 128         struct ocfs2_xattr_entry *here;
 129         int not_found;
 130 };
 131
 132 /* Operations on struct ocfs2_xa_entry */
 133 struct ocfs2_xa_loc;
 134 struct ocfs2_xa_loc_operations {
 135         /*
 136          * Journal functions
 137          */
 138         int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
 139                                   int type);
 140         void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
 141
 142         /*
 143          * Return a pointer to the appropriate buffer in loc->xl_storage
 144          * at the given offset from loc->xl_header.
 145          */
 146         void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
 147
 148         /* Can we reuse the existing entry for the new value? */
 149         int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
 150                              struct ocfs2_xattr_info *xi);
 151
 152         /* How much space is needed for the new value? */
 153         int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
 154                                struct ocfs2_xattr_info *xi);
 155
 156         /*
 157          * Return the offset of the first name+value pair.  This is
 158          * the start of our downward-filling free space.
 159          */
 160         int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
 161
 162         /*
 163          * Remove the name+value at this location.  Do whatever is
 164          * appropriate with the remaining name+value pairs.
 165          */
 166         void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
 167
 168         /* Fill xl_entry with a new entry */
 169         void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
 170
 171         /* Add name+value storage to an entry */
 172         void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
 173
 174         /*
 175          * Initialize the value buf's access and bh fields for this entry.
 176          * ocfs2_xa_fill_value_buf() will handle the xv pointer.
 177          */
 178         void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
 179                                    struct ocfs2_xattr_value_buf *vb);
 180 };
 181
 182 /*
 183  * Describes an xattr entry location.  This is a memory structure
 184  * tracking the on-disk structure.
 185  */
 186 struct ocfs2_xa_loc {
 187         /* This xattr belongs to this inode */
 188         struct inode *xl_inode;
 189
 190         /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
 191         struct ocfs2_xattr_header *xl_header;
 192
 193         /* Bytes from xl_header to the end of the storage */
 194         int xl_size;
 195
 196         /*
 197          * The ocfs2_xattr_entry this location describes.  If this is
 198          * NULL, this location describes the on-disk structure where it
 199          * would have been.
 200          */
 201         struct ocfs2_xattr_entry *xl_entry;
 202
 203         /*
 204          * Internal housekeeping
 205          */
 206
 207         /* Buffer(s) containing this entry */
 208         void *xl_storage;
 209
 210         /* Operations on the storage backing this location */
 211         const struct ocfs2_xa_loc_operations *xl_ops;
 212 };
 213
 214 /*
 215  * Convenience functions to calculate how much space is needed for a
 216  * given name+value pair
 217  */
 218 static int namevalue_size(int name_len, uint64_t value_len)
 219 {
 220         if (value_len > OCFS2_XATTR_INLINE_SIZE)
 221                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 222         else
 223                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 224 }
 225
 226 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
 227 {
 228         return namevalue_size(xi->xi_name_len, xi->xi_value_len);
 229 }
 230
 231 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
 232 {
 233         u64 value_len = le64_to_cpu(xe->xe_value_size);
 234
 235         BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
 236                ocfs2_xattr_is_local(xe));
 237         return namevalue_size(xe->xe_name_len, value_len);
 238 }
 239
 240
 241 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 242                                              struct ocfs2_xattr_header *xh,
 243                                              int index,
 244                                              int *block_off,
 245                                              int *new_offset);
 246
 247 static int ocfs2_xattr_block_find(struct inode *inode,
 248                                   int name_index,
 249                                   const char *name,
 250                                   struct ocfs2_xattr_search *xs);
 251 static int ocfs2_xattr_index_block_find(struct inode *inode,
 252                                         struct buffer_head *root_bh,
 253                                         int name_index,
 254                                         const char *name,
 255                                         struct ocfs2_xattr_search *xs);
 256
 257 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 258                                         struct buffer_head *blk_bh,
 259                                         char *buffer,
 260                                         size_t buffer_size);
 261
 262 static int ocfs2_xattr_create_index_block(struct inode *inode,
 263                                           struct ocfs2_xattr_search *xs,
 264                                           struct ocfs2_xattr_set_ctxt *ctxt);
 265
 266 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 267                                              struct ocfs2_xattr_info *xi,
 268                                              struct ocfs2_xattr_search *xs,
 269                                              struct ocfs2_xattr_set_ctxt *ctxt);
 270
 271 typedef int (xattr_tree_rec_func)(struct inode *inode,
 272                                   struct buffer_head *root_bh,
 273                                   u64 blkno, u32 cpos, u32 len, void *para);
 274 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
 275                                            struct buffer_head *root_bh,
 276                                            xattr_tree_rec_func *rec_func,
 277                                            void *para);
 278 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
 279                                         struct ocfs2_xattr_bucket *bucket,
 280                                         void *para);
 281 static int ocfs2_rm_xattr_cluster(struct inode *inode,
 282                                   struct buffer_head *root_bh,
 283                                   u64 blkno,
 284                                   u32 cpos,
 285                                   u32 len,
 286                                   void *para);
 287
 288 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 289                                   u64 src_blk, u64 last_blk, u64 to_blk,
 290                                   unsigned int start_bucket,
 291                                   u32 *first_hash);
 292 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
 293                                         struct ocfs2_dinode *di,
 294                                         struct ocfs2_xattr_info *xi,
 295                                         struct ocfs2_xattr_search *xis,
 296                                         struct ocfs2_xattr_search *xbs,
 297                                         struct ocfs2_refcount_tree **ref_tree,
 298                                         int *meta_need,
 299                                         int *credits);
 300 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
 301                                            struct ocfs2_xattr_bucket *bucket,
 302                                            int offset,
 303                                            struct ocfs2_xattr_value_root **xv,
 304                                            struct buffer_head **bh);
 305
 306 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 307 {
 308         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 309 }
 310
 311 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 312 {
 313         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 314 }
 315
 316 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 317 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 318 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 319
 320 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 321 {
 322         struct ocfs2_xattr_bucket *bucket;
 323         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 324
 325         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 326
 327         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 328         if (bucket) {
 329                 bucket->bu_inode = inode;
 330                 bucket->bu_blocks = blks;
 331         }
 332
 333         return bucket;
 334 }
 335
 336 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 337 {
 338         int i;
 339
 340         for (i = 0; i < bucket->bu_blocks; i++) {
 341                 brelse(bucket->bu_bhs[i]);
 342                 bucket->bu_bhs[i] = NULL;
 343         }
 344 }
 345
 346 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 347 {
 348         if (bucket) {
 349                 ocfs2_xattr_bucket_relse(bucket);
 350                 bucket->bu_inode = NULL;
 351                 kfree(bucket);
 352         }
 353 }
 354
 355 /*
 356  * A bucket that has never been written to disk doesn't need to be
 357  * read.  We just need the buffer_heads.  Don't call this for
 358  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 359  * them fully.
 360  */
 361 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 362                                    u64 xb_blkno, int new)
 363 {
 364         int i, rc = 0;
 365
 366         for (i = 0; i < bucket->bu_blocks; i++) {
 367                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 368                                               xb_blkno + i);
 369                 if (!bucket->bu_bhs[i]) {
 370                         rc = -ENOMEM;
 371                         mlog_errno(rc);
 372                         break;
 373                 }
 374
 375                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 376                                            bucket->bu_bhs[i])) {
 377                         if (new)
 378                                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 379                                                               bucket->bu_bhs[i]);
 380                         else {
 381                                 set_buffer_uptodate(bucket->bu_bhs[i]);
 382                                 ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 383                                                           bucket->bu_bhs[i]);
 384                         }
 385                 }
 386         }
 387
 388         if (rc)
 389                 ocfs2_xattr_bucket_relse(bucket);
 390         return rc;
 391 }
 392
 393 /* Read the xattr bucket at xb_blkno */
 394 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 395                                    u64 xb_blkno)
 396 {
 397         int rc;
 398
 399         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 400                                bucket->bu_blocks, bucket->bu_bhs, 0,
 401                                NULL);
 402         if (!rc) {
 403                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 404                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 405                                                  bucket->bu_bhs,
 406                                                  bucket->bu_blocks,
 407                                                  &bucket_xh(bucket)->xh_check);
 408                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 409                 if (rc)
 410                         mlog_errno(rc);
 411         }
 412
 413         if (rc)
 414                 ocfs2_xattr_bucket_relse(bucket);
 415         return rc;
 416 }
 417
 418 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 419                                              struct ocfs2_xattr_bucket *bucket,
 420                                              int type)
 421 {
 422         int i, rc = 0;
 423
 424         for (i = 0; i < bucket->bu_blocks; i++) {
 425                 rc = ocfs2_journal_access(handle,
 426                                           INODE_CACHE(bucket->bu_inode),
 427                                           bucket->bu_bhs[i], type);
 428                 if (rc) {
 429                         mlog_errno(rc);
 430                         break;
 431                 }
 432         }
 433
 434         return rc;
 435 }
 436
 437 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 438                                              struct ocfs2_xattr_bucket *bucket)
 439 {
 440         int i;
 441
 442         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 443         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 444                                    bucket->bu_bhs, bucket->bu_blocks,
 445                                    &bucket_xh(bucket)->xh_check);
 446         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 447
 448         for (i = 0; i < bucket->bu_blocks; i++)
 449                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 450 }
 451
 452 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 453                                          struct ocfs2_xattr_bucket *src)
 454 {
 455         int i;
 456         int blocksize = src->bu_inode->i_sb->s_blocksize;
 457
 458         BUG_ON(dest->bu_blocks != src->bu_blocks);
 459         BUG_ON(dest->bu_inode != src->bu_inode);
 460
 461         for (i = 0; i < src->bu_blocks; i++) {
 462                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 463                        blocksize);
 464         }
 465 }
 466
 467 static int ocfs2_validate_xattr_block(struct super_block *sb,
 468                                       struct buffer_head *bh)
 469 {
 470         int rc;
 471         struct ocfs2_xattr_block *xb =
 472                 (struct ocfs2_xattr_block *)bh->b_data;
 473
 474         trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
 475
 476         BUG_ON(!buffer_uptodate(bh));
 477
 478         /*
 479          * If the ecc fails, we return the error but otherwise
 480          * leave the filesystem running.  We know any error is
 481          * local to this block.
 482          */
 483         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 484         if (rc)
 485                 return rc;
 486
 487         /*
 488          * Errors after here are fatal
 489          */
 490
 491         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 492                 return ocfs2_error(sb,
 493                                    "Extended attribute block #%llu has bad signature %.*s\n",
 494                                    (unsigned long long)bh->b_blocknr, 7,
 495                                    xb->xb_signature);
 496         }
 497
 498         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 499                 return ocfs2_error(sb,
 500                                    "Extended attribute block #%llu has an invalid xb_blkno of %llu\n",
 501                                    (unsigned long long)bh->b_blocknr,
 502                                    (unsigned long long)le64_to_cpu(xb->xb_blkno));
 503         }
 504
 505         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 506                 return ocfs2_error(sb,
 507                                    "Extended attribute block #%llu has an invalid xb_fs_generation of #%u\n",
 508                                    (unsigned long long)bh->b_blocknr,
 509                                    le32_to_cpu(xb->xb_fs_generation));
 510         }
 511
 512         return 0;
 513 }
 514
 515 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 516                                   struct buffer_head **bh)
 517 {
 518         int rc;
 519         struct buffer_head *tmp = *bh;
 520
 521         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 522                               ocfs2_validate_xattr_block);
 523
 524         /* If ocfs2_read_block() got us a new bh, pass it up. */
 525         if (!rc && !*bh)
 526                 *bh = tmp;
 527
 528         return rc;
 529 }
 530
 531 static inline const char *ocfs2_xattr_prefix(int name_index)
 532 {
 533         const struct xattr_handler *handler = NULL;
 534
 535         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 536                 handler = ocfs2_xattr_handler_map[name_index];
 537         return handler ? xattr_prefix(handler) : NULL;
 538 }
 539
 540 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 541                                  const char *name,
 542                                  int name_len)
 543 {
 544         /* Get hash value of uuid from super block */
 545         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 546         int i;
 547
 548         /* hash extended attribute name */
 549         for (i = 0; i < name_len; i++) {
 550                 hash = (hash << OCFS2_HASH_SHIFT) ^
 551                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 552                        *name++;
 553         }
 554
 555         return hash;
 556 }
 557
 558 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 559 {
 560         return namevalue_size(name_len, value_len) +
 561                 sizeof(struct ocfs2_xattr_entry);
 562 }
 563
 564 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 565 {
 566         return namevalue_size_xi(xi) +
 567                 sizeof(struct ocfs2_xattr_entry);
 568 }
 569
 570 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 571 {
 572         return namevalue_size_xe(xe) +
 573                 sizeof(struct ocfs2_xattr_entry);
 574 }
 575
 576 int ocfs2_calc_security_init(struct inode *dir,
 577                              struct ocfs2_security_xattr_info *si,
 578                              int *want_clusters,
 579                              int *xattr_credits,
 580                              struct ocfs2_alloc_context **xattr_ac)
 581 {
 582         int ret = 0;
 583         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 584         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 585                                                  si->value_len);
 586
 587         /*
 588          * The max space of security xattr taken inline is
 589          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 590          * So reserve one metadata block for it is ok.
 591          */
 592         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 593             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 594                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 595                 if (ret) {
 596                         mlog_errno(ret);
 597                         return ret;
 598                 }
 599                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 600         }
 601
 602         /* reserve clusters for xattr value which will be set in B tree*/
 603         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 604                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 605                                                             si->value_len);
 606
 607                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 608                                                            new_clusters);
 609                 *want_clusters += new_clusters;
 610         }
 611         return ret;
 612 }
 613
 614 int ocfs2_calc_xattr_init(struct inode *dir,
 615                           struct buffer_head *dir_bh,
 616                           umode_t mode,
 617                           struct ocfs2_security_xattr_info *si,
 618                           int *want_clusters,
 619                           int *xattr_credits,
 620                           int *want_meta)
 621 {
 622         int ret = 0;
 623         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 624         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 625
 626         if (si->enable)
 627                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 628                                                      si->value_len);
 629
 630         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 631                 down_read(&OCFS2_I(dir)->ip_xattr_sem);
 632                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 633                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 634                                         "", NULL, 0);
 635                 up_read(&OCFS2_I(dir)->ip_xattr_sem);
 636                 if (acl_len > 0) {
 637                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 638                         if (S_ISDIR(mode))
 639                                 a_size <<= 1;
 640                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 641                         ret = acl_len;
 642                         mlog_errno(ret);
 643                         return ret;
 644                 }
 645         }
 646
 647         if (!(s_size + a_size))
 648                 return ret;
 649
 650         /*
 651          * The max space of security xattr taken inline is
 652          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 653          * The max space of acl xattr taken inline is
 654          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 655          * when blocksize = 512, may reserve one more cluser for
 656          * xattr bucket, otherwise reserve one metadata block
 657          * for them is ok.
 658          * If this is a new directory with inline data,
 659          * we choose to reserve the entire inline area for
 660          * directory contents and force an external xattr block.
 661          */
 662         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 663             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 664             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 665                 *want_meta = *want_meta + 1;
 666                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 667         }
 668
 669         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 670             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 671                 *want_clusters += 1;
 672                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 673         }
 674
 675         /*
 676          * reserve credits and clusters for xattrs which has large value
 677          * and have to be set outside
 678          */
 679         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 680                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 681                                                         si->value_len);
 682                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 683                                                            new_clusters);
 684                 *want_clusters += new_clusters;
 685         }
 686         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 687             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 688                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 689                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 690                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 691                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 692                                                            new_clusters);
 693                 *want_clusters += new_clusters;
 694         }
 695
 696         return ret;
 697 }
 698
 699 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 700                                          u32 clusters_to_add,
 701                                          struct ocfs2_xattr_value_buf *vb,
 702                                          struct ocfs2_xattr_set_ctxt *ctxt)
 703 {
 704         int status = 0, credits;
 705         handle_t *handle = ctxt->handle;
 706         enum ocfs2_alloc_restarted why;
 707         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 708         struct ocfs2_extent_tree et;
 709
 710         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 711
 712         while (clusters_to_add) {
 713                 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 714
 715                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 716                                        OCFS2_JOURNAL_ACCESS_WRITE);
 717                 if (status < 0) {
 718                         mlog_errno(status);
 719                         break;
 720                 }
 721
 722                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 723                 status = ocfs2_add_clusters_in_btree(handle,
 724                                                      &et,
 725                                                      &logical_start,
 726                                                      clusters_to_add,
 727                                                      0,
 728                                                      ctxt->data_ac,
 729                                                      ctxt->meta_ac,
 730                                                      &why);
 731                 if ((status < 0) && (status != -EAGAIN)) {
 732                         if (status != -ENOSPC)
 733                                 mlog_errno(status);
 734                         break;
 735                 }
 736
 737                 ocfs2_journal_dirty(handle, vb->vb_bh);
 738
 739                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 740                                          prev_clusters;
 741
 742                 if (why != RESTART_NONE && clusters_to_add) {
 743                         /*
 744                          * We can only fail in case the alloc file doesn't give
 745                          * up enough clusters.
 746                          */
 747                         BUG_ON(why == RESTART_META);
 748
 749                         credits = ocfs2_calc_extend_credits(inode->i_sb,
 750                                                             &vb->vb_xv->xr_list);
 751                         status = ocfs2_extend_trans(handle, credits);
 752                         if (status < 0) {
 753                                 status = -ENOMEM;
 754                                 mlog_errno(status);
 755                                 break;
 756                         }
 757                 }
 758         }
 759
 760         return status;
 761 }
 762
 763 static int __ocfs2_remove_xattr_range(struct inode *inode,
 764                                       struct ocfs2_xattr_value_buf *vb,
 765                                       u32 cpos, u32 phys_cpos, u32 len,
 766                                       unsigned int ext_flags,
 767                                       struct ocfs2_xattr_set_ctxt *ctxt)
 768 {
 769         int ret;
 770         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 771         handle_t *handle = ctxt->handle;
 772         struct ocfs2_extent_tree et;
 773
 774         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 775
 776         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 777                             OCFS2_JOURNAL_ACCESS_WRITE);
 778         if (ret) {
 779                 mlog_errno(ret);
 780                 goto out;
 781         }
 782
 783         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 784                                   &ctxt->dealloc);
 785         if (ret) {
 786                 mlog_errno(ret);
 787                 goto out;
 788         }
 789
 790         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 791         ocfs2_journal_dirty(handle, vb->vb_bh);
 792
 793         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 794                 ret = ocfs2_decrease_refcount(inode, handle,
 795                                         ocfs2_blocks_to_clusters(inode->i_sb,
 796                                                                  phys_blkno),
 797                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 798         else
 799                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 800                                                   phys_blkno, len);
 801         if (ret)
 802                 mlog_errno(ret);
 803
 804 out:
 805         return ret;
 806 }
 807
 808 static int ocfs2_xattr_shrink_size(struct inode *inode,
 809                                    u32 old_clusters,
 810                                    u32 new_clusters,
 811                                    struct ocfs2_xattr_value_buf *vb,
 812                                    struct ocfs2_xattr_set_ctxt *ctxt)
 813 {
 814         int ret = 0;
 815         unsigned int ext_flags;
 816         u32 trunc_len, cpos, phys_cpos, alloc_size;
 817         u64 block;
 818
 819         if (old_clusters <= new_clusters)
 820                 return 0;
 821
 822         cpos = new_clusters;
 823         trunc_len = old_clusters - new_clusters;
 824         while (trunc_len) {
 825                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 826                                                &alloc_size,
 827                                                &vb->vb_xv->xr_list, &ext_flags);
 828                 if (ret) {
 829                         mlog_errno(ret);
 830                         goto out;
 831                 }
 832
 833                 if (alloc_size > trunc_len)
 834                         alloc_size = trunc_len;
 835
 836                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 837                                                  phys_cpos, alloc_size,
 838                                                  ext_flags, ctxt);
 839                 if (ret) {
 840                         mlog_errno(ret);
 841                         goto out;
 842                 }
 843
 844                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 845                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 846                                                        block, alloc_size);
 847                 cpos += alloc_size;
 848                 trunc_len -= alloc_size;
 849         }
 850
 851 out:
 852         return ret;
 853 }
 854
 855 static int ocfs2_xattr_value_truncate(struct inode *inode,
 856                                       struct ocfs2_xattr_value_buf *vb,
 857                                       int len,
 858                                       struct ocfs2_xattr_set_ctxt *ctxt)
 859 {
 860         int ret;
 861         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 862         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 863
 864         if (new_clusters == old_clusters)
 865                 return 0;
 866
 867         if (new_clusters > old_clusters)
 868                 ret = ocfs2_xattr_extend_allocation(inode,
 869                                                     new_clusters - old_clusters,
 870                                                     vb, ctxt);
 871         else
 872                 ret = ocfs2_xattr_shrink_size(inode,
 873                                               old_clusters, new_clusters,
 874                                               vb, ctxt);
 875
 876         return ret;
 877 }
 878
 879 static int ocfs2_xattr_list_entry(struct super_block *sb,
 880                                   char *buffer, size_t size,
 881                                   size_t *result, int type,
 882                                   const char *name, int name_len)
 883 {
 884         char *p = buffer + *result;
 885         const char *prefix;
 886         int prefix_len;
 887         int total_len;
 888
 889         switch(type) {
 890         case OCFS2_XATTR_INDEX_USER:
 891                 if (OCFS2_SB(sb)->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
 892                         return 0;
 893                 break;
 894
 895         case OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS:
 896         case OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT:
 897                 if (!(sb->s_flags & SB_POSIXACL))
 898                         return 0;
 899                 break;
 900
 901         case OCFS2_XATTR_INDEX_TRUSTED:
 902                 if (!capable(CAP_SYS_ADMIN))
 903                         return 0;
 904                 break;
 905         }
 906
 907         prefix = ocfs2_xattr_prefix(type);
 908         if (!prefix)
 909                 return 0;
 910         prefix_len = strlen(prefix);
 911         total_len = prefix_len + name_len + 1;
 912         *result += total_len;
 913
 914         /* we are just looking for how big our buffer needs to be */
 915         if (!size)
 916                 return 0;
 917
 918         if (*result > size)
 919                 return -ERANGE;
 920
 921         memcpy(p, prefix, prefix_len);
 922         memcpy(p + prefix_len, name, name_len);
 923         p[prefix_len + name_len] = '\0';
 924
 925         return 0;
 926 }
 927
 928 static int ocfs2_xattr_list_entries(struct inode *inode,
 929                                     struct ocfs2_xattr_header *header,
 930                                     char *buffer, size_t buffer_size)
 931 {
 932         size_t result = 0;
 933         int i, type, ret;
 934         const char *name;
 935
 936         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 937                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 938                 type = ocfs2_xattr_get_type(entry);
 939                 name = (const char *)header +
 940                         le16_to_cpu(entry->xe_name_offset);
 941
 942                 ret = ocfs2_xattr_list_entry(inode->i_sb,
 943                                              buffer, buffer_size,
 944                                              &result, type, name,
 945                                              entry->xe_name_len);
 946                 if (ret)
 947                         return ret;
 948         }
 949
 950         return result;
 951 }
 952
 953 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 954                                          struct ocfs2_dinode *di)
 955 {
 956         struct ocfs2_xattr_header *xh;
 957         int i;
 958
 959         xh = (struct ocfs2_xattr_header *)
 960                  ((void *)di + inode->i_sb->s_blocksize -
 961                  le16_to_cpu(di->i_xattr_inline_size));
 962
 963         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 964                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 965                         return 1;
 966
 967         return 0;
 968 }
 969
 970 static int ocfs2_xattr_ibody_list(struct inode *inode,
 971                                   struct ocfs2_dinode *di,
 972                                   char *buffer,
 973                                   size_t buffer_size)
 974 {
 975         struct ocfs2_xattr_header *header = NULL;
 976         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 977         int ret = 0;
 978
 979         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 980                 return ret;
 981
 982         header = (struct ocfs2_xattr_header *)
 983                  ((void *)di + inode->i_sb->s_blocksize -
 984                  le16_to_cpu(di->i_xattr_inline_size));
 985
 986         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 987
 988         return ret;
 989 }
 990
 991 static int ocfs2_xattr_block_list(struct inode *inode,
 992                                   struct ocfs2_dinode *di,
 993                                   char *buffer,
 994                                   size_t buffer_size)
 995 {
 996         struct buffer_head *blk_bh = NULL;
 997         struct ocfs2_xattr_block *xb;
 998         int ret = 0;
 999
1000         if (!di->i_xattr_loc)
1001                 return ret;
1002
1003         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
1004                                      &blk_bh);
1005         if (ret < 0) {
1006                 mlog_errno(ret);
1007                 return ret;
1008         }
1009
1010         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1011         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1012                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1013                 ret = ocfs2_xattr_list_entries(inode, header,
1014                                                buffer, buffer_size);
1015         } else
1016                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1017                                                    buffer, buffer_size);
1018
1019         brelse(blk_bh);
1020
1021         return ret;
1022 }
1023
1024 ssize_t ocfs2_listxattr(struct dentry *dentry,
1025                         char *buffer,
1026                         size_t size)
1027 {
1028         int ret = 0, i_ret = 0, b_ret = 0;
1029         struct buffer_head *di_bh = NULL;
1030         struct ocfs2_dinode *di = NULL;
1031         struct ocfs2_inode_info *oi = OCFS2_I(d_inode(dentry));
1032
1033         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1034                 return -EOPNOTSUPP;
1035
1036         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1037                 return ret;
1038
1039         ret = ocfs2_inode_lock(d_inode(dentry), &di_bh, 0);
1040         if (ret < 0) {
1041                 mlog_errno(ret);
1042                 return ret;
1043         }
1044
1045         di = (struct ocfs2_dinode *)di_bh->b_data;
1046
1047         down_read(&oi->ip_xattr_sem);
1048         i_ret = ocfs2_xattr_ibody_list(d_inode(dentry), di, buffer, size);
1049         if (i_ret < 0)
1050                 b_ret = 0;
1051         else {
1052                 if (buffer) {
1053                         buffer += i_ret;
1054                         size -= i_ret;
1055                 }
1056                 b_ret = ocfs2_xattr_block_list(d_inode(dentry), di,
1057                                                buffer, size);
1058                 if (b_ret < 0)
1059                         i_ret = 0;
1060         }
1061         up_read(&oi->ip_xattr_sem);
1062         ocfs2_inode_unlock(d_inode(dentry), 0);
1063
1064         brelse(di_bh);
1065
1066         return i_ret + b_ret;
1067 }
1068
1069 static int ocfs2_xattr_find_entry(int name_index,
1070                                   const char *name,
1071                                   struct ocfs2_xattr_search *xs)
1072 {
1073         struct ocfs2_xattr_entry *entry;
1074         size_t name_len;
1075         int i, cmp = 1;
1076
1077         if (name == NULL)
1078                 return -EINVAL;
1079
1080         name_len = strlen(name);
1081         entry = xs->here;
1082         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1083                 cmp = name_index - ocfs2_xattr_get_type(entry);
1084                 if (!cmp)
1085                         cmp = name_len - entry->xe_name_len;
1086                 if (!cmp)
1087                         cmp = memcmp(name, (xs->base +
1088                                      le16_to_cpu(entry->xe_name_offset)),
1089                                      name_len);
1090                 if (cmp == 0)
1091                         break;
1092                 entry += 1;
1093         }
1094         xs->here = entry;
1095
1096         return cmp ? -ENODATA : 0;
1097 }
1098
1099 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1100                                          struct ocfs2_xattr_value_root *xv,
1101                                          void *buffer,
1102                                          size_t len)
1103 {
1104         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1105         u64 blkno;
1106         int i, ret = 0;
1107         size_t cplen, blocksize;
1108         struct buffer_head *bh = NULL;
1109         struct ocfs2_extent_list *el;
1110
1111         el = &xv->xr_list;
1112         clusters = le32_to_cpu(xv->xr_clusters);
1113         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1114         blocksize = inode->i_sb->s_blocksize;
1115
1116         cpos = 0;
1117         while (cpos < clusters) {
1118                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1119                                                &num_clusters, el, NULL);
1120                 if (ret) {
1121                         mlog_errno(ret);
1122                         goto out;
1123                 }
1124
1125                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1126                 /* Copy ocfs2_xattr_value */
1127                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1128                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1129                                                &bh, NULL);
1130                         if (ret) {
1131                                 mlog_errno(ret);
1132                                 goto out;
1133                         }
1134
1135                         cplen = len >= blocksize ? blocksize : len;
1136                         memcpy(buffer, bh->b_data, cplen);
1137                         len -= cplen;
1138                         buffer += cplen;
1139
1140                         brelse(bh);
1141                         bh = NULL;
1142                         if (len == 0)
1143                                 break;
1144                 }
1145                 cpos += num_clusters;
1146         }
1147 out:
1148         return ret;
1149 }
1150
1151 static int ocfs2_xattr_ibody_get(struct inode *inode,
1152                                  int name_index,
1153                                  const char *name,
1154                                  void *buffer,
1155                                  size_t buffer_size,
1156                                  struct ocfs2_xattr_search *xs)
1157 {
1158         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1159         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1160         struct ocfs2_xattr_value_root *xv;
1161         size_t size;
1162         int ret = 0;
1163
1164         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1165                 return -ENODATA;
1166
1167         xs->end = (void *)di + inode->i_sb->s_blocksize;
1168         xs->header = (struct ocfs2_xattr_header *)
1169                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1170         xs->base = (void *)xs->header;
1171         xs->here = xs->header->xh_entries;
1172
1173         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1174         if (ret)
1175                 return ret;
1176         size = le64_to_cpu(xs->here->xe_value_size);
1177         if (buffer) {
1178                 if (size > buffer_size)
1179                         return -ERANGE;
1180                 if (ocfs2_xattr_is_local(xs->here)) {
1181                         memcpy(buffer, (void *)xs->base +
1182                                le16_to_cpu(xs->here->xe_name_offset) +
1183                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1184                 } else {
1185                         xv = (struct ocfs2_xattr_value_root *)
1186                                 (xs->base + le16_to_cpu(
1187                                  xs->here->xe_name_offset) +
1188                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1189                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1190                                                             buffer, size);
1191                         if (ret < 0) {
1192                                 mlog_errno(ret);
1193                                 return ret;
1194                         }
1195                 }
1196         }
1197
1198         return size;
1199 }
1200
1201 static int ocfs2_xattr_block_get(struct inode *inode,
1202                                  int name_index,
1203                                  const char *name,
1204                                  void *buffer,
1205                                  size_t buffer_size,
1206                                  struct ocfs2_xattr_search *xs)
1207 {
1208         struct ocfs2_xattr_block *xb;
1209         struct ocfs2_xattr_value_root *xv;
1210         size_t size;
1211         int ret = -ENODATA, name_offset, name_len, i;
1212         int block_off;
1213
1214         xs->bucket = ocfs2_xattr_bucket_new(inode);
1215         if (!xs->bucket) {
1216                 ret = -ENOMEM;
1217                 mlog_errno(ret);
1218                 goto cleanup;
1219         }
1220
1221         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1222         if (ret) {
1223                 mlog_errno(ret);
1224                 goto cleanup;
1225         }
1226
1227         if (xs->not_found) {
1228                 ret = -ENODATA;
1229                 goto cleanup;
1230         }
1231
1232         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1233         size = le64_to_cpu(xs->here->xe_value_size);
1234         if (buffer) {
1235                 ret = -ERANGE;
1236                 if (size > buffer_size)
1237                         goto cleanup;
1238
1239                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1240                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1241                 i = xs->here - xs->header->xh_entries;
1242
1243                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1244                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1245                                                                 bucket_xh(xs->bucket),
1246                                                                 i,
1247                                                                 &block_off,
1248                                                                 &name_offset);
1249                         if (ret) {
1250                                 mlog_errno(ret);
1251                                 goto cleanup;
1252                         }
1253                         xs->base = bucket_block(xs->bucket, block_off);
1254                 }
1255                 if (ocfs2_xattr_is_local(xs->here)) {
1256                         memcpy(buffer, (void *)xs->base +
1257                                name_offset + name_len, size);
1258                 } else {
1259                         xv = (struct ocfs2_xattr_value_root *)
1260                                 (xs->base + name_offset + name_len);
1261                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1262                                                             buffer, size);
1263                         if (ret < 0) {
1264                                 mlog_errno(ret);
1265                                 goto cleanup;
1266                         }
1267                 }
1268         }
1269         ret = size;
1270 cleanup:
1271         ocfs2_xattr_bucket_free(xs->bucket);
1272
1273         brelse(xs->xattr_bh);
1274         xs->xattr_bh = NULL;
1275         return ret;
1276 }
1277
1278 int ocfs2_xattr_get_nolock(struct inode *inode,
1279                            struct buffer_head *di_bh,
1280                            int name_index,
1281                            const char *name,
1282                            void *buffer,
1283                            size_t buffer_size)
1284 {
1285         int ret;
1286         struct ocfs2_dinode *di = NULL;
1287         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1288         struct ocfs2_xattr_search xis = {
1289                 .not_found = -ENODATA,
1290         };
1291         struct ocfs2_xattr_search xbs = {
1292                 .not_found = -ENODATA,
1293         };
1294
1295         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1296                 return -EOPNOTSUPP;
1297
1298         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1299                 return -ENODATA;
1300
1301         xis.inode_bh = xbs.inode_bh = di_bh;
1302         di = (struct ocfs2_dinode *)di_bh->b_data;
1303
1304         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1305                                     buffer_size, &xis);
1306         if (ret == -ENODATA && di->i_xattr_loc)
1307                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1308                                             buffer_size, &xbs);
1309
1310         return ret;
1311 }
1312
1313 /* ocfs2_xattr_get()
1314  *
1315  * Copy an extended attribute into the buffer provided.
1316  * Buffer is NULL to compute the size of buffer required.
1317  */
1318 static int ocfs2_xattr_get(struct inode *inode,
1319                            int name_index,
1320                            const char *name,
1321                            void *buffer,
1322                            size_t buffer_size)
1323 {
1324         int ret, had_lock;
1325         struct buffer_head *di_bh = NULL;
1326         struct ocfs2_lock_holder oh;
1327
1328         had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
1329         if (had_lock < 0) {
1330                 mlog_errno(had_lock);
1331                 return had_lock;
1332         }
1333         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1334         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1335                                      name, buffer, buffer_size);
1336         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1337
1338         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1339
1340         brelse(di_bh);
1341
1342         return ret;
1343 }
1344
1345 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1346                                            handle_t *handle,
1347                                            struct ocfs2_xattr_value_buf *vb,
1348                                            const void *value,
1349                                            int value_len)
1350 {
1351         int ret = 0, i, cp_len;
1352         u16 blocksize = inode->i_sb->s_blocksize;
1353         u32 p_cluster, num_clusters;
1354         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1355         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1356         u64 blkno;
1357         struct buffer_head *bh = NULL;
1358         unsigned int ext_flags;
1359         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1360
1361         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1362
1363         while (cpos < clusters) {
1364                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1365                                                &num_clusters, &xv->xr_list,
1366                                                &ext_flags);
1367                 if (ret) {
1368                         mlog_errno(ret);
1369                         goto out;
1370                 }
1371
1372                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1373
1374                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1375
1376                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1377                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1378                                                &bh, NULL);
1379                         if (ret) {
1380                                 mlog_errno(ret);
1381                                 goto out;
1382                         }
1383
1384                         ret = ocfs2_journal_access(handle,
1385                                                    INODE_CACHE(inode),
1386                                                    bh,
1387                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1388                         if (ret < 0) {
1389                                 mlog_errno(ret);
1390                                 goto out;
1391                         }
1392
1393                         cp_len = value_len > blocksize ? blocksize : value_len;
1394                         memcpy(bh->b_data, value, cp_len);
1395                         value_len -= cp_len;
1396                         value += cp_len;
1397                         if (cp_len < blocksize)
1398                                 memset(bh->b_data + cp_len, 0,
1399                                        blocksize - cp_len);
1400
1401                         ocfs2_journal_dirty(handle, bh);
1402                         brelse(bh);
1403                         bh = NULL;
1404
1405                         /*
1406                          * XXX: do we need to empty all the following
1407                          * blocks in this cluster?
1408                          */
1409                         if (!value_len)
1410                                 break;
1411                 }
1412                 cpos += num_clusters;
1413         }
1414 out:
1415         brelse(bh);
1416
1417         return ret;
1418 }
1419
1420 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1421                                        int num_entries)
1422 {
1423         int free_space;
1424
1425         if (!needed_space)
1426                 return 0;
1427
1428         free_space = free_start -
1429                 sizeof(struct ocfs2_xattr_header) -
1430                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1431                 OCFS2_XATTR_HEADER_GAP;
1432         if (free_space < 0)
1433                 return -EIO;
1434         if (free_space < needed_space)
1435                 return -ENOSPC;
1436
1437         return 0;
1438 }
1439
1440 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1441                                    int type)
1442 {
1443         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1444 }
1445
1446 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1447 {
1448         loc->xl_ops->xlo_journal_dirty(handle, loc);
1449 }
1450
1451 /* Give a pointer into the storage for the given offset */
1452 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1453 {
1454         BUG_ON(offset >= loc->xl_size);
1455         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1456 }
1457
1458 /*
1459  * Wipe the name+value pair and allow the storage to reclaim it.  This
1460  * must be followed by either removal of the entry or a call to
1461  * ocfs2_xa_add_namevalue().
1462  */
1463 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1464 {
1465         loc->xl_ops->xlo_wipe_namevalue(loc);
1466 }
1467
1468 /*
1469  * Find lowest offset to a name+value pair.  This is the start of our
1470  * downward-growing free space.
1471  */
1472 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1473 {
1474         return loc->xl_ops->xlo_get_free_start(loc);
1475 }
1476
1477 /* Can we reuse loc->xl_entry for xi? */
1478 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1479                                     struct ocfs2_xattr_info *xi)
1480 {
1481         return loc->xl_ops->xlo_can_reuse(loc, xi);
1482 }
1483
1484 /* How much free space is needed to set the new value */
1485 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1486                                 struct ocfs2_xattr_info *xi)
1487 {
1488         return loc->xl_ops->xlo_check_space(loc, xi);
1489 }
1490
1491 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1492 {
1493         loc->xl_ops->xlo_add_entry(loc, name_hash);
1494         loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1495         /*
1496          * We can't leave the new entry's xe_name_offset at zero or
1497          * add_namevalue() will go nuts.  We set it to the size of our
1498          * storage so that it can never be less than any other entry.
1499          */
1500         loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1501 }
1502
1503 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1504                                    struct ocfs2_xattr_info *xi)
1505 {
1506         int size = namevalue_size_xi(xi);
1507         int nameval_offset;
1508         char *nameval_buf;
1509
1510         loc->xl_ops->xlo_add_namevalue(loc, size);
1511         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1512         loc->xl_entry->xe_name_len = xi->xi_name_len;
1513         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1514         ocfs2_xattr_set_local(loc->xl_entry,
1515                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1516
1517         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1518         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1519         memset(nameval_buf, 0, size);
1520         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1521 }
1522
1523 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1524                                     struct ocfs2_xattr_value_buf *vb)
1525 {
1526         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1527         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1528
1529         /* Value bufs are for value trees */
1530         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1531         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1532                (name_size + OCFS2_XATTR_ROOT_SIZE));
1533
1534         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1535         vb->vb_xv =
1536                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1537                                                         nameval_offset +
1538                                                         name_size);
1539 }
1540
1541 static int ocfs2_xa_block_journal_access(handle_t *handle,
1542                                          struct ocfs2_xa_loc *loc, int type)
1543 {
1544         struct buffer_head *bh = loc->xl_storage;
1545         ocfs2_journal_access_func access;
1546
1547         if (loc->xl_size == (bh->b_size -
1548                              offsetof(struct ocfs2_xattr_block,
1549                                       xb_attrs.xb_header)))
1550                 access = ocfs2_journal_access_xb;
1551         else
1552                 access = ocfs2_journal_access_di;
1553         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1554 }
1555
1556 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1557                                          struct ocfs2_xa_loc *loc)
1558 {
1559         struct buffer_head *bh = loc->xl_storage;
1560
1561         ocfs2_journal_dirty(handle, bh);
1562 }
1563
1564 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1565                                            int offset)
1566 {
1567         return (char *)loc->xl_header + offset;
1568 }
1569
1570 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1571                                     struct ocfs2_xattr_info *xi)
1572 {
1573         /*
1574          * Block storage is strict.  If the sizes aren't exact, we will
1575          * remove the old one and reinsert the new.
1576          */
1577         return namevalue_size_xe(loc->xl_entry) ==
1578                 namevalue_size_xi(xi);
1579 }
1580
1581 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1582 {
1583         struct ocfs2_xattr_header *xh = loc->xl_header;
1584         int i, count = le16_to_cpu(xh->xh_count);
1585         int offset, free_start = loc->xl_size;
1586
1587         for (i = 0; i < count; i++) {
1588                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1589                 if (offset < free_start)
1590                         free_start = offset;
1591         }
1592
1593         return free_start;
1594 }
1595
1596 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1597                                       struct ocfs2_xattr_info *xi)
1598 {
1599         int count = le16_to_cpu(loc->xl_header->xh_count);
1600         int free_start = ocfs2_xa_get_free_start(loc);
1601         int needed_space = ocfs2_xi_entry_usage(xi);
1602
1603         /*
1604          * Block storage will reclaim the original entry before inserting
1605          * the new value, so we only need the difference.  If the new
1606          * entry is smaller than the old one, we don't need anything.
1607          */
1608         if (loc->xl_entry) {
1609                 /* Don't need space if we're reusing! */
1610                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1611                         needed_space = 0;
1612                 else
1613                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1614         }
1615         if (needed_space < 0)
1616                 needed_space = 0;
1617         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1618 }
1619
1620 /*
1621  * Block storage for xattrs keeps the name+value pairs compacted.  When
1622  * we remove one, we have to shift any that preceded it towards the end.
1623  */
1624 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1625 {
1626         int i, offset;
1627         int namevalue_offset, first_namevalue_offset, namevalue_size;
1628         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1629         struct ocfs2_xattr_header *xh = loc->xl_header;
1630         int count = le16_to_cpu(xh->xh_count);
1631
1632         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1633         namevalue_size = namevalue_size_xe(entry);
1634         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1635
1636         /* Shift the name+value pairs */
1637         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1638                 (char *)xh + first_namevalue_offset,
1639                 namevalue_offset - first_namevalue_offset);
1640         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1641
1642         /* Now tell xh->xh_entries about it */
1643         for (i = 0; i < count; i++) {
1644                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1645                 if (offset <= namevalue_offset)
1646                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1647                                      namevalue_size);
1648         }
1649
1650         /*
1651          * Note that we don't update xh_free_start or xh_name_value_len
1652          * because they're not used in block-stored xattrs.
1653          */
1654 }
1655
1656 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1657 {
1658         int count = le16_to_cpu(loc->xl_header->xh_count);
1659         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1660         le16_add_cpu(&loc->xl_header->xh_count, 1);
1661         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1662 }
1663
1664 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1665 {
1666         int free_start = ocfs2_xa_get_free_start(loc);
1667
1668         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1669 }
1670
1671 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1672                                           struct ocfs2_xattr_value_buf *vb)
1673 {
1674         struct buffer_head *bh = loc->xl_storage;
1675
1676         if (loc->xl_size == (bh->b_size -
1677                              offsetof(struct ocfs2_xattr_block,
1678                                       xb_attrs.xb_header)))
1679                 vb->vb_access = ocfs2_journal_access_xb;
1680         else
1681                 vb->vb_access = ocfs2_journal_access_di;
1682         vb->vb_bh = bh;
1683 }
1684
1685 /*
1686  * Operations for xattrs stored in blocks.  This includes inline inode
1687  * storage and unindexed ocfs2_xattr_blocks.
1688  */
1689 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1690         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1691         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1692         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1693         .xlo_check_space        = ocfs2_xa_block_check_space,
1694         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1695         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1696         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1697         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1698         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1699         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1700 };
1701
1702 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1703                                           struct ocfs2_xa_loc *loc, int type)
1704 {
1705         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1706
1707         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1708 }
1709
1710 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1711                                           struct ocfs2_xa_loc *loc)
1712 {
1713         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1714
1715         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1716 }
1717
1718 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1719                                             int offset)
1720 {
1721         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1722         int block, block_offset;
1723
1724         /* The header is at the front of the bucket */
1725         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1726         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1727
1728         return bucket_block(bucket, block) + block_offset;
1729 }
1730
1731 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1732                                      struct ocfs2_xattr_info *xi)
1733 {
1734         return namevalue_size_xe(loc->xl_entry) >=
1735                 namevalue_size_xi(xi);
1736 }
1737
1738 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1739 {
1740         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1741         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1742 }
1743
1744 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1745                                          int free_start, int size)
1746 {
1747         /*
1748          * We need to make sure that the name+value pair fits within
1749          * one block.
1750          */
1751         if (((free_start - size) >> sb->s_blocksize_bits) !=
1752             ((free_start - 1) >> sb->s_blocksize_bits))
1753                 free_start -= free_start % sb->s_blocksize;
1754
1755         return free_start;
1756 }
1757
1758 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1759                                        struct ocfs2_xattr_info *xi)
1760 {
1761         int rc;
1762         int count = le16_to_cpu(loc->xl_header->xh_count);
1763         int free_start = ocfs2_xa_get_free_start(loc);
1764         int needed_space = ocfs2_xi_entry_usage(xi);
1765         int size = namevalue_size_xi(xi);
1766         struct super_block *sb = loc->xl_inode->i_sb;
1767
1768         /*
1769          * Bucket storage does not reclaim name+value pairs it cannot
1770          * reuse.  They live as holes until the bucket fills, and then
1771          * the bucket is defragmented.  However, the bucket can reclaim
1772          * the ocfs2_xattr_entry.
1773          */
1774         if (loc->xl_entry) {
1775                 /* Don't need space if we're reusing! */
1776                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1777                         needed_space = 0;
1778                 else
1779                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1780         }
1781         BUG_ON(needed_space < 0);
1782
1783         if (free_start < size) {
1784                 if (needed_space)
1785                         return -ENOSPC;
1786         } else {
1787                 /*
1788                  * First we check if it would fit in the first place.
1789                  * Below, we align the free start to a block.  This may
1790                  * slide us below the minimum gap.  By checking unaligned
1791                  * first, we avoid that error.
1792                  */
1793                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1794                                                  count);
1795                 if (rc)
1796                         return rc;
1797                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1798                                                            size);
1799         }
1800         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1801 }
1802
1803 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1804 {
1805         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1806                      -namevalue_size_xe(loc->xl_entry));
1807 }
1808
1809 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1810 {
1811         struct ocfs2_xattr_header *xh = loc->xl_header;
1812         int count = le16_to_cpu(xh->xh_count);
1813         int low = 0, high = count - 1, tmp;
1814         struct ocfs2_xattr_entry *tmp_xe;
1815
1816         /*
1817          * We keep buckets sorted by name_hash, so we need to find
1818          * our insert place.
1819          */
1820         while (low <= high && count) {
1821                 tmp = (low + high) / 2;
1822                 tmp_xe = &xh->xh_entries[tmp];
1823
1824                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1825                         low = tmp + 1;
1826                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1827                         high = tmp - 1;
1828                 else {
1829                         low = tmp;
1830                         break;
1831                 }
1832         }
1833
1834         if (low != count)
1835                 memmove(&xh->xh_entries[low + 1],
1836                         &xh->xh_entries[low],
1837                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1838
1839         le16_add_cpu(&xh->xh_count, 1);
1840         loc->xl_entry = &xh->xh_entries[low];
1841         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1842 }
1843
1844 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1845 {
1846         int free_start = ocfs2_xa_get_free_start(loc);
1847         struct ocfs2_xattr_header *xh = loc->xl_header;
1848         struct super_block *sb = loc->xl_inode->i_sb;
1849         int nameval_offset;
1850
1851         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1852         nameval_offset = free_start - size;
1853         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1854         xh->xh_free_start = cpu_to_le16(nameval_offset);
1855         le16_add_cpu(&xh->xh_name_value_len, size);
1856
1857 }
1858
1859 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1860                                            struct ocfs2_xattr_value_buf *vb)
1861 {
1862         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1863         struct super_block *sb = loc->xl_inode->i_sb;
1864         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1865         int size = namevalue_size_xe(loc->xl_entry);
1866         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1867
1868         /* Values are not allowed to straddle block boundaries */
1869         BUG_ON(block_offset !=
1870                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1871         /* We expect the bucket to be filled in */
1872         BUG_ON(!bucket->bu_bhs[block_offset]);
1873
1874         vb->vb_access = ocfs2_journal_access;
1875         vb->vb_bh = bucket->bu_bhs[block_offset];
1876 }
1877
1878 /* Operations for xattrs stored in buckets. */
1879 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1880         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1881         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1882         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1883         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1884         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1885         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1886         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1887         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1888         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1889         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1890 };
1891
1892 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1893 {
1894         struct ocfs2_xattr_value_buf vb;
1895
1896         if (ocfs2_xattr_is_local(loc->xl_entry))
1897                 return 0;
1898
1899         ocfs2_xa_fill_value_buf(loc, &vb);
1900         return le32_to_cpu(vb.vb_xv->xr_clusters);
1901 }
1902
1903 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1904                                    struct ocfs2_xattr_set_ctxt *ctxt)
1905 {
1906         int trunc_rc, access_rc;
1907         struct ocfs2_xattr_value_buf vb;
1908
1909         ocfs2_xa_fill_value_buf(loc, &vb);
1910         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1911                                               ctxt);
1912
1913         /*
1914          * The caller of ocfs2_xa_value_truncate() has already called
1915          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1916          * calls ocfs2_extend_trans().  This may commit the previous
1917          * transaction and open a new one.  If this is a bucket, truncate
1918          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1919          * the caller is expecting to dirty the entire bucket.  So we must
1920          * reset the journal work.  We do this even if truncate has failed,
1921          * as it could have failed after committing the extend.
1922          */
1923         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1924                                             OCFS2_JOURNAL_ACCESS_WRITE);
1925
1926         /* Errors in truncate take precedence */
1927         return trunc_rc ? trunc_rc : access_rc;
1928 }
1929
1930 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1931 {
1932         int index, count;
1933         struct ocfs2_xattr_header *xh = loc->xl_header;
1934         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1935
1936         ocfs2_xa_wipe_namevalue(loc);
1937         loc->xl_entry = NULL;
1938
1939         le16_add_cpu(&xh->xh_count, -1);
1940         count = le16_to_cpu(xh->xh_count);
1941
1942         /*
1943          * Only zero out the entry if there are more remaining.  This is
1944          * important for an empty bucket, as it keeps track of the
1945          * bucket's hash value.  It doesn't hurt empty block storage.
1946          */
1947         if (count) {
1948                 index = ((char *)entry - (char *)&xh->xh_entries) /
1949                         sizeof(struct ocfs2_xattr_entry);
1950                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1951                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1952                 memset(&xh->xh_entries[count], 0,
1953                        sizeof(struct ocfs2_xattr_entry));
1954         }
1955 }
1956
1957 /*
1958  * If we have a problem adjusting the size of an external value during
1959  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1960  * in an intermediate state.  For example, the value may be partially
1961  * truncated.
1962  *
1963  * If the value tree hasn't changed, the extend/truncate went nowhere.
1964  * We have nothing to do.  The caller can treat it as a straight error.
1965  *
1966  * If the value tree got partially truncated, we now have a corrupted
1967  * extended attribute.  We're going to wipe its entry and leak the
1968  * clusters.  Better to leak some storage than leave a corrupt entry.
1969  *
1970  * If the value tree grew, it obviously didn't grow enough for the
1971  * new entry.  We're not going to try and reclaim those clusters either.
1972  * If there was already an external value there (orig_clusters != 0),
1973  * the new clusters are attached safely and we can just leave the old
1974  * value in place.  If there was no external value there, we remove
1975  * the entry.
1976  *
1977  * This way, the xattr block we store in the journal will be consistent.
1978  * If the size change broke because of the journal, no changes will hit
1979  * disk anyway.
1980  */
1981 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1982                                             const char *what,
1983                                             unsigned int orig_clusters)
1984 {
1985         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1986         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1987                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1988
1989         if (new_clusters < orig_clusters) {
1990                 mlog(ML_ERROR,
1991                      "Partial truncate while %s xattr %.*s.  Leaking "
1992                      "%u clusters and removing the entry\n",
1993                      what, loc->xl_entry->xe_name_len, nameval_buf,
1994                      orig_clusters - new_clusters);
1995                 ocfs2_xa_remove_entry(loc);
1996         } else if (!orig_clusters) {
1997                 mlog(ML_ERROR,
1998                      "Unable to allocate an external value for xattr "
1999                      "%.*s safely.  Leaking %u clusters and removing the "
2000                      "entry\n",
2001                      loc->xl_entry->xe_name_len, nameval_buf,
2002                      new_clusters - orig_clusters);
2003                 ocfs2_xa_remove_entry(loc);
2004         } else if (new_clusters > orig_clusters)
2005                 mlog(ML_ERROR,
2006                      "Unable to grow xattr %.*s safely.  %u new clusters "
2007                      "have been added, but the value will not be "
2008                      "modified\n",
2009                      loc->xl_entry->xe_name_len, nameval_buf,
2010                      new_clusters - orig_clusters);
2011 }
2012
2013 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2014                            struct ocfs2_xattr_set_ctxt *ctxt)
2015 {
2016         int rc = 0;
2017         unsigned int orig_clusters;
2018
2019         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2020                 orig_clusters = ocfs2_xa_value_clusters(loc);
2021                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2022                 if (rc) {
2023                         mlog_errno(rc);
2024                         /*
2025                          * Since this is remove, we can return 0 if
2026                          * ocfs2_xa_cleanup_value_truncate() is going to
2027                          * wipe the entry anyway.  So we check the
2028                          * cluster count as well.
2029                          */
2030                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2031                                 rc = 0;
2032                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2033                                                         orig_clusters);
2034                         if (rc)
2035                                 goto out;
2036                 }
2037         }
2038
2039         ocfs2_xa_remove_entry(loc);
2040
2041 out:
2042         return rc;
2043 }
2044
2045 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2046 {
2047         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2048         char *nameval_buf;
2049
2050         nameval_buf = ocfs2_xa_offset_pointer(loc,
2051                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2052         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2053 }
2054
2055 /*
2056  * Take an existing entry and make it ready for the new value.  This
2057  * won't allocate space, but it may free space.  It should be ready for
2058  * ocfs2_xa_prepare_entry() to finish the work.
2059  */
2060 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2061                                 struct ocfs2_xattr_info *xi,
2062                                 struct ocfs2_xattr_set_ctxt *ctxt)
2063 {
2064         int rc = 0;
2065         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2066         unsigned int orig_clusters;
2067         char *nameval_buf;
2068         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2069         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2070
2071         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2072                name_size);
2073
2074         nameval_buf = ocfs2_xa_offset_pointer(loc,
2075                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2076         if (xe_local) {
2077                 memset(nameval_buf + name_size, 0,
2078                        namevalue_size_xe(loc->xl_entry) - name_size);
2079                 if (!xi_local)
2080                         ocfs2_xa_install_value_root(loc);
2081         } else {
2082                 orig_clusters = ocfs2_xa_value_clusters(loc);
2083                 if (xi_local) {
2084                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2085                         if (rc < 0)
2086                                 mlog_errno(rc);
2087                         else
2088                                 memset(nameval_buf + name_size, 0,
2089                                        namevalue_size_xe(loc->xl_entry) -
2090                                        name_size);
2091                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2092                            xi->xi_value_len) {
2093                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2094                                                      ctxt);
2095                         if (rc < 0)
2096                                 mlog_errno(rc);
2097                 }
2098
2099                 if (rc) {
2100                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2101                                                         orig_clusters);
2102                         goto out;
2103                 }
2104         }
2105
2106         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2107         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2108
2109 out:
2110         return rc;
2111 }
2112
2113 /*
2114  * Prepares loc->xl_entry to receive the new xattr.  This includes
2115  * properly setting up the name+value pair region.  If loc->xl_entry
2116  * already exists, it will take care of modifying it appropriately.
2117  *
2118  * Note that this modifies the data.  You did journal_access already,
2119  * right?
2120  */
2121 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2122                                   struct ocfs2_xattr_info *xi,
2123                                   u32 name_hash,
2124                                   struct ocfs2_xattr_set_ctxt *ctxt)
2125 {
2126         int rc = 0;
2127         unsigned int orig_clusters;
2128         __le64 orig_value_size = 0;
2129
2130         rc = ocfs2_xa_check_space(loc, xi);
2131         if (rc)
2132                 goto out;
2133
2134         if (loc->xl_entry) {
2135                 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2136                         orig_value_size = loc->xl_entry->xe_value_size;
2137                         rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2138                         if (rc)
2139                                 goto out;
2140                         goto alloc_value;
2141                 }
2142
2143                 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2144                         orig_clusters = ocfs2_xa_value_clusters(loc);
2145                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2146                         if (rc) {
2147                                 mlog_errno(rc);
2148                                 ocfs2_xa_cleanup_value_truncate(loc,
2149                                                                 "overwriting",
2150                                                                 orig_clusters);
2151                                 goto out;
2152                         }
2153                 }
2154                 ocfs2_xa_wipe_namevalue(loc);
2155         } else
2156                 ocfs2_xa_add_entry(loc, name_hash);
2157
2158         /*
2159          * If we get here, we have a blank entry.  Fill it.  We grow our
2160          * name+value pair back from the end.
2161          */
2162         ocfs2_xa_add_namevalue(loc, xi);
2163         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2164                 ocfs2_xa_install_value_root(loc);
2165
2166 alloc_value:
2167         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2168                 orig_clusters = ocfs2_xa_value_clusters(loc);
2169                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2170                 if (rc < 0) {
2171                         ctxt->set_abort = 1;
2172                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2173                                                         orig_clusters);
2174                         /*
2175                          * If we were growing an existing value,
2176                          * ocfs2_xa_cleanup_value_truncate() won't remove
2177                          * the entry. We need to restore the original value
2178                          * size.
2179                          */
2180                         if (loc->xl_entry) {
2181                                 BUG_ON(!orig_value_size);
2182                                 loc->xl_entry->xe_value_size = orig_value_size;
2183                         }
2184                         mlog_errno(rc);
2185                 }
2186         }
2187
2188 out:
2189         return rc;
2190 }
2191
2192 /*
2193  * Store the value portion of the name+value pair.  This will skip
2194  * values that are stored externally.  Their tree roots were set up
2195  * by ocfs2_xa_prepare_entry().
2196  */
2197 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2198                                 struct ocfs2_xattr_info *xi,
2199                                 struct ocfs2_xattr_set_ctxt *ctxt)
2200 {
2201         int rc = 0;
2202         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2203         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2204         char *nameval_buf;
2205         struct ocfs2_xattr_value_buf vb;
2206
2207         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2208         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2209                 ocfs2_xa_fill_value_buf(loc, &vb);
2210                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2211                                                      ctxt->handle, &vb,
2212                                                      xi->xi_value,
2213                                                      xi->xi_value_len);
2214         } else
2215                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2216
2217         return rc;
2218 }
2219
2220 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2221                         struct ocfs2_xattr_info *xi,
2222                         struct ocfs2_xattr_set_ctxt *ctxt)
2223 {
2224         int ret;
2225         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2226                                               xi->xi_name_len);
2227
2228         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2229                                       OCFS2_JOURNAL_ACCESS_WRITE);
2230         if (ret) {
2231                 mlog_errno(ret);
2232                 goto out;
2233         }
2234
2235         /*
2236          * From here on out, everything is going to modify the buffer a
2237          * little.  Errors are going to leave the xattr header in a
2238          * sane state.  Thus, even with errors we dirty the sucker.
2239          */
2240
2241         /* Don't worry, we are never called with !xi_value and !xl_entry */
2242         if (!xi->xi_value) {
2243                 ret = ocfs2_xa_remove(loc, ctxt);
2244                 goto out_dirty;
2245         }
2246
2247         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2248         if (ret) {
2249                 if (ret != -ENOSPC)
2250                         mlog_errno(ret);
2251                 goto out_dirty;
2252         }
2253
2254         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2255         if (ret)
2256                 mlog_errno(ret);
2257
2258 out_dirty:
2259         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2260
2261 out:
2262         return ret;
2263 }
2264
2265 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2266                                      struct inode *inode,
2267                                      struct buffer_head *bh,
2268                                      struct ocfs2_xattr_entry *entry)
2269 {
2270         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2271
2272         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2273
2274         loc->xl_inode = inode;
2275         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2276         loc->xl_storage = bh;
2277         loc->xl_entry = entry;
2278         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2279         loc->xl_header =
2280                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2281                                               loc->xl_size);
2282 }
2283
2284 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2285                                           struct inode *inode,
2286                                           struct buffer_head *bh,
2287                                           struct ocfs2_xattr_entry *entry)
2288 {
2289         struct ocfs2_xattr_block *xb =
2290                 (struct ocfs2_xattr_block *)bh->b_data;
2291
2292         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2293
2294         loc->xl_inode = inode;
2295         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2296         loc->xl_storage = bh;
2297         loc->xl_header = &(xb->xb_attrs.xb_header);
2298         loc->xl_entry = entry;
2299         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2300                                              xb_attrs.xb_header);
2301 }
2302
2303 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2304                                            struct ocfs2_xattr_bucket *bucket,
2305                                            struct ocfs2_xattr_entry *entry)
2306 {
2307         loc->xl_inode = bucket->bu_inode;
2308         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2309         loc->xl_storage = bucket;
2310         loc->xl_header = bucket_xh(bucket);
2311         loc->xl_entry = entry;
2312         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2313 }
2314
2315 /*
2316  * In xattr remove, if it is stored outside and refcounted, we may have
2317  * the chance to split the refcount tree. So need the allocators.
2318  */
2319 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2320                                         struct ocfs2_xattr_value_root *xv,
2321                                         struct ocfs2_caching_info *ref_ci,
2322                                         struct buffer_head *ref_root_bh,
2323                                         struct ocfs2_alloc_context **meta_ac,
2324                                         int *ref_credits)
2325 {
2326         int ret, meta_add = 0;
2327         u32 p_cluster, num_clusters;
2328         unsigned int ext_flags;
2329
2330         *ref_credits = 0;
2331         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2332                                        &num_clusters,
2333                                        &xv->xr_list,
2334                                        &ext_flags);
2335         if (ret) {
2336                 mlog_errno(ret);
2337                 goto out;
2338         }
2339
2340         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2341                 goto out;
2342
2343         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2344                                                  ref_root_bh, xv,
2345                                                  &meta_add, ref_credits);
2346         if (ret) {
2347                 mlog_errno(ret);
2348                 goto out;
2349         }
2350
2351         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2352                                                 meta_add, meta_ac);
2353         if (ret)
2354                 mlog_errno(ret);
2355
2356 out:
2357         return ret;
2358 }
2359
2360 static int ocfs2_remove_value_outside(struct inode*inode,
2361                                       struct ocfs2_xattr_value_buf *vb,
2362                                       struct ocfs2_xattr_header *header,
2363                                       struct ocfs2_caching_info *ref_ci,
2364                                       struct buffer_head *ref_root_bh)
2365 {
2366         int ret = 0, i, ref_credits;
2367         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2368         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2369         void *val;
2370
2371         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2372
2373         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2374                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2375
2376                 if (ocfs2_xattr_is_local(entry))
2377                         continue;
2378
2379                 val = (void *)header +
2380                         le16_to_cpu(entry->xe_name_offset);
2381                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2382                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2383
2384                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2385                                                          ref_ci, ref_root_bh,
2386                                                          &ctxt.meta_ac,
2387                                                          &ref_credits);
2388
2389                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2390                                         ocfs2_remove_extent_credits(osb->sb));
2391                 if (IS_ERR(ctxt.handle)) {
2392                         ret = PTR_ERR(ctxt.handle);
2393                         mlog_errno(ret);
2394                         break;
2395                 }
2396
2397                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2398
2399                 ocfs2_commit_trans(osb, ctxt.handle);
2400                 if (ctxt.meta_ac) {
2401                         ocfs2_free_alloc_context(ctxt.meta_ac);
2402                         ctxt.meta_ac = NULL;
2403                 }
2404
2405                 if (ret < 0) {
2406                         mlog_errno(ret);
2407                         break;
2408                 }
2409
2410         }
2411
2412         if (ctxt.meta_ac)
2413                 ocfs2_free_alloc_context(ctxt.meta_ac);
2414         ocfs2_schedule_truncate_log_flush(osb, 1);
2415         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2416         return ret;
2417 }
2418
2419 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2420                                     struct buffer_head *di_bh,
2421                                     struct ocfs2_caching_info *ref_ci,
2422                                     struct buffer_head *ref_root_bh)
2423 {
2424
2425         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2426         struct ocfs2_xattr_header *header;
2427         int ret;
2428         struct ocfs2_xattr_value_buf vb = {
2429                 .vb_bh = di_bh,
2430                 .vb_access = ocfs2_journal_access_di,
2431         };
2432
2433         header = (struct ocfs2_xattr_header *)
2434                  ((void *)di + inode->i_sb->s_blocksize -
2435                  le16_to_cpu(di->i_xattr_inline_size));
2436
2437         ret = ocfs2_remove_value_outside(inode, &vb, header,
2438                                          ref_ci, ref_root_bh);
2439
2440         return ret;
2441 }
2442
2443 struct ocfs2_rm_xattr_bucket_para {
2444         struct ocfs2_caching_info *ref_ci;
2445         struct buffer_head *ref_root_bh;
2446 };
2447
2448 static int ocfs2_xattr_block_remove(struct inode *inode,
2449                                     struct buffer_head *blk_bh,
2450                                     struct ocfs2_caching_info *ref_ci,
2451                                     struct buffer_head *ref_root_bh)
2452 {
2453         struct ocfs2_xattr_block *xb;
2454         int ret = 0;
2455         struct ocfs2_xattr_value_buf vb = {
2456                 .vb_bh = blk_bh,
2457                 .vb_access = ocfs2_journal_access_xb,
2458         };
2459         struct ocfs2_rm_xattr_bucket_para args = {
2460                 .ref_ci = ref_ci,
2461                 .ref_root_bh = ref_root_bh,
2462         };
2463
2464         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2465         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2466                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2467                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2468                                                  ref_ci, ref_root_bh);
2469         } else
2470                 ret = ocfs2_iterate_xattr_index_block(inode,
2471                                                 blk_bh,
2472                                                 ocfs2_rm_xattr_cluster,
2473                                                 &args);
2474
2475         return ret;
2476 }
2477
2478 static int ocfs2_xattr_free_block(struct inode *inode,
2479                                   u64 block,
2480                                   struct ocfs2_caching_info *ref_ci,
2481                                   struct buffer_head *ref_root_bh)
2482 {
2483         struct inode *xb_alloc_inode;
2484         struct buffer_head *xb_alloc_bh = NULL;
2485         struct buffer_head *blk_bh = NULL;
2486         struct ocfs2_xattr_block *xb;
2487         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2488         handle_t *handle;
2489         int ret = 0;
2490         u64 blk, bg_blkno;
2491         u16 bit;
2492
2493         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2494         if (ret < 0) {
2495                 mlog_errno(ret);
2496                 goto out;
2497         }
2498
2499         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2500         if (ret < 0) {
2501                 mlog_errno(ret);
2502                 goto out;
2503         }
2504
2505         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2506         blk = le64_to_cpu(xb->xb_blkno);
2507         bit = le16_to_cpu(xb->xb_suballoc_bit);
2508         if (xb->xb_suballoc_loc)
2509                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2510         else
2511                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2512
2513         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2514                                 EXTENT_ALLOC_SYSTEM_INODE,
2515                                 le16_to_cpu(xb->xb_suballoc_slot));
2516         if (!xb_alloc_inode) {
2517                 ret = -ENOMEM;
2518                 mlog_errno(ret);
2519                 goto out;
2520         }
2521         inode_lock(xb_alloc_inode);
2522
2523         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2524         if (ret < 0) {
2525                 mlog_errno(ret);
2526                 goto out_mutex;
2527         }
2528
2529         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2530         if (IS_ERR(handle)) {
2531                 ret = PTR_ERR(handle);
2532                 mlog_errno(ret);
2533                 goto out_unlock;
2534         }
2535
2536         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2537                                        bit, bg_blkno, 1);
2538         if (ret < 0)
2539                 mlog_errno(ret);
2540
2541         ocfs2_commit_trans(osb, handle);
2542 out_unlock:
2543         ocfs2_inode_unlock(xb_alloc_inode, 1);
2544         brelse(xb_alloc_bh);
2545 out_mutex:
2546         inode_unlock(xb_alloc_inode);
2547         iput(xb_alloc_inode);
2548 out:
2549         brelse(blk_bh);
2550         return ret;
2551 }
2552
2553 /*
2554  * ocfs2_xattr_remove()
2555  *
2556  * Free extended attribute resources associated with this inode.
2557  */
2558 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2559 {
2560         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2561         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2562         struct ocfs2_refcount_tree *ref_tree = NULL;
2563         struct buffer_head *ref_root_bh = NULL;
2564         struct ocfs2_caching_info *ref_ci = NULL;
2565         handle_t *handle;
2566         int ret;
2567
2568         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2569                 return 0;
2570
2571         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2572                 return 0;
2573
2574         if (ocfs2_is_refcount_inode(inode)) {
2575                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2576                                                le64_to_cpu(di->i_refcount_loc),
2577                                                1, &ref_tree, &ref_root_bh);
2578                 if (ret) {
2579                         mlog_errno(ret);
2580                         goto out;
2581                 }
2582                 ref_ci = &ref_tree->rf_ci;
2583
2584         }
2585
2586         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2587                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2588                                                ref_ci, ref_root_bh);
2589                 if (ret < 0) {
2590                         mlog_errno(ret);
2591                         goto out;
2592                 }
2593         }
2594
2595         if (di->i_xattr_loc) {
2596                 ret = ocfs2_xattr_free_block(inode,
2597                                              le64_to_cpu(di->i_xattr_loc),
2598                                              ref_ci, ref_root_bh);
2599                 if (ret < 0) {
2600                         mlog_errno(ret);
2601                         goto out;
2602                 }
2603         }
2604
2605         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2606                                    OCFS2_INODE_UPDATE_CREDITS);
2607         if (IS_ERR(handle)) {
2608                 ret = PTR_ERR(handle);
2609                 mlog_errno(ret);
2610                 goto out;
2611         }
2612         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2613                                       OCFS2_JOURNAL_ACCESS_WRITE);
2614         if (ret) {
2615                 mlog_errno(ret);
2616                 goto out_commit;
2617         }
2618
2619         di->i_xattr_loc = 0;
2620
2621         spin_lock(&oi->ip_lock);
2622         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2623         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2624         spin_unlock(&oi->ip_lock);
2625         ocfs2_update_inode_fsync_trans(handle, inode, 0);
2626
2627         ocfs2_journal_dirty(handle, di_bh);
2628 out_commit:
2629         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2630 out:
2631         if (ref_tree)
2632                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2633         brelse(ref_root_bh);
2634         return ret;
2635 }
2636
2637 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2638                                         struct ocfs2_dinode *di)
2639 {
2640         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2641         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2642         int free;
2643
2644         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2645                 return 0;
2646
2647         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2648                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2649                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2650         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2651                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2652                         le64_to_cpu(di->i_size);
2653         } else {
2654                 struct ocfs2_extent_list *el = &di->id2.i_list;
2655                 free = (le16_to_cpu(el->l_count) -
2656                         le16_to_cpu(el->l_next_free_rec)) *
2657                         sizeof(struct ocfs2_extent_rec);
2658         }
2659         if (free >= xattrsize)
2660                 return 1;
2661
2662         return 0;
2663 }
2664
2665 /*
2666  * ocfs2_xattr_ibody_find()
2667  *
2668  * Find extended attribute in inode block and
2669  * fill search info into struct ocfs2_xattr_search.
2670  */
2671 static int ocfs2_xattr_ibody_find(struct inode *inode,
2672                                   int name_index,
2673                                   const char *name,
2674                                   struct ocfs2_xattr_search *xs)
2675 {
2676         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2677         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2678         int ret;
2679         int has_space = 0;
2680
2681         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2682                 return 0;
2683
2684         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2685                 down_read(&oi->ip_alloc_sem);
2686                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2687                 up_read(&oi->ip_alloc_sem);
2688                 if (!has_space)
2689                         return 0;
2690         }
2691
2692         xs->xattr_bh = xs->inode_bh;
2693         xs->end = (void *)di + inode->i_sb->s_blocksize;
2694         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2695                 xs->header = (struct ocfs2_xattr_header *)
2696                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2697         else
2698                 xs->header = (struct ocfs2_xattr_header *)
2699                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2700         xs->base = (void *)xs->header;
2701         xs->here = xs->header->xh_entries;
2702
2703         /* Find the named attribute. */
2704         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2705                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2706                 if (ret && ret != -ENODATA)
2707                         return ret;
2708                 xs->not_found = ret;
2709         }
2710
2711         return 0;
2712 }
2713
2714 static int ocfs2_xattr_ibody_init(struct inode *inode,
2715                                   struct buffer_head *di_bh,
2716                                   struct ocfs2_xattr_set_ctxt *ctxt)
2717 {
2718         int ret;
2719         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2720         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2721         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2722         unsigned int xattrsize = osb->s_xattr_inline_size;
2723
2724         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2725                 ret = -ENOSPC;
2726                 goto out;
2727         }
2728
2729         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2730                                       OCFS2_JOURNAL_ACCESS_WRITE);
2731         if (ret) {
2732                 mlog_errno(ret);
2733                 goto out;
2734         }
2735
2736         /*
2737          * Adjust extent record count or inline data size
2738          * to reserve space for extended attribute.
2739          */
2740         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2741                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2742                 le16_add_cpu(&idata->id_count, -xattrsize);
2743         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2744                 struct ocfs2_extent_list *el = &di->id2.i_list;
2745                 le16_add_cpu(&el->l_count, -(xattrsize /
2746                                              sizeof(struct ocfs2_extent_rec)));
2747         }
2748         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2749
2750         spin_lock(&oi->ip_lock);
2751         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2752         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2753         spin_unlock(&oi->ip_lock);
2754
2755         ocfs2_journal_dirty(ctxt->handle, di_bh);
2756
2757 out:
2758         return ret;
2759 }
2760
2761 /*
2762  * ocfs2_xattr_ibody_set()
2763  *
2764  * Set, replace or remove an extended attribute into inode block.
2765  *
2766  */
2767 static int ocfs2_xattr_ibody_set(struct inode *inode,
2768                                  struct ocfs2_xattr_info *xi,
2769                                  struct ocfs2_xattr_search *xs,
2770                                  struct ocfs2_xattr_set_ctxt *ctxt)
2771 {
2772         int ret;
2773         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2774         struct ocfs2_xa_loc loc;
2775
2776         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2777                 return -ENOSPC;
2778
2779         down_write(&oi->ip_alloc_sem);
2780         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2781                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2782                 if (ret) {
2783                         if (ret != -ENOSPC)
2784                                 mlog_errno(ret);
2785                         goto out;
2786                 }
2787         }
2788
2789         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2790                                  xs->not_found ? NULL : xs->here);
2791         ret = ocfs2_xa_set(&loc, xi, ctxt);
2792         if (ret) {
2793                 if (ret != -ENOSPC)
2794                         mlog_errno(ret);
2795                 goto out;
2796         }
2797         xs->here = loc.xl_entry;
2798
2799 out:
2800         up_write(&oi->ip_alloc_sem);
2801
2802         return ret;
2803 }
2804
2805 /*
2806  * ocfs2_xattr_block_find()
2807  *
2808  * Find extended attribute in external block and
2809  * fill search info into struct ocfs2_xattr_search.
2810  */
2811 static int ocfs2_xattr_block_find(struct inode *inode,
2812                                   int name_index,
2813                                   const char *name,
2814                                   struct ocfs2_xattr_search *xs)
2815 {
2816         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2817         struct buffer_head *blk_bh = NULL;
2818         struct ocfs2_xattr_block *xb;
2819         int ret = 0;
2820
2821         if (!di->i_xattr_loc)
2822                 return ret;
2823
2824         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2825                                      &blk_bh);
2826         if (ret < 0) {
2827                 mlog_errno(ret);
2828                 return ret;
2829         }
2830
2831         xs->xattr_bh = blk_bh;
2832         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2833
2834         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2835                 xs->header = &xb->xb_attrs.xb_header;
2836                 xs->base = (void *)xs->header;
2837                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2838                 xs->here = xs->header->xh_entries;
2839
2840                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2841         } else
2842                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2843                                                    name_index,
2844                                                    name, xs);
2845
2846         if (ret && ret != -ENODATA) {
2847                 xs->xattr_bh = NULL;
2848                 goto cleanup;
2849         }
2850         xs->not_found = ret;
2851         return 0;
2852 cleanup:
2853         brelse(blk_bh);
2854
2855         return ret;
2856 }
2857
2858 static int ocfs2_create_xattr_block(struct inode *inode,
2859                                     struct buffer_head *inode_bh,
2860                                     struct ocfs2_xattr_set_ctxt *ctxt,
2861                                     int indexed,
2862                                     struct buffer_head **ret_bh)
2863 {
2864         int ret;
2865         u16 suballoc_bit_start;
2866         u32 num_got;
2867         u64 suballoc_loc, first_blkno;
2868         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2869         struct buffer_head *new_bh = NULL;
2870         struct ocfs2_xattr_block *xblk;
2871
2872         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2873                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2874         if (ret < 0) {
2875                 mlog_errno(ret);
2876                 goto end;
2877         }
2878
2879         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2880                                    &suballoc_loc, &suballoc_bit_start,
2881                                    &num_got, &first_blkno);
2882         if (ret < 0) {
2883                 mlog_errno(ret);
2884                 goto end;
2885         }
2886
2887         new_bh = sb_getblk(inode->i_sb, first_blkno);
2888         if (!new_bh) {
2889                 ret = -ENOMEM;
2890                 mlog_errno(ret);
2891                 goto end;
2892         }
2893
2894         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2895
2896         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2897                                       new_bh,
2898                                       OCFS2_JOURNAL_ACCESS_CREATE);
2899         if (ret < 0) {
2900                 mlog_errno(ret);
2901                 goto end;
2902         }
2903
2904         /* Initialize ocfs2_xattr_block */
2905         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2906         memset(xblk, 0, inode->i_sb->s_blocksize);
2907         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2908         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2909         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2910         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2911         xblk->xb_fs_generation =
2912                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2913         xblk->xb_blkno = cpu_to_le64(first_blkno);
2914         if (indexed) {
2915                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2916                 xr->xt_clusters = cpu_to_le32(1);
2917                 xr->xt_last_eb_blk = 0;
2918                 xr->xt_list.l_tree_depth = 0;
2919                 xr->xt_list.l_count = cpu_to_le16(
2920                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2921                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2922                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2923         }
2924         ocfs2_journal_dirty(ctxt->handle, new_bh);
2925
2926         /* Add it to the inode */
2927         di->i_xattr_loc = cpu_to_le64(first_blkno);
2928
2929         spin_lock(&OCFS2_I(inode)->ip_lock);
2930         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2931         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2932         spin_unlock(&OCFS2_I(inode)->ip_lock);
2933
2934         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2935
2936         *ret_bh = new_bh;
2937         new_bh = NULL;
2938
2939 end:
2940         brelse(new_bh);
2941         return ret;
2942 }
2943
2944 /*
2945  * ocfs2_xattr_block_set()
2946  *
2947  * Set, replace or remove an extended attribute into external block.
2948  *
2949  */
2950 static int ocfs2_xattr_block_set(struct inode *inode,
2951                                  struct ocfs2_xattr_info *xi,
2952                                  struct ocfs2_xattr_search *xs,
2953                                  struct ocfs2_xattr_set_ctxt *ctxt)
2954 {
2955         struct buffer_head *new_bh = NULL;
2956         struct ocfs2_xattr_block *xblk = NULL;
2957         int ret;
2958         struct ocfs2_xa_loc loc;
2959
2960         if (!xs->xattr_bh) {
2961                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2962                                                0, &new_bh);
2963                 if (ret) {
2964                         mlog_errno(ret);
2965                         goto end;
2966                 }
2967
2968                 xs->xattr_bh = new_bh;
2969                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2970                 xs->header = &xblk->xb_attrs.xb_header;
2971                 xs->base = (void *)xs->header;
2972                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2973                 xs->here = xs->header->xh_entries;
2974         } else
2975                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2976
2977         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2978                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2979                                               xs->not_found ? NULL : xs->here);
2980
2981                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2982                 if (!ret)
2983                         xs->here = loc.xl_entry;
2984                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2985                         goto end;
2986                 else {
2987                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2988                         if (ret)
2989                                 goto end;
2990                 }
2991         }
2992
2993         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2994                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2995
2996 end:
2997         return ret;
2998 }
2999
3000 /* Check whether the new xattr can be inserted into the inode. */
3001 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
3002                                        struct ocfs2_xattr_info *xi,
3003                                        struct ocfs2_xattr_search *xs)
3004 {
3005         struct ocfs2_xattr_entry *last;
3006         int free, i;
3007         size_t min_offs = xs->end - xs->base;
3008
3009         if (!xs->header)
3010                 return 0;
3011
3012         last = xs->header->xh_entries;
3013
3014         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3015                 size_t offs = le16_to_cpu(last->xe_name_offset);
3016                 if (offs < min_offs)
3017                         min_offs = offs;
3018                 last += 1;
3019         }
3020
3021         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3022         if (free < 0)
3023                 return 0;
3024
3025         BUG_ON(!xs->not_found);
3026
3027         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3028                 return 1;
3029
3030         return 0;
3031 }
3032
3033 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3034                                      struct ocfs2_dinode *di,
3035                                      struct ocfs2_xattr_info *xi,
3036                                      struct ocfs2_xattr_search *xis,
3037                                      struct ocfs2_xattr_search *xbs,
3038                                      int *clusters_need,
3039                                      int *meta_need,
3040                                      int *credits_need)
3041 {
3042         int ret = 0, old_in_xb = 0;
3043         int clusters_add = 0, meta_add = 0, credits = 0;
3044         struct buffer_head *bh = NULL;
3045         struct ocfs2_xattr_block *xb = NULL;
3046         struct ocfs2_xattr_entry *xe = NULL;
3047         struct ocfs2_xattr_value_root *xv = NULL;
3048         char *base = NULL;
3049         int name_offset, name_len = 0;
3050         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3051                                                     xi->xi_value_len);
3052         u64 value_size;
3053
3054         /*
3055          * Calculate the clusters we need to write.
3056          * No matter whether we replace an old one or add a new one,
3057          * we need this for writing.
3058          */
3059         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3060                 credits += new_clusters *
3061                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3062
3063         if (xis->not_found && xbs->not_found) {
3064                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3065
3066                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3067                         clusters_add += new_clusters;
3068                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3069                                                         &def_xv.xv.xr_list);
3070                 }
3071
3072                 goto meta_guess;
3073         }
3074
3075         if (!xis->not_found) {
3076                 xe = xis->here;
3077                 name_offset = le16_to_cpu(xe->xe_name_offset);
3078                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3079                 base = xis->base;
3080                 credits += OCFS2_INODE_UPDATE_CREDITS;
3081         } else {
3082                 int i, block_off = 0;
3083                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3084                 xe = xbs->here;
3085                 name_offset = le16_to_cpu(xe->xe_name_offset);
3086                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3087                 i = xbs->here - xbs->header->xh_entries;
3088                 old_in_xb = 1;
3089
3090                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3091                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3092                                                         bucket_xh(xbs->bucket),
3093                                                         i, &block_off,
3094                                                         &name_offset);
3095                         base = bucket_block(xbs->bucket, block_off);
3096                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3097                 } else {
3098                         base = xbs->base;
3099                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3100                 }
3101         }
3102
3103         /*
3104          * delete a xattr doesn't need metadata and cluster allocation.
3105          * so just calculate the credits and return.
3106          *
3107          * The credits for removing the value tree will be extended
3108          * by ocfs2_remove_extent itself.
3109          */
3110         if (!xi->xi_value) {
3111                 if (!ocfs2_xattr_is_local(xe))
3112                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3113
3114                 goto out;
3115         }
3116
3117         /* do cluster allocation guess first. */
3118         value_size = le64_to_cpu(xe->xe_value_size);
3119
3120         if (old_in_xb) {
3121                 /*
3122                  * In xattr set, we always try to set the xe in inode first,
3123                  * so if it can be inserted into inode successfully, the old
3124                  * one will be removed from the xattr block, and this xattr
3125                  * will be inserted into inode as a new xattr in inode.
3126                  */
3127                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3128                         clusters_add += new_clusters;
3129                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3130                                     OCFS2_INODE_UPDATE_CREDITS;
3131                         if (!ocfs2_xattr_is_local(xe))
3132                                 credits += ocfs2_calc_extend_credits(
3133                                                         inode->i_sb,
3134                                                         &def_xv.xv.xr_list);
3135                         goto out;
3136                 }
3137         }
3138
3139         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3140                 /* the new values will be stored outside. */
3141                 u32 old_clusters = 0;
3142
3143                 if (!ocfs2_xattr_is_local(xe)) {
3144                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3145                                                                  value_size);
3146                         xv = (struct ocfs2_xattr_value_root *)
3147                              (base + name_offset + name_len);
3148                         value_size = OCFS2_XATTR_ROOT_SIZE;
3149                 } else
3150                         xv = &def_xv.xv;
3151
3152                 if (old_clusters >= new_clusters) {
3153                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3154                         goto out;
3155                 } else {
3156                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3157                         clusters_add += new_clusters - old_clusters;
3158                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3159                                                              &xv->xr_list);
3160                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3161                                 goto out;
3162                 }
3163         } else {
3164                 /*
3165                  * Now the new value will be stored inside. So if the new
3166                  * value is smaller than the size of value root or the old
3167                  * value, we don't need any allocation, otherwise we have
3168                  * to guess metadata allocation.
3169                  */
3170                 if ((ocfs2_xattr_is_local(xe) &&
3171                      (value_size >= xi->xi_value_len)) ||
3172                     (!ocfs2_xattr_is_local(xe) &&
3173                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3174                         goto out;
3175         }
3176
3177 meta_guess:
3178         /* calculate metadata allocation. */
3179         if (di->i_xattr_loc) {
3180                 if (!xbs->xattr_bh) {
3181                         ret = ocfs2_read_xattr_block(inode,
3182                                                      le64_to_cpu(di->i_xattr_loc),
3183                                                      &bh);
3184                         if (ret) {
3185                                 mlog_errno(ret);
3186                                 goto out;
3187                         }
3188
3189                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3190                 } else
3191                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3192
3193                 /*
3194                  * If there is already an xattr tree, good, we can calculate
3195                  * like other b-trees. Otherwise we may have the chance of
3196                  * create a tree, the credit calculation is borrowed from
3197                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3198                  * new tree will be cluster based, so no meta is needed.
3199                  */
3200                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3201                         struct ocfs2_extent_list *el =
3202                                  &xb->xb_attrs.xb_root.xt_list;
3203                         meta_add += ocfs2_extend_meta_needed(el);
3204                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3205                                                              el);
3206                 } else
3207                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3208
3209                 /*
3210                  * This cluster will be used either for new bucket or for
3211                  * new xattr block.
3212                  * If the cluster size is the same as the bucket size, one
3213                  * more is needed since we may need to extend the bucket
3214                  * also.
3215                  */
3216                 clusters_add += 1;
3217                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3218                 if (OCFS2_XATTR_BUCKET_SIZE ==
3219                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3220                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3221                         clusters_add += 1;
3222                 }
3223         } else {
3224                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3225                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3226                         struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3227                         meta_add += ocfs2_extend_meta_needed(el);
3228                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3229                                                              el);
3230                 } else {
3231                         meta_add += 1;
3232                 }
3233         }
3234 out:
3235         if (clusters_need)
3236                 *clusters_need = clusters_add;
3237         if (meta_need)
3238                 *meta_need = meta_add;
3239         if (credits_need)
3240                 *credits_need = credits;
3241         brelse(bh);
3242         return ret;
3243 }
3244
3245 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3246                                      struct ocfs2_dinode *di,
3247                                      struct ocfs2_xattr_info *xi,
3248                                      struct ocfs2_xattr_search *xis,
3249                                      struct ocfs2_xattr_search *xbs,
3250                                      struct ocfs2_xattr_set_ctxt *ctxt,
3251                                      int extra_meta,
3252                                      int *credits)
3253 {
3254         int clusters_add, meta_add, ret;
3255         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3256
3257         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3258
3259         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3260
3261         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3262                                         &clusters_add, &meta_add, credits);
3263         if (ret) {
3264                 mlog_errno(ret);
3265                 return ret;
3266         }
3267
3268         meta_add += extra_meta;
3269         trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3270                                         clusters_add, *credits);
3271
3272         if (meta_add) {
3273                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3274                                                         &ctxt->meta_ac);
3275                 if (ret) {
3276                         mlog_errno(ret);
3277                         goto out;
3278                 }
3279         }
3280
3281         if (clusters_add) {
3282                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3283                 if (ret)
3284                         mlog_errno(ret);
3285         }
3286 out:
3287         if (ret) {
3288                 if (ctxt->meta_ac) {
3289                         ocfs2_free_alloc_context(ctxt->meta_ac);
3290                         ctxt->meta_ac = NULL;
3291                 }
3292
3293                 /*
3294                  * We cannot have an error and a non null ctxt->data_ac.
3295                  */
3296         }
3297
3298         return ret;
3299 }
3300
3301 static int __ocfs2_xattr_set_handle(struct inode *inode,
3302                                     struct ocfs2_dinode *di,
3303                                     struct ocfs2_xattr_info *xi,
3304                                     struct ocfs2_xattr_search *xis,
3305                                     struct ocfs2_xattr_search *xbs,
3306                                     struct ocfs2_xattr_set_ctxt *ctxt)
3307 {
3308         int ret = 0, credits, old_found;
3309
3310         if (!xi->xi_value) {
3311                 /* Remove existing extended attribute */
3312                 if (!xis->not_found)
3313                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3314                 else if (!xbs->not_found)
3315                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3316         } else {
3317                 /* We always try to set extended attribute into inode first*/
3318                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3319                 if (!ret && !xbs->not_found) {
3320                         /*
3321                          * If succeed and that extended attribute existing in
3322                          * external block, then we will remove it.
3323                          */
3324                         xi->xi_value = NULL;
3325                         xi->xi_value_len = 0;
3326
3327                         old_found = xis->not_found;
3328                         xis->not_found = -ENODATA;
3329                         ret = ocfs2_calc_xattr_set_need(inode,
3330                                                         di,
3331                                                         xi,
3332                                                         xis,
3333                                                         xbs,
3334                                                         NULL,
3335                                                         NULL,
3336                                                         &credits);
3337                         xis->not_found = old_found;
3338                         if (ret) {
3339                                 mlog_errno(ret);
3340                                 goto out;
3341                         }
3342
3343                         ret = ocfs2_extend_trans(ctxt->handle, credits);
3344                         if (ret) {
3345                                 mlog_errno(ret);
3346                                 goto out;
3347                         }
3348                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3349                 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3350                         if (di->i_xattr_loc && !xbs->xattr_bh) {
3351                                 ret = ocfs2_xattr_block_find(inode,
3352                                                              xi->xi_name_index,
3353                                                              xi->xi_name, xbs);
3354                                 if (ret)
3355                                         goto out;
3356
3357                                 old_found = xis->not_found;
3358                                 xis->not_found = -ENODATA;
3359                                 ret = ocfs2_calc_xattr_set_need(inode,
3360                                                                 di,
3361                                                                 xi,
3362                                                                 xis,
3363                                                                 xbs,
3364                                                                 NULL,
3365                                                                 NULL,
3366                                                                 &credits);
3367                                 xis->not_found = old_found;
3368                                 if (ret) {
3369                                         mlog_errno(ret);
3370                                         goto out;
3371                                 }
3372
3373                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3374                                 if (ret) {
3375                                         mlog_errno(ret);
3376                                         goto out;
3377                                 }
3378                         }
3379                         /*
3380                          * If no space in inode, we will set extended attribute
3381                          * into external block.
3382                          */
3383                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3384                         if (ret)
3385                                 goto out;
3386                         if (!xis->not_found) {
3387                                 /*
3388                                  * If succeed and that extended attribute
3389                                  * existing in inode, we will remove it.
3390                                  */
3391                                 xi->xi_value = NULL;
3392                                 xi->xi_value_len = 0;
3393                                 xbs->not_found = -ENODATA;
3394                                 ret = ocfs2_calc_xattr_set_need(inode,
3395                                                                 di,
3396                                                                 xi,
3397                                                                 xis,
3398                                                                 xbs,
3399                                                                 NULL,
3400                                                                 NULL,
3401                                                                 &credits);
3402                                 if (ret) {
3403                                         mlog_errno(ret);
3404                                         goto out;
3405                                 }
3406
3407                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3408                                 if (ret) {
3409                                         mlog_errno(ret);
3410                                         goto out;
3411                                 }
3412                                 ret = ocfs2_xattr_ibody_set(inode, xi,
3413                                                             xis, ctxt);
3414                         }
3415                 }
3416         }
3417
3418         if (!ret) {
3419                 /* Update inode ctime. */
3420                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3421                                               xis->inode_bh,
3422                                               OCFS2_JOURNAL_ACCESS_WRITE);
3423                 if (ret) {
3424                         mlog_errno(ret);
3425                         goto out;
3426                 }
3427
3428                 inode->i_ctime = current_time(inode);
3429                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3430                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3431                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3432         }
3433 out:
3434         return ret;
3435 }
3436
3437 /*
3438  * This function only called duing creating inode
3439  * for init security/acl xattrs of the new inode.
3440  * All transanction credits have been reserved in mknod.
3441  */
3442 int ocfs2_xattr_set_handle(handle_t *handle,
3443                            struct inode *inode,
3444                            struct buffer_head *di_bh,
3445                            int name_index,
3446                            const char *name,
3447                            const void *value,
3448                            size_t value_len,
3449                            int flags,
3450                            struct ocfs2_alloc_context *meta_ac,
3451                            struct ocfs2_alloc_context *data_ac)
3452 {
3453         struct ocfs2_dinode *di;
3454         int ret;
3455
3456         struct ocfs2_xattr_info xi = {
3457                 .xi_name_index = name_index,
3458                 .xi_name = name,
3459                 .xi_name_len = strlen(name),
3460                 .xi_value = value,
3461                 .xi_value_len = value_len,
3462         };
3463
3464         struct ocfs2_xattr_search xis = {
3465                 .not_found = -ENODATA,
3466         };
3467
3468         struct ocfs2_xattr_search xbs = {
3469                 .not_found = -ENODATA,
3470         };
3471
3472         struct ocfs2_xattr_set_ctxt ctxt = {
3473                 .handle = handle,
3474                 .meta_ac = meta_ac,
3475                 .data_ac = data_ac,
3476         };
3477
3478         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3479                 return -EOPNOTSUPP;
3480
3481         /*
3482          * In extreme situation, may need xattr bucket when
3483          * block size is too small. And we have already reserved
3484          * the credits for bucket in mknod.
3485          */
3486         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3487                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3488                 if (!xbs.bucket) {
3489                         mlog_errno(-ENOMEM);
3490                         return -ENOMEM;
3491                 }
3492         }
3493
3494         xis.inode_bh = xbs.inode_bh = di_bh;
3495         di = (struct ocfs2_dinode *)di_bh->b_data;
3496
3497         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3498
3499         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3500         if (ret)
3501                 goto cleanup;
3502         if (xis.not_found) {
3503                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3504                 if (ret)
3505                         goto cleanup;
3506         }
3507
3508         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3509
3510 cleanup:
3511         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3512         brelse(xbs.xattr_bh);
3513         ocfs2_xattr_bucket_free(xbs.bucket);
3514
3515         return ret;
3516 }
3517
3518 /*
3519  * ocfs2_xattr_set()
3520  *
3521  * Set, replace or remove an extended attribute for this inode.
3522  * value is NULL to remove an existing extended attribute, else either
3523  * create or replace an extended attribute.
3524  */
3525 int ocfs2_xattr_set(struct inode *inode,
3526                     int name_index,
3527                     const char *name,
3528                     const void *value,
3529                     size_t value_len,
3530                     int flags)
3531 {
3532         struct buffer_head *di_bh = NULL;
3533         struct ocfs2_dinode *di;
3534         int ret, credits, had_lock, ref_meta = 0, ref_credits = 0;
3535         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3536         struct inode *tl_inode = osb->osb_tl_inode;
3537         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3538         struct ocfs2_refcount_tree *ref_tree = NULL;
3539         struct ocfs2_lock_holder oh;
3540
3541         struct ocfs2_xattr_info xi = {
3542                 .xi_name_index = name_index,
3543                 .xi_name = name,
3544                 .xi_name_len = strlen(name),
3545                 .xi_value = value,
3546                 .xi_value_len = value_len,
3547         };
3548
3549         struct ocfs2_xattr_search xis = {
3550                 .not_found = -ENODATA,
3551         };
3552
3553         struct ocfs2_xattr_search xbs = {
3554                 .not_found = -ENODATA,
3555         };
3556
3557         if (!ocfs2_supports_xattr(osb))
3558                 return -EOPNOTSUPP;
3559
3560         /*
3561          * Only xbs will be used on indexed trees.  xis doesn't need a
3562          * bucket.
3563          */
3564         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3565         if (!xbs.bucket) {
3566                 mlog_errno(-ENOMEM);
3567                 return -ENOMEM;
3568         }
3569
3570         had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 1, &oh);
3571         if (had_lock < 0) {
3572                 ret = had_lock;
3573                 mlog_errno(ret);
3574                 goto cleanup_nolock;
3575         }
3576         xis.inode_bh = xbs.inode_bh = di_bh;
3577         di = (struct ocfs2_dinode *)di_bh->b_data;
3578
3579         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3580         /*
3581          * Scan inode and external block to find the same name
3582          * extended attribute and collect search information.
3583          */
3584         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3585         if (ret)
3586                 goto cleanup;
3587         if (xis.not_found) {
3588                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3589                 if (ret)
3590                         goto cleanup;
3591         }
3592
3593         if (xis.not_found && xbs.not_found) {
3594                 ret = -ENODATA;
3595                 if (flags & XATTR_REPLACE)
3596                         goto cleanup;
3597                 ret = 0;
3598                 if (!value)
3599                         goto cleanup;
3600         } else {
3601                 ret = -EEXIST;
3602                 if (flags & XATTR_CREATE)
3603                         goto cleanup;
3604         }
3605
3606         /* Check whether the value is refcounted and do some preparation. */
3607         if (ocfs2_is_refcount_inode(inode) &&
3608             (!xis.not_found || !xbs.not_found)) {
3609                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3610                                                    &xis, &xbs, &ref_tree,
3611                                                    &ref_meta, &ref_credits);
3612                 if (ret) {
3613                         mlog_errno(ret);
3614                         goto cleanup;
3615                 }
3616         }
3617
3618         inode_lock(tl_inode);
3619
3620         if (ocfs2_truncate_log_needs_flush(osb)) {
3621                 ret = __ocfs2_flush_truncate_log(osb);
3622                 if (ret < 0) {
3623                         inode_unlock(tl_inode);
3624                         mlog_errno(ret);
3625                         goto cleanup;
3626                 }
3627         }
3628         inode_unlock(tl_inode);
3629
3630         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3631                                         &xbs, &ctxt, ref_meta, &credits);
3632         if (ret) {
3633                 mlog_errno(ret);
3634                 goto cleanup;
3635         }
3636
3637         /* we need to update inode's ctime field, so add credit for it. */
3638         credits += OCFS2_INODE_UPDATE_CREDITS;
3639         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3640         if (IS_ERR(ctxt.handle)) {
3641                 ret = PTR_ERR(ctxt.handle);
3642                 mlog_errno(ret);
3643                 goto out_free_ac;
3644         }
3645
3646         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3647         ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3648
3649         ocfs2_commit_trans(osb, ctxt.handle);
3650
3651 out_free_ac:
3652         if (ctxt.data_ac)
3653                 ocfs2_free_alloc_context(ctxt.data_ac);
3654         if (ctxt.meta_ac)
3655                 ocfs2_free_alloc_context(ctxt.meta_ac);
3656         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3657                 ocfs2_schedule_truncate_log_flush(osb, 1);
3658         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3659
3660 cleanup:
3661         if (ref_tree)
3662                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3663         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3664         if (!value && !ret) {
3665                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3666                 if (ret)
3667                         mlog_errno(ret);
3668         }
3669         ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
3670 cleanup_nolock:
3671         brelse(di_bh);
3672         brelse(xbs.xattr_bh);
3673         ocfs2_xattr_bucket_free(xbs.bucket);
3674
3675         return ret;
3676 }
3677
3678 /*
3679  * Find the xattr extent rec which may contains name_hash.
3680  * e_cpos will be the first name hash of the xattr rec.
3681  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3682  */
3683 static int ocfs2_xattr_get_rec(struct inode *inode,
3684                                u32 name_hash,
3685                                u64 *p_blkno,
3686                                u32 *e_cpos,
3687                                u32 *num_clusters,
3688                                struct ocfs2_extent_list *el)
3689 {
3690         int ret = 0, i;
3691         struct buffer_head *eb_bh = NULL;
3692         struct ocfs2_extent_block *eb;
3693         struct ocfs2_extent_rec *rec = NULL;
3694         u64 e_blkno = 0;
3695
3696         if (el->l_tree_depth) {
3697                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3698                                       &eb_bh);
3699                 if (ret) {
3700                         mlog_errno(ret);
3701                         goto out;
3702                 }
3703
3704                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3705                 el = &eb->h_list;
3706
3707                 if (el->l_tree_depth) {
3708                         ret = ocfs2_error(inode->i_sb,
3709                                           "Inode %lu has non zero tree depth in xattr tree block %llu\n",
3710                                           inode->i_ino,
3711                                           (unsigned long long)eb_bh->b_blocknr);
3712                         goto out;
3713                 }
3714         }
3715
3716         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3717                 rec = &el->l_recs[i];
3718
3719                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3720                         e_blkno = le64_to_cpu(rec->e_blkno);
3721                         break;
3722                 }
3723         }
3724
3725         if (!e_blkno) {
3726                 ret = ocfs2_error(inode->i_sb, "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
3727                                   inode->i_ino,
3728                                   le32_to_cpu(rec->e_cpos),
3729                                   ocfs2_rec_clusters(el, rec));
3730                 goto out;
3731         }
3732
3733         *p_blkno = le64_to_cpu(rec->e_blkno);
3734         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3735         if (e_cpos)
3736                 *e_cpos = le32_to_cpu(rec->e_cpos);
3737 out:
3738         brelse(eb_bh);
3739         return ret;
3740 }
3741
3742 typedef int (xattr_bucket_func)(struct inode *inode,
3743                                 struct ocfs2_xattr_bucket *bucket,
3744                                 void *para);
3745
3746 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3747                                    struct ocfs2_xattr_bucket *bucket,
3748                                    int name_index,
3749                                    const char *name,
3750                                    u32 name_hash,
3751                                    u16 *xe_index,
3752                                    int *found)
3753 {
3754         int i, ret = 0, cmp = 1, block_off, new_offset;
3755         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3756         size_t name_len = strlen(name);
3757         struct ocfs2_xattr_entry *xe = NULL;
3758         char *xe_name;
3759
3760         /*
3761          * We don't use binary search in the bucket because there
3762          * may be multiple entries with the same name hash.
3763          */
3764         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3765                 xe = &xh->xh_entries[i];
3766
3767                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3768                         continue;
3769                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3770                         break;
3771
3772                 cmp = name_index - ocfs2_xattr_get_type(xe);
3773                 if (!cmp)
3774                         cmp = name_len - xe->xe_name_len;
3775                 if (cmp)
3776                         continue;
3777
3778                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3779                                                         xh,
3780                                                         i,
3781                                                         &block_off,
3782                                                         &new_offset);
3783                 if (ret) {
3784                         mlog_errno(ret);
3785                         break;
3786                 }
3787
3788
3789                 xe_name = bucket_block(bucket, block_off) + new_offset;
3790                 if (!memcmp(name, xe_name, name_len)) {
3791                         *xe_index = i;
3792                         *found = 1;
3793                         ret = 0;
3794                         break;
3795                 }
3796         }
3797
3798         return ret;
3799 }
3800
3801 /*
3802  * Find the specified xattr entry in a series of buckets.
3803  * This series start from p_blkno and last for num_clusters.
3804  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3805  * the num of the valid buckets.
3806  *
3807  * Return the buffer_head this xattr should reside in. And if the xattr's
3808  * hash is in the gap of 2 buckets, return the lower bucket.
3809  */
3810 static int ocfs2_xattr_bucket_find(struct inode *inode,
3811                                    int name_index,
3812                                    const char *name,
3813                                    u32 name_hash,
3814                                    u64 p_blkno,
3815                                    u32 first_hash,
3816                                    u32 num_clusters,
3817                                    struct ocfs2_xattr_search *xs)
3818 {
3819         int ret, found = 0;
3820         struct ocfs2_xattr_header *xh = NULL;
3821         struct ocfs2_xattr_entry *xe = NULL;
3822         u16 index = 0;
3823         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3824         int low_bucket = 0, bucket, high_bucket;
3825         struct ocfs2_xattr_bucket *search;
3826         u64 blkno, lower_blkno = 0;
3827
3828         search = ocfs2_xattr_bucket_new(inode);
3829         if (!search) {
3830                 ret = -ENOMEM;
3831                 mlog_errno(ret);
3832                 goto out;
3833         }
3834
3835         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3836         if (ret) {
3837                 mlog_errno(ret);
3838                 goto out;
3839         }
3840
3841         xh = bucket_xh(search);
3842         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3843         while (low_bucket <= high_bucket) {
3844                 ocfs2_xattr_bucket_relse(search);
3845
3846                 bucket = (low_bucket + high_bucket) / 2;
3847                 blkno = p_blkno + bucket * blk_per_bucket;
3848                 ret = ocfs2_read_xattr_bucket(search, blkno);
3849                 if (ret) {
3850                         mlog_errno(ret);
3851                         goto out;
3852                 }
3853
3854                 xh = bucket_xh(search);
3855                 xe = &xh->xh_entries[0];
3856                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3857                         high_bucket = bucket - 1;
3858                         continue;
3859                 }
3860
3861                 /*
3862                  * Check whether the hash of the last entry in our
3863                  * bucket is larger than the search one. for an empty
3864                  * bucket, the last one is also the first one.
3865                  */
3866                 if (xh->xh_count)
3867                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3868
3869                 /* record lower_blkno which may be the insert place. */
3870                 lower_blkno = blkno;
3871
3872                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3873                         low_bucket = bucket + 1;
3874                         continue;
3875                 }
3876
3877                 /* the searched xattr should reside in this bucket if exists. */
3878                 ret = ocfs2_find_xe_in_bucket(inode, search,
3879                                               name_index, name, name_hash,
3880                                               &index, &found);
3881                 if (ret) {
3882                         mlog_errno(ret);
3883                         goto out;
3884                 }
3885                 break;
3886         }
3887
3888         /*
3889          * Record the bucket we have found.
3890          * When the xattr's hash value is in the gap of 2 buckets, we will
3891          * always set it to the previous bucket.
3892          */
3893         if (!lower_blkno)
3894                 lower_blkno = p_blkno;
3895
3896         /* This should be in cache - we just read it during the search */
3897         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3898         if (ret) {
3899                 mlog_errno(ret);
3900                 goto out;
3901         }
3902
3903         xs->header = bucket_xh(xs->bucket);
3904         xs->base = bucket_block(xs->bucket, 0);
3905         xs->end = xs->base + inode->i_sb->s_blocksize;
3906
3907         if (found) {
3908                 xs->here = &xs->header->xh_entries[index];
3909                 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3910                         name, name_index, name_hash,
3911                         (unsigned long long)bucket_blkno(xs->bucket),
3912                         index);
3913         } else
3914                 ret = -ENODATA;
3915
3916 out:
3917         ocfs2_xattr_bucket_free(search);
3918         return ret;
3919 }
3920
3921 static int ocfs2_xattr_index_block_find(struct inode *inode,
3922                                         struct buffer_head *root_bh,
3923                                         int name_index,
3924                                         const char *name,
3925                                         struct ocfs2_xattr_search *xs)
3926 {
3927         int ret;
3928         struct ocfs2_xattr_block *xb =
3929                         (struct ocfs2_xattr_block *)root_bh->b_data;
3930         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3931         struct ocfs2_extent_list *el = &xb_root->xt_list;
3932         u64 p_blkno = 0;
3933         u32 first_hash, num_clusters = 0;
3934         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3935
3936         if (le16_to_cpu(el->l_next_free_rec) == 0)
3937                 return -ENODATA;
3938
3939         trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3940                                         name, name_index, name_hash,
3941                                         (unsigned long long)root_bh->b_blocknr,
3942                                         -1);
3943
3944         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3945                                   &num_clusters, el);
3946         if (ret) {
3947                 mlog_errno(ret);
3948                 goto out;
3949         }
3950
3951         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3952
3953         trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3954                                         name, name_index, first_hash,
3955                                         (unsigned long long)p_blkno,
3956                                         num_clusters);
3957
3958         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3959                                       p_blkno, first_hash, num_clusters, xs);
3960
3961 out:
3962         return ret;
3963 }
3964
3965 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3966                                        u64 blkno,
3967                                        u32 clusters,
3968                                        xattr_bucket_func *func,
3969                                        void *para)
3970 {
3971         int i, ret = 0;
3972         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3973         u32 num_buckets = clusters * bpc;
3974         struct ocfs2_xattr_bucket *bucket;
3975
3976         bucket = ocfs2_xattr_bucket_new(inode);
3977         if (!bucket) {
3978                 mlog_errno(-ENOMEM);
3979                 return -ENOMEM;
3980         }
3981
3982         trace_ocfs2_iterate_xattr_buckets(
3983                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3984                 (unsigned long long)blkno, clusters);
3985
3986         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3987                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3988                 if (ret) {
3989                         mlog_errno(ret);
3990                         break;
3991                 }
3992
3993                 /*
3994                  * The real bucket num in this series of blocks is stored
3995                  * in the 1st bucket.
3996                  */
3997                 if (i == 0)
3998                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3999
4000                 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
4001                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
4002                 if (func) {
4003                         ret = func(inode, bucket, para);
4004                         if (ret && ret != -ERANGE)
4005                                 mlog_errno(ret);
4006                         /* Fall through to bucket_relse() */
4007                 }
4008
4009                 ocfs2_xattr_bucket_relse(bucket);
4010                 if (ret)
4011                         break;
4012         }
4013
4014         ocfs2_xattr_bucket_free(bucket);
4015         return ret;
4016 }
4017
4018 struct ocfs2_xattr_tree_list {
4019         char *buffer;
4020         size_t buffer_size;
4021         size_t result;
4022 };
4023
4024 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4025                                              struct ocfs2_xattr_header *xh,
4026                                              int index,
4027                                              int *block_off,
4028                                              int *new_offset)
4029 {
4030         u16 name_offset;
4031
4032         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4033                 return -EINVAL;
4034
4035         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4036
4037         *block_off = name_offset >> sb->s_blocksize_bits;
4038         *new_offset = name_offset % sb->s_blocksize;
4039
4040         return 0;
4041 }
4042
4043 static int ocfs2_list_xattr_bucket(struct inode *inode,
4044                                    struct ocfs2_xattr_bucket *bucket,
4045                                    void *para)
4046 {
4047         int ret = 0, type;
4048         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4049         int i, block_off, new_offset;
4050         const char *name;
4051
4052         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4053                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4054                 type = ocfs2_xattr_get_type(entry);
4055
4056                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4057                                                         bucket_xh(bucket),
4058                                                         i,
4059                                                         &block_off,
4060                                                         &new_offset);
4061                 if (ret)
4062                         break;
4063
4064                 name = (const char *)bucket_block(bucket, block_off) +
4065                         new_offset;
4066                 ret = ocfs2_xattr_list_entry(inode->i_sb,
4067                                              xl->buffer,
4068                                              xl->buffer_size,
4069                                              &xl->result,
4070                                              type, name,
4071                                              entry->xe_name_len);
4072                 if (ret)
4073                         break;
4074         }
4075
4076         return ret;
4077 }
4078
4079 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4080                                            struct buffer_head *blk_bh,
4081                                            xattr_tree_rec_func *rec_func,
4082                                            void *para)
4083 {
4084         struct ocfs2_xattr_block *xb =
4085                         (struct ocfs2_xattr_block *)blk_bh->b_data;
4086         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4087         int ret = 0;
4088         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4089         u64 p_blkno = 0;
4090
4091         if (!el->l_next_free_rec || !rec_func)
4092                 return 0;
4093
4094         while (name_hash > 0) {
4095                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4096                                           &e_cpos, &num_clusters, el);
4097                 if (ret) {
4098                         mlog_errno(ret);
4099                         break;
4100                 }
4101
4102                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4103                                num_clusters, para);
4104                 if (ret) {
4105                         if (ret != -ERANGE)
4106                                 mlog_errno(ret);
4107                         break;
4108                 }
4109
4110                 if (e_cpos == 0)
4111                         break;
4112
4113                 name_hash = e_cpos - 1;
4114         }
4115
4116         return ret;
4117
4118 }
4119
4120 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4121                                      struct buffer_head *root_bh,
4122                                      u64 blkno, u32 cpos, u32 len, void *para)
4123 {
4124         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4125                                            ocfs2_list_xattr_bucket, para);
4126 }
4127
4128 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4129                                              struct buffer_head *blk_bh,
4130                                              char *buffer,
4131                                              size_t buffer_size)
4132 {
4133         int ret;
4134         struct ocfs2_xattr_tree_list xl = {
4135                 .buffer = buffer,
4136                 .buffer_size = buffer_size,
4137                 .result = 0,
4138         };
4139
4140         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4141                                               ocfs2_list_xattr_tree_rec, &xl);
4142         if (ret) {
4143                 mlog_errno(ret);
4144                 goto out;
4145         }
4146
4147         ret = xl.result;
4148 out:
4149         return ret;
4150 }
4151
4152 static int cmp_xe(const void *a, const void *b)
4153 {
4154         const struct ocfs2_xattr_entry *l = a, *r = b;
4155         u32 l_hash = le32_to_cpu(l->xe_name_hash);
4156         u32 r_hash = le32_to_cpu(r->xe_name_hash);
4157
4158         if (l_hash > r_hash)
4159                 return 1;
4160         if (l_hash < r_hash)
4161                 return -1;
4162         return 0;
4163 }
4164
4165 static void swap_xe(void *a, void *b, int size)
4166 {
4167         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4168
4169         tmp = *l;
4170         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4171         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4172 }
4173
4174 /*
4175  * When the ocfs2_xattr_block is filled up, new bucket will be created
4176  * and all the xattr entries will be moved to the new bucket.
4177  * The header goes at the start of the bucket, and the names+values are
4178  * filled from the end.  This is why *target starts as the last buffer.
4179  * Note: we need to sort the entries since they are not saved in order
4180  * in the ocfs2_xattr_block.
4181  */
4182 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4183                                            struct buffer_head *xb_bh,
4184                                            struct ocfs2_xattr_bucket *bucket)
4185 {
4186         int i, blocksize = inode->i_sb->s_blocksize;
4187         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4188         u16 offset, size, off_change;
4189         struct ocfs2_xattr_entry *xe;
4190         struct ocfs2_xattr_block *xb =
4191                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
4192         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4193         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4194         u16 count = le16_to_cpu(xb_xh->xh_count);
4195         char *src = xb_bh->b_data;
4196         char *target = bucket_block(bucket, blks - 1);
4197
4198         trace_ocfs2_cp_xattr_block_to_bucket_begin(
4199                                 (unsigned long long)xb_bh->b_blocknr,
4200                                 (unsigned long long)bucket_blkno(bucket));
4201
4202         for (i = 0; i < blks; i++)
4203                 memset(bucket_block(bucket, i), 0, blocksize);
4204
4205         /*
4206          * Since the xe_name_offset is based on ocfs2_xattr_header,
4207          * there is a offset change corresponding to the change of
4208          * ocfs2_xattr_header's position.
4209          */
4210         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4211         xe = &xb_xh->xh_entries[count - 1];
4212         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4213         size = blocksize - offset;
4214
4215         /* copy all the names and values. */
4216         memcpy(target + offset, src + offset, size);
4217
4218         /* Init new header now. */
4219         xh->xh_count = xb_xh->xh_count;
4220         xh->xh_num_buckets = cpu_to_le16(1);
4221         xh->xh_name_value_len = cpu_to_le16(size);
4222         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4223
4224         /* copy all the entries. */
4225         target = bucket_block(bucket, 0);
4226         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4227         size = count * sizeof(struct ocfs2_xattr_entry);
4228         memcpy(target + offset, (char *)xb_xh + offset, size);
4229
4230         /* Change the xe offset for all the xe because of the move. */
4231         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4232                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4233         for (i = 0; i < count; i++)
4234                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4235
4236         trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4237
4238         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4239              cmp_xe, swap_xe);
4240 }
4241
4242 /*
4243  * After we move xattr from block to index btree, we have to
4244  * update ocfs2_xattr_search to the new xe and base.
4245  *
4246  * When the entry is in xattr block, xattr_bh indicates the storage place.
4247  * While if the entry is in index b-tree, "bucket" indicates the
4248  * real place of the xattr.
4249  */
4250 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4251                                             struct ocfs2_xattr_search *xs,
4252                                             struct buffer_head *old_bh)
4253 {
4254         char *buf = old_bh->b_data;
4255         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4256         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4257         int i;
4258
4259         xs->header = bucket_xh(xs->bucket);
4260         xs->base = bucket_block(xs->bucket, 0);
4261         xs->end = xs->base + inode->i_sb->s_blocksize;
4262
4263         if (xs->not_found)
4264                 return;
4265
4266         i = xs->here - old_xh->xh_entries;
4267         xs->here = &xs->header->xh_entries[i];
4268 }
4269
4270 static int ocfs2_xattr_create_index_block(struct inode *inode,
4271                                           struct ocfs2_xattr_search *xs,
4272                                           struct ocfs2_xattr_set_ctxt *ctxt)
4273 {
4274         int ret;
4275         u32 bit_off, len;
4276         u64 blkno;
4277         handle_t *handle = ctxt->handle;
4278         struct ocfs2_inode_info *oi = OCFS2_I(inode);
4279         struct buffer_head *xb_bh = xs->xattr_bh;
4280         struct ocfs2_xattr_block *xb =
4281                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4282         struct ocfs2_xattr_tree_root *xr;
4283         u16 xb_flags = le16_to_cpu(xb->xb_flags);
4284
4285         trace_ocfs2_xattr_create_index_block_begin(
4286                                 (unsigned long long)xb_bh->b_blocknr);
4287
4288         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4289         BUG_ON(!xs->bucket);
4290
4291         /*
4292          * XXX:
4293          * We can use this lock for now, and maybe move to a dedicated mutex
4294          * if performance becomes a problem later.
4295          */
4296         down_write(&oi->ip_alloc_sem);
4297
4298         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4299                                       OCFS2_JOURNAL_ACCESS_WRITE);
4300         if (ret) {
4301                 mlog_errno(ret);
4302                 goto out;
4303         }
4304
4305         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4306                                      1, 1, &bit_off, &len);
4307         if (ret) {
4308                 mlog_errno(ret);
4309                 goto out;
4310         }
4311
4312         /*
4313          * The bucket may spread in many blocks, and
4314          * we will only touch the 1st block and the last block
4315          * in the whole bucket(one for entry and one for data).
4316          */
4317         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4318
4319         trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4320
4321         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4322         if (ret) {
4323                 mlog_errno(ret);
4324                 goto out;
4325         }
4326
4327         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4328                                                 OCFS2_JOURNAL_ACCESS_CREATE);
4329         if (ret) {
4330                 mlog_errno(ret);
4331                 goto out;
4332         }
4333
4334         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4335         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4336
4337         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4338
4339         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4340         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4341                offsetof(struct ocfs2_xattr_block, xb_attrs));
4342
4343         xr = &xb->xb_attrs.xb_root;
4344         xr->xt_clusters = cpu_to_le32(1);
4345         xr->xt_last_eb_blk = 0;
4346         xr->xt_list.l_tree_depth = 0;
4347         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4348         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4349
4350         xr->xt_list.l_recs[0].e_cpos = 0;
4351         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4352         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4353
4354         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4355
4356         ocfs2_journal_dirty(handle, xb_bh);
4357
4358 out:
4359         up_write(&oi->ip_alloc_sem);
4360
4361         return ret;
4362 }
4363
4364 static int cmp_xe_offset(const void *a, const void *b)
4365 {
4366         const struct ocfs2_xattr_entry *l = a, *r = b;
4367         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4368         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4369
4370         if (l_name_offset < r_name_offset)
4371                 return 1;
4372         if (l_name_offset > r_name_offset)
4373                 return -1;
4374         return 0;
4375 }
4376
4377 /*
4378  * defrag a xattr bucket if we find that the bucket has some
4379  * holes beteen name/value pairs.
4380  * We will move all the name/value pairs to the end of the bucket
4381  * so that we can spare some space for insertion.
4382  */
4383 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4384                                      handle_t *handle,
4385                                      struct ocfs2_xattr_bucket *bucket)
4386 {
4387         int ret, i;
4388         size_t end, offset, len;
4389         struct ocfs2_xattr_header *xh;
4390         char *entries, *buf, *bucket_buf = NULL;
4391         u64 blkno = bucket_blkno(bucket);
4392         u16 xh_free_start;
4393         size_t blocksize = inode->i_sb->s_blocksize;
4394         struct ocfs2_xattr_entry *xe;
4395
4396         /*
4397          * In order to make the operation more efficient and generic,
4398          * we copy all the blocks into a contiguous memory and do the
4399          * defragment there, so if anything is error, we will not touch
4400          * the real block.
4401          */
4402         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4403         if (!bucket_buf) {
4404                 ret = -EIO;
4405                 goto out;
4406         }
4407
4408         buf = bucket_buf;
4409         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4410                 memcpy(buf, bucket_block(bucket, i), blocksize);
4411
4412         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4413                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4414         if (ret < 0) {
4415                 mlog_errno(ret);
4416                 goto out;
4417         }
4418
4419         xh = (struct ocfs2_xattr_header *)bucket_buf;
4420         entries = (char *)xh->xh_entries;
4421         xh_free_start = le16_to_cpu(xh->xh_free_start);
4422
4423         trace_ocfs2_defrag_xattr_bucket(
4424              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4425              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4426
4427         /*
4428          * sort all the entries by their offset.
4429          * the largest will be the first, so that we can
4430          * move them to the end one by one.
4431          */
4432         sort(entries, le16_to_cpu(xh->xh_count),
4433              sizeof(struct ocfs2_xattr_entry),
4434              cmp_xe_offset, swap_xe);
4435
4436         /* Move all name/values to the end of the bucket. */
4437         xe = xh->xh_entries;
4438         end = OCFS2_XATTR_BUCKET_SIZE;
4439         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4440                 offset = le16_to_cpu(xe->xe_name_offset);
4441                 len = namevalue_size_xe(xe);
4442
4443                 /*
4444                  * We must make sure that the name/value pair
4445                  * exist in the same block. So adjust end to
4446                  * the previous block end if needed.
4447                  */
4448                 if (((end - len) / blocksize !=
4449                         (end - 1) / blocksize))
4450                         end = end - end % blocksize;
4451
4452                 if (end > offset + len) {
4453                         memmove(bucket_buf + end - len,
4454                                 bucket_buf + offset, len);
4455                         xe->xe_name_offset = cpu_to_le16(end - len);
4456                 }
4457
4458                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4459                                 "bucket %llu\n", (unsigned long long)blkno);
4460
4461                 end -= len;
4462         }
4463
4464         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4465                         "bucket %llu\n", (unsigned long long)blkno);
4466
4467         if (xh_free_start == end)
4468                 goto out;
4469
4470         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4471         xh->xh_free_start = cpu_to_le16(end);
4472
4473         /* sort the entries by their name_hash. */
4474         sort(entries, le16_to_cpu(xh->xh_count),
4475              sizeof(struct ocfs2_xattr_entry),
4476              cmp_xe, swap_xe);
4477
4478         buf = bucket_buf;
4479         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4480                 memcpy(bucket_block(bucket, i), buf, blocksize);
4481         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4482
4483 out:
4484         kfree(bucket_buf);
4485         return ret;
4486 }
4487
4488 /*
4489  * prev_blkno points to the start of an existing extent.  new_blkno
4490  * points to a newly allocated extent.  Because we know each of our
4491  * clusters contains more than bucket, we can easily split one cluster
4492  * at a bucket boundary.  So we take the last cluster of the existing
4493  * extent and split it down the middle.  We move the last half of the
4494  * buckets in the last cluster of the existing extent over to the new
4495  * extent.
4496  *
4497  * first_bh is the buffer at prev_blkno so we can update the existing
4498  * extent's bucket count.  header_bh is the bucket were we were hoping
4499  * to insert our xattr.  If the bucket move places the target in the new
4500  * extent, we'll update first_bh and header_bh after modifying the old
4501  * extent.
4502  *
4503  * first_hash will be set as the 1st xe's name_hash in the new extent.
4504  */
4505 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4506                                                handle_t *handle,
4507                                                struct ocfs2_xattr_bucket *first,
4508                                                struct ocfs2_xattr_bucket *target,
4509                                                u64 new_blkno,
4510                                                u32 num_clusters,
4511                                                u32 *first_hash)
4512 {
4513         int ret;
4514         struct super_block *sb = inode->i_sb;
4515         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4516         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4517         int to_move = num_buckets / 2;
4518         u64 src_blkno;
4519         u64 last_cluster_blkno = bucket_blkno(first) +
4520                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4521
4522         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4523         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4524
4525         trace_ocfs2_mv_xattr_bucket_cross_cluster(
4526                                 (unsigned long long)last_cluster_blkno,
4527                                 (unsigned long long)new_blkno);
4528
4529         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4530                                      last_cluster_blkno, new_blkno,
4531                                      to_move, first_hash);
4532         if (ret) {
4533                 mlog_errno(ret);
4534                 goto out;
4535         }
4536
4537         /* This is the first bucket that got moved */
4538         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4539
4540         /*
4541          * If the target bucket was part of the moved buckets, we need to
4542          * update first and target.
4543          */
4544         if (bucket_blkno(target) >= src_blkno) {
4545                 /* Find the block for the new target bucket */
4546                 src_blkno = new_blkno +
4547                         (bucket_blkno(target) - src_blkno);
4548
4549                 ocfs2_xattr_bucket_relse(first);
4550                 ocfs2_xattr_bucket_relse(target);
4551
4552                 /*
4553                  * These shouldn't fail - the buffers are in the
4554                  * journal from ocfs2_cp_xattr_bucket().
4555                  */
4556                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4557                 if (ret) {
4558                         mlog_errno(ret);
4559                         goto out;
4560                 }
4561                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4562                 if (ret)
4563                         mlog_errno(ret);
4564
4565         }
4566
4567 out:
4568         return ret;
4569 }
4570
4571 /*
4572  * Find the suitable pos when we divide a bucket into 2.
4573  * We have to make sure the xattrs with the same hash value exist
4574  * in the same bucket.
4575  *
4576  * If this ocfs2_xattr_header covers more than one hash value, find a
4577  * place where the hash value changes.  Try to find the most even split.
4578  * The most common case is that all entries have different hash values,
4579  * and the first check we make will find a place to split.
4580  */
4581 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4582 {
4583         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4584         int count = le16_to_cpu(xh->xh_count);
4585         int delta, middle = count / 2;
4586
4587         /*
4588          * We start at the middle.  Each step gets farther away in both
4589          * directions.  We therefore hit the change in hash value
4590          * nearest to the middle.  Note that this loop does not execute for
4591          * count < 2.
4592          */
4593         for (delta = 0; delta < middle; delta++) {
4594                 /* Let's check delta earlier than middle */
4595                 if (cmp_xe(&entries[middle - delta - 1],
4596                            &entries[middle - delta]))
4597                         return middle - delta;
4598
4599                 /* For even counts, don't walk off the end */
4600                 if ((middle + delta + 1) == count)
4601                         continue;
4602
4603                 /* Now try delta past middle */
4604                 if (cmp_xe(&entries[middle + delta],
4605                            &entries[middle + delta + 1]))
4606                         return middle + delta + 1;
4607         }
4608
4609         /* Every entry had the same hash */
4610         return count;
4611 }
4612
4613 /*
4614  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4615  * first_hash will record the 1st hash of the new bucket.
4616  *
4617  * Normally half of the xattrs will be moved.  But we have to make
4618  * sure that the xattrs with the same hash value are stored in the
4619  * same bucket. If all the xattrs in this bucket have the same hash
4620  * value, the new bucket will be initialized as an empty one and the
4621  * first_hash will be initialized as (hash_value+1).
4622  */
4623 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4624                                     handle_t *handle,
4625                                     u64 blk,
4626                                     u64 new_blk,
4627                                     u32 *first_hash,
4628                                     int new_bucket_head)
4629 {
4630         int ret, i;
4631         int count, start, len, name_value_len = 0, name_offset = 0;
4632         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4633         struct ocfs2_xattr_header *xh;
4634         struct ocfs2_xattr_entry *xe;
4635         int blocksize = inode->i_sb->s_blocksize;
4636
4637         trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4638                                               (unsigned long long)new_blk);
4639
4640         s_bucket = ocfs2_xattr_bucket_new(inode);
4641         t_bucket = ocfs2_xattr_bucket_new(inode);
4642         if (!s_bucket || !t_bucket) {
4643                 ret = -ENOMEM;
4644                 mlog_errno(ret);
4645                 goto out;
4646         }
4647
4648         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4649         if (ret) {
4650                 mlog_errno(ret);
4651                 goto out;
4652         }
4653
4654         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4655                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4656         if (ret) {
4657                 mlog_errno(ret);
4658                 goto out;
4659         }
4660
4661         /*
4662          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4663          * there's no need to read it.
4664          */
4665         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4666         if (ret) {
4667                 mlog_errno(ret);
4668                 goto out;
4669         }
4670
4671         /*
4672          * Hey, if we're overwriting t_bucket, what difference does
4673          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4674          * same part of ocfs2_cp_xattr_bucket().
4675          */
4676         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4677                                                 new_bucket_head ?
4678                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4679                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4680         if (ret) {
4681                 mlog_errno(ret);
4682                 goto out;
4683         }
4684
4685         xh = bucket_xh(s_bucket);
4686         count = le16_to_cpu(xh->xh_count);
4687         start = ocfs2_xattr_find_divide_pos(xh);
4688
4689         if (start == count) {
4690                 xe = &xh->xh_entries[start-1];
4691
4692                 /*
4693                  * initialized a new empty bucket here.
4694                  * The hash value is set as one larger than
4695                  * that of the last entry in the previous bucket.
4696                  */
4697                 for (i = 0; i < t_bucket->bu_blocks; i++)
4698                         memset(bucket_block(t_bucket, i), 0, blocksize);
4699
4700                 xh = bucket_xh(t_bucket);
4701                 xh->xh_free_start = cpu_to_le16(blocksize);
4702                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4703                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4704
4705                 goto set_num_buckets;
4706         }
4707
4708         /* copy the whole bucket to the new first. */
4709         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4710
4711         /* update the new bucket. */
4712         xh = bucket_xh(t_bucket);
4713
4714         /*
4715          * Calculate the total name/value len and xh_free_start for
4716          * the old bucket first.
4717          */
4718         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4719         name_value_len = 0;
4720         for (i = 0; i < start; i++) {
4721                 xe = &xh->xh_entries[i];
4722                 name_value_len += namevalue_size_xe(xe);
4723                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4724                         name_offset = le16_to_cpu(xe->xe_name_offset);
4725         }
4726
4727         /*
4728          * Now begin the modification to the new bucket.
4729          *
4730          * In the new bucket, We just move the xattr entry to the beginning
4731          * and don't touch the name/value. So there will be some holes in the
4732          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4733          * called.
4734          */
4735         xe = &xh->xh_entries[start];
4736         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4737         trace_ocfs2_divide_xattr_bucket_move(len,
4738                         (int)((char *)xe - (char *)xh),
4739                         (int)((char *)xh->xh_entries - (char *)xh));
4740         memmove((char *)xh->xh_entries, (char *)xe, len);
4741         xe = &xh->xh_entries[count - start];
4742         len = sizeof(struct ocfs2_xattr_entry) * start;
4743         memset((char *)xe, 0, len);
4744
4745         le16_add_cpu(&xh->xh_count, -start);
4746         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4747
4748         /* Calculate xh_free_start for the new bucket. */
4749         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4750         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4751                 xe = &xh->xh_entries[i];
4752                 if (le16_to_cpu(xe->xe_name_offset) <
4753                     le16_to_cpu(xh->xh_free_start))
4754                         xh->xh_free_start = xe->xe_name_offset;
4755         }
4756
4757 set_num_buckets:
4758         /* set xh->xh_num_buckets for the new xh. */
4759         if (new_bucket_head)
4760                 xh->xh_num_buckets = cpu_to_le16(1);
4761         else
4762                 xh->xh_num_buckets = 0;
4763
4764         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4765
4766         /* store the first_hash of the new bucket. */
4767         if (first_hash)
4768                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4769
4770         /*
4771          * Now only update the 1st block of the old bucket.  If we
4772          * just added a new empty bucket, there is no need to modify
4773          * it.
4774          */
4775         if (start == count)
4776                 goto out;
4777
4778         xh = bucket_xh(s_bucket);
4779         memset(&xh->xh_entries[start], 0,
4780                sizeof(struct ocfs2_xattr_entry) * (count - start));
4781         xh->xh_count = cpu_to_le16(start);
4782         xh->xh_free_start = cpu_to_le16(name_offset);
4783         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4784
4785         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4786
4787 out:
4788         ocfs2_xattr_bucket_free(s_bucket);
4789         ocfs2_xattr_bucket_free(t_bucket);
4790
4791         return ret;
4792 }
4793
4794 /*
4795  * Copy xattr from one bucket to another bucket.
4796  *
4797  * The caller must make sure that the journal transaction
4798  * has enough space for journaling.
4799  */
4800 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4801                                  handle_t *handle,
4802                                  u64 s_blkno,
4803                                  u64 t_blkno,
4804                                  int t_is_new)
4805 {
4806         int ret;
4807         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4808
4809         BUG_ON(s_blkno == t_blkno);
4810
4811         trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4812                                     (unsigned long long)t_blkno,
4813                                     t_is_new);
4814
4815         s_bucket = ocfs2_xattr_bucket_new(inode);
4816         t_bucket = ocfs2_xattr_bucket_new(inode);
4817         if (!s_bucket || !t_bucket) {
4818                 ret = -ENOMEM;
4819                 mlog_errno(ret);
4820                 goto out;
4821         }
4822
4823         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4824         if (ret)
4825                 goto out;
4826
4827         /*
4828          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4829          * there's no need to read it.
4830          */
4831         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4832         if (ret)
4833                 goto out;
4834
4835         /*
4836          * Hey, if we're overwriting t_bucket, what difference does
4837          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4838          * cluster to fill, we came here from
4839          * ocfs2_mv_xattr_buckets(), and it is really new -
4840          * ACCESS_CREATE is required.  But we also might have moved data
4841          * out of t_bucket before extending back into it.
4842          * ocfs2_add_new_xattr_bucket() can do this - its call to
4843          * ocfs2_add_new_xattr_cluster() may have created a new extent
4844          * and copied out the end of the old extent.  Then it re-extends
4845          * the old extent back to create space for new xattrs.  That's
4846          * how we get here, and the bucket isn't really new.
4847          */
4848         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4849                                                 t_is_new ?
4850                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4851                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4852         if (ret)
4853                 goto out;
4854
4855         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4856         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4857
4858 out:
4859         ocfs2_xattr_bucket_free(t_bucket);
4860         ocfs2_xattr_bucket_free(s_bucket);
4861
4862         return ret;
4863 }
4864
4865 /*
4866  * src_blk points to the start of an existing extent.  last_blk points to
4867  * last cluster in that extent.  to_blk points to a newly allocated
4868  * extent.  We copy the buckets from the cluster at last_blk to the new
4869  * extent.  If start_bucket is non-zero, we skip that many buckets before
4870  * we start copying.  The new extent's xh_num_buckets gets set to the
4871  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4872  * by the same amount.
4873  */
4874 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4875                                   u64 src_blk, u64 last_blk, u64 to_blk,
4876                                   unsigned int start_bucket,
4877                                   u32 *first_hash)
4878 {
4879         int i, ret, credits;
4880         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4881         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4882         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4883         struct ocfs2_xattr_bucket *old_first, *new_first;
4884
4885         trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4886                                      (unsigned long long)to_blk);
4887
4888         BUG_ON(start_bucket >= num_buckets);
4889         if (start_bucket) {
4890                 num_buckets -= start_bucket;
4891                 last_blk += (start_bucket * blks_per_bucket);
4892         }
4893
4894         /* The first bucket of the original extent */
4895         old_first = ocfs2_xattr_bucket_new(inode);
4896         /* The first bucket of the new extent */
4897         new_first = ocfs2_xattr_bucket_new(inode);
4898         if (!old_first || !new_first) {
4899                 ret = -ENOMEM;
4900                 mlog_errno(ret);
4901                 goto out;
4902         }
4903
4904         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4905         if (ret) {
4906                 mlog_errno(ret);
4907                 goto out;
4908         }
4909
4910         /*
4911          * We need to update the first bucket of the old extent and all
4912          * the buckets going to the new extent.
4913          */
4914         credits = ((num_buckets + 1) * blks_per_bucket);
4915         ret = ocfs2_extend_trans(handle, credits);
4916         if (ret) {
4917                 mlog_errno(ret);
4918                 goto out;
4919         }
4920
4921         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4922                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4923         if (ret) {
4924                 mlog_errno(ret);
4925                 goto out;
4926         }
4927
4928         for (i = 0; i < num_buckets; i++) {
4929                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4930                                             last_blk + (i * blks_per_bucket),
4931                                             to_blk + (i * blks_per_bucket),
4932                                             1);
4933                 if (ret) {
4934                         mlog_errno(ret);
4935                         goto out;
4936                 }
4937         }
4938
4939         /*
4940          * Get the new bucket ready before we dirty anything
4941          * (This actually shouldn't fail, because we already dirtied
4942          * it once in ocfs2_cp_xattr_bucket()).
4943          */
4944         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4945         if (ret) {
4946                 mlog_errno(ret);
4947                 goto out;
4948         }
4949         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4950                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4951         if (ret) {
4952                 mlog_errno(ret);
4953                 goto out;
4954         }
4955
4956         /* Now update the headers */
4957         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4958         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4959
4960         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4961         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4962
4963         if (first_hash)
4964                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4965
4966 out:
4967         ocfs2_xattr_bucket_free(new_first);
4968         ocfs2_xattr_bucket_free(old_first);
4969         return ret;
4970 }
4971
4972 /*
4973  * Move some xattrs in this cluster to the new cluster.
4974  * This function should only be called when bucket size == cluster size.
4975  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4976  */
4977 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4978                                       handle_t *handle,
4979                                       u64 prev_blk,
4980                                       u64 new_blk,
4981                                       u32 *first_hash)
4982 {
4983         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4984         int ret, credits = 2 * blk_per_bucket;
4985
4986         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4987
4988         ret = ocfs2_extend_trans(handle, credits);
4989         if (ret) {
4990                 mlog_errno(ret);
4991                 return ret;
4992         }
4993
4994         /* Move half of the xattr in start_blk to the next bucket. */
4995         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4996                                           new_blk, first_hash, 1);
4997 }
4998
4999 /*
5000  * Move some xattrs from the old cluster to the new one since they are not
5001  * contiguous in ocfs2 xattr tree.
5002  *
5003  * new_blk starts a new separate cluster, and we will move some xattrs from
5004  * prev_blk to it. v_start will be set as the first name hash value in this
5005  * new cluster so that it can be used as e_cpos during tree insertion and
5006  * don't collide with our original b-tree operations. first_bh and header_bh
5007  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5008  * to extend the insert bucket.
5009  *
5010  * The problem is how much xattr should we move to the new one and when should
5011  * we update first_bh and header_bh?
5012  * 1. If cluster size > bucket size, that means the previous cluster has more
5013  *    than 1 bucket, so just move half nums of bucket into the new cluster and
5014  *    update the first_bh and header_bh if the insert bucket has been moved
5015  *    to the new cluster.
5016  * 2. If cluster_size == bucket_size:
5017  *    a) If the previous extent rec has more than one cluster and the insert
5018  *       place isn't in the last cluster, copy the entire last cluster to the
5019  *       new one. This time, we don't need to upate the first_bh and header_bh
5020  *       since they will not be moved into the new cluster.
5021  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
5022  *       the new one. And we set the extend flag to zero if the insert place is
5023  *       moved into the new allocated cluster since no extend is needed.
5024  */
5025 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5026                                             handle_t *handle,
5027                                             struct ocfs2_xattr_bucket *first,
5028                                             struct ocfs2_xattr_bucket *target,
5029                                             u64 new_blk,
5030                                             u32 prev_clusters,
5031                                             u32 *v_start,
5032                                             int *extend)
5033 {
5034         int ret;
5035
5036         trace_ocfs2_adjust_xattr_cross_cluster(
5037                         (unsigned long long)bucket_blkno(first),
5038                         (unsigned long long)new_blk, prev_clusters);
5039
5040         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5041                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5042                                                           handle,
5043                                                           first, target,
5044                                                           new_blk,
5045                                                           prev_clusters,
5046                                                           v_start);
5047                 if (ret)
5048                         mlog_errno(ret);
5049         } else {
5050                 /* The start of the last cluster in the first extent */
5051                 u64 last_blk = bucket_blkno(first) +
5052                         ((prev_clusters - 1) *
5053                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
5054
5055                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5056                         ret = ocfs2_mv_xattr_buckets(inode, handle,
5057                                                      bucket_blkno(first),
5058                                                      last_blk, new_blk, 0,
5059                                                      v_start);
5060                         if (ret)
5061                                 mlog_errno(ret);
5062                 } else {
5063                         ret = ocfs2_divide_xattr_cluster(inode, handle,
5064                                                          last_blk, new_blk,
5065                                                          v_start);
5066                         if (ret)
5067                                 mlog_errno(ret);
5068
5069                         if ((bucket_blkno(target) == last_blk) && extend)
5070                                 *extend = 0;
5071                 }
5072         }
5073
5074         return ret;
5075 }
5076
5077 /*
5078  * Add a new cluster for xattr storage.
5079  *
5080  * If the new cluster is contiguous with the previous one, it will be
5081  * appended to the same extent record, and num_clusters will be updated.
5082  * If not, we will insert a new extent for it and move some xattrs in
5083  * the last cluster into the new allocated one.
5084  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5085  * lose the benefits of hashing because we'll have to search large leaves.
5086  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5087  * if it's bigger).
5088  *
5089  * first_bh is the first block of the previous extent rec and header_bh
5090  * indicates the bucket we will insert the new xattrs. They will be updated
5091  * when the header_bh is moved into the new cluster.
5092  */
5093 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5094                                        struct buffer_head *root_bh,
5095                                        struct ocfs2_xattr_bucket *first,
5096                                        struct ocfs2_xattr_bucket *target,
5097                                        u32 *num_clusters,
5098                                        u32 prev_cpos,
5099                                        int *extend,
5100                                        struct ocfs2_xattr_set_ctxt *ctxt)
5101 {
5102         int ret;
5103         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5104         u32 prev_clusters = *num_clusters;
5105         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5106         u64 block;
5107         handle_t *handle = ctxt->handle;
5108         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5109         struct ocfs2_extent_tree et;
5110
5111         trace_ocfs2_add_new_xattr_cluster_begin(
5112                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5113                 (unsigned long long)bucket_blkno(first),
5114                 prev_cpos, prev_clusters);
5115
5116         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5117
5118         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5119                                       OCFS2_JOURNAL_ACCESS_WRITE);
5120         if (ret < 0) {
5121                 mlog_errno(ret);
5122                 goto leave;
5123         }
5124
5125         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5126                                      clusters_to_add, &bit_off, &num_bits);
5127         if (ret < 0) {
5128                 if (ret != -ENOSPC)
5129                         mlog_errno(ret);
5130                 goto leave;
5131         }
5132
5133         BUG_ON(num_bits > clusters_to_add);
5134
5135         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5136         trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5137
5138         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5139             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5140              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5141                 /*
5142                  * If this cluster is contiguous with the old one and
5143                  * adding this new cluster, we don't surpass the limit of
5144                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5145                  * initialized and used like other buckets in the previous
5146                  * cluster.
5147                  * So add it as a contiguous one. The caller will handle
5148                  * its init process.
5149                  */
5150                 v_start = prev_cpos + prev_clusters;
5151                 *num_clusters = prev_clusters + num_bits;
5152         } else {
5153                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5154                                                        handle,
5155                                                        first,
5156                                                        target,
5157                                                        block,
5158                                                        prev_clusters,
5159                                                        &v_start,
5160                                                        extend);
5161                 if (ret) {
5162                         mlog_errno(ret);
5163                         goto leave;
5164                 }
5165         }
5166
5167         trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5168                                                  v_start, num_bits);
5169         ret = ocfs2_insert_extent(handle, &et, v_start, block,
5170                                   num_bits, 0, ctxt->meta_ac);
5171         if (ret < 0) {
5172                 mlog_errno(ret);
5173                 goto leave;
5174         }
5175
5176         ocfs2_journal_dirty(handle, root_bh);
5177
5178 leave:
5179         return ret;
5180 }
5181
5182 /*
5183  * We are given an extent.  'first' is the bucket at the very front of
5184  * the extent.  The extent has space for an additional bucket past
5185  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5186  * of the target bucket.  We wish to shift every bucket past the target
5187  * down one, filling in that additional space.  When we get back to the
5188  * target, we split the target between itself and the now-empty bucket
5189  * at target+1 (aka, target_blkno + blks_per_bucket).
5190  */
5191 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5192                                      handle_t *handle,
5193                                      struct ocfs2_xattr_bucket *first,
5194                                      u64 target_blk,
5195                                      u32 num_clusters)
5196 {
5197         int ret, credits;
5198         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5199         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5200         u64 end_blk;
5201         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5202
5203         trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5204                                         (unsigned long long)bucket_blkno(first),
5205                                         num_clusters, new_bucket);
5206
5207         /* The extent must have room for an additional bucket */
5208         BUG_ON(new_bucket >=
5209                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5210
5211         /* end_blk points to the last existing bucket */
5212         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5213
5214         /*
5215          * end_blk is the start of the last existing bucket.
5216          * Thus, (end_blk - target_blk) covers the target bucket and
5217          * every bucket after it up to, but not including, the last
5218          * existing bucket.  Then we add the last existing bucket, the
5219          * new bucket, and the first bucket (3 * blk_per_bucket).
5220          */
5221         credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5222         ret = ocfs2_extend_trans(handle, credits);
5223         if (ret) {
5224                 mlog_errno(ret);
5225                 goto out;
5226         }
5227
5228         ret = ocfs2_xattr_bucket_journal_access(handle, first,
5229                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5230         if (ret) {
5231                 mlog_errno(ret);
5232                 goto out;
5233         }
5234
5235         while (end_blk != target_blk) {
5236                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5237                                             end_blk + blk_per_bucket, 0);
5238                 if (ret)
5239                         goto out;
5240                 end_blk -= blk_per_bucket;
5241         }
5242
5243         /* Move half of the xattr in target_blkno to the next bucket. */
5244         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5245                                         target_blk + blk_per_bucket, NULL, 0);
5246
5247         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5248         ocfs2_xattr_bucket_journal_dirty(handle, first);
5249
5250 out:
5251         return ret;
5252 }
5253
5254 /*
5255  * Add new xattr bucket in an extent record and adjust the buckets
5256  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5257  * bucket we want to insert into.
5258  *
5259  * In the easy case, we will move all the buckets after target down by
5260  * one. Half of target's xattrs will be moved to the next bucket.
5261  *
5262  * If current cluster is full, we'll allocate a new one.  This may not
5263  * be contiguous.  The underlying calls will make sure that there is
5264  * space for the insert, shifting buckets around if necessary.
5265  * 'target' may be moved by those calls.
5266  */
5267 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5268                                       struct buffer_head *xb_bh,
5269                                       struct ocfs2_xattr_bucket *target,
5270                                       struct ocfs2_xattr_set_ctxt *ctxt)
5271 {
5272         struct ocfs2_xattr_block *xb =
5273                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5274         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5275         struct ocfs2_extent_list *el = &xb_root->xt_list;
5276         u32 name_hash =
5277                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5278         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5279         int ret, num_buckets, extend = 1;
5280         u64 p_blkno;
5281         u32 e_cpos, num_clusters;
5282         /* The bucket at the front of the extent */
5283         struct ocfs2_xattr_bucket *first;
5284
5285         trace_ocfs2_add_new_xattr_bucket(
5286                                 (unsigned long long)bucket_blkno(target));
5287
5288         /* The first bucket of the original extent */
5289         first = ocfs2_xattr_bucket_new(inode);
5290         if (!first) {
5291                 ret = -ENOMEM;
5292                 mlog_errno(ret);
5293                 goto out;
5294         }
5295
5296         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5297                                   &num_clusters, el);
5298         if (ret) {
5299                 mlog_errno(ret);
5300                 goto out;
5301         }
5302
5303         ret = ocfs2_read_xattr_bucket(first, p_blkno);
5304         if (ret) {
5305                 mlog_errno(ret);
5306                 goto out;
5307         }
5308
5309         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5310         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5311                 /*
5312                  * This can move first+target if the target bucket moves
5313                  * to the new extent.
5314                  */
5315                 ret = ocfs2_add_new_xattr_cluster(inode,
5316                                                   xb_bh,
5317                                                   first,
5318                                                   target,
5319                                                   &num_clusters,
5320                                                   e_cpos,
5321                                                   &extend,
5322                                                   ctxt);
5323                 if (ret) {
5324                         mlog_errno(ret);
5325                         goto out;
5326                 }
5327         }
5328
5329         if (extend) {
5330                 ret = ocfs2_extend_xattr_bucket(inode,
5331                                                 ctxt->handle,
5332                                                 first,
5333                                                 bucket_blkno(target),
5334                                                 num_clusters);
5335                 if (ret)
5336                         mlog_errno(ret);
5337         }
5338
5339 out:
5340         ocfs2_xattr_bucket_free(first);
5341
5342         return ret;
5343 }
5344
5345 /*
5346  * Truncate the specified xe_off entry in xattr bucket.
5347  * bucket is indicated by header_bh and len is the new length.
5348  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5349  *
5350  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5351  */
5352 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5353                                              struct ocfs2_xattr_bucket *bucket,
5354                                              int xe_off,
5355                                              int len,
5356                                              struct ocfs2_xattr_set_ctxt *ctxt)
5357 {
5358         int ret, offset;
5359         u64 value_blk;
5360         struct ocfs2_xattr_entry *xe;
5361         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5362         size_t blocksize = inode->i_sb->s_blocksize;
5363         struct ocfs2_xattr_value_buf vb = {
5364                 .vb_access = ocfs2_journal_access,
5365         };
5366
5367         xe = &xh->xh_entries[xe_off];
5368
5369         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5370
5371         offset = le16_to_cpu(xe->xe_name_offset) +
5372                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5373
5374         value_blk = offset / blocksize;
5375
5376         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5377         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5378
5379         vb.vb_bh = bucket->bu_bhs[value_blk];
5380         BUG_ON(!vb.vb_bh);
5381
5382         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5383                 (vb.vb_bh->b_data + offset % blocksize);
5384
5385         /*
5386          * From here on out we have to dirty the bucket.  The generic
5387          * value calls only modify one of the bucket's bhs, but we need
5388          * to send the bucket at once.  So if they error, they *could* have
5389          * modified something.  We have to assume they did, and dirty
5390          * the whole bucket.  This leaves us in a consistent state.
5391          */
5392         trace_ocfs2_xattr_bucket_value_truncate(
5393                         (unsigned long long)bucket_blkno(bucket), xe_off, len);
5394         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5395         if (ret) {
5396                 mlog_errno(ret);
5397                 goto out;
5398         }
5399
5400         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5401                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5402         if (ret) {
5403                 mlog_errno(ret);
5404                 goto out;
5405         }
5406
5407         xe->xe_value_size = cpu_to_le64(len);
5408
5409         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5410
5411 out:
5412         return ret;
5413 }
5414
5415 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5416                                   struct buffer_head *root_bh,
5417                                   u64 blkno,
5418                                   u32 cpos,
5419                                   u32 len,
5420                                   void *para)
5421 {
5422         int ret;
5423         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5424         struct inode *tl_inode = osb->osb_tl_inode;
5425         handle_t *handle;
5426         struct ocfs2_xattr_block *xb =
5427                         (struct ocfs2_xattr_block *)root_bh->b_data;
5428         struct ocfs2_alloc_context *meta_ac = NULL;
5429         struct ocfs2_cached_dealloc_ctxt dealloc;
5430         struct ocfs2_extent_tree et;
5431
5432         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5433                                           ocfs2_delete_xattr_in_bucket, para);
5434         if (ret) {
5435                 mlog_errno(ret);
5436                 return ret;
5437         }
5438
5439         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5440
5441         ocfs2_init_dealloc_ctxt(&dealloc);
5442
5443         trace_ocfs2_rm_xattr_cluster(
5444                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
5445                         (unsigned long long)blkno, cpos, len);
5446
5447         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5448                                                len);
5449
5450         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5451         if (ret) {
5452                 mlog_errno(ret);
5453                 return ret;
5454         }
5455
5456         inode_lock(tl_inode);
5457
5458         if (ocfs2_truncate_log_needs_flush(osb)) {
5459                 ret = __ocfs2_flush_truncate_log(osb);
5460                 if (ret < 0) {
5461                         mlog_errno(ret);
5462                         goto out;
5463                 }
5464         }
5465
5466         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5467         if (IS_ERR(handle)) {
5468                 ret = -ENOMEM;
5469                 mlog_errno(ret);
5470                 goto out;
5471         }
5472
5473         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5474                                       OCFS2_JOURNAL_ACCESS_WRITE);
5475         if (ret) {
5476                 mlog_errno(ret);
5477                 goto out_commit;
5478         }
5479
5480         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5481                                   &dealloc);
5482         if (ret) {
5483                 mlog_errno(ret);
5484                 goto out_commit;
5485         }
5486
5487         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5488         ocfs2_journal_dirty(handle, root_bh);
5489
5490         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5491         if (ret)
5492                 mlog_errno(ret);
5493         ocfs2_update_inode_fsync_trans(handle, inode, 0);
5494
5495 out_commit:
5496         ocfs2_commit_trans(osb, handle);
5497 out:
5498         ocfs2_schedule_truncate_log_flush(osb, 1);
5499
5500         inode_unlock(tl_inode);
5501
5502         if (meta_ac)
5503                 ocfs2_free_alloc_context(meta_ac);
5504
5505         ocfs2_run_deallocs(osb, &dealloc);
5506
5507         return ret;
5508 }
5509
5510 /*
5511  * check whether the xattr bucket is filled up with the same hash value.
5512  * If we want to insert the xattr with the same hash, return -ENOSPC.
5513  * If we want to insert a xattr with different hash value, go ahead
5514  * and ocfs2_divide_xattr_bucket will handle this.
5515  */
5516 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5517                                               struct ocfs2_xattr_bucket *bucket,
5518                                               const char *name)
5519 {
5520         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5521         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5522
5523         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5524                 return 0;
5525
5526         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5527             xh->xh_entries[0].xe_name_hash) {
5528                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5529                      "hash = %u\n",
5530                      (unsigned long long)bucket_blkno(bucket),
5531                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5532                 return -ENOSPC;
5533         }
5534
5535         return 0;
5536 }
5537
5538 /*
5539  * Try to set the entry in the current bucket.  If we fail, the caller
5540  * will handle getting us another bucket.
5541  */
5542 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5543                                         struct ocfs2_xattr_info *xi,
5544                                         struct ocfs2_xattr_search *xs,
5545                                         struct ocfs2_xattr_set_ctxt *ctxt)
5546 {
5547         int ret;
5548         struct ocfs2_xa_loc loc;
5549
5550         trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5551
5552         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5553                                        xs->not_found ? NULL : xs->here);
5554         ret = ocfs2_xa_set(&loc, xi, ctxt);
5555         if (!ret) {
5556                 xs->here = loc.xl_entry;
5557                 goto out;
5558         }
5559         if (ret != -ENOSPC) {
5560                 mlog_errno(ret);
5561                 goto out;
5562         }
5563
5564         /* Ok, we need space.  Let's try defragmenting the bucket. */
5565         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5566                                         xs->bucket);
5567         if (ret) {
5568                 mlog_errno(ret);
5569                 goto out;
5570         }
5571
5572         ret = ocfs2_xa_set(&loc, xi, ctxt);
5573         if (!ret) {
5574                 xs->here = loc.xl_entry;
5575                 goto out;
5576         }
5577         if (ret != -ENOSPC)
5578                 mlog_errno(ret);
5579
5580
5581 out:
5582         return ret;
5583 }
5584
5585 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5586                                              struct ocfs2_xattr_info *xi,
5587                                              struct ocfs2_xattr_search *xs,
5588                                              struct ocfs2_xattr_set_ctxt *ctxt)
5589 {
5590         int ret;
5591
5592         trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5593
5594         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5595         if (!ret)
5596                 goto out;
5597         if (ret != -ENOSPC) {
5598                 mlog_errno(ret);
5599                 goto out;
5600         }
5601
5602         /* Ack, need more space.  Let's try to get another bucket! */
5603
5604         /*
5605          * We do not allow for overlapping ranges between buckets. And
5606          * the maximum number of collisions we will allow for then is
5607          * one bucket's worth, so check it here whether we need to
5608          * add a new bucket for the insert.
5609          */
5610         ret = ocfs2_check_xattr_bucket_collision(inode,
5611                                                  xs->bucket,
5612                                                  xi->xi_name);
5613         if (ret) {
5614                 mlog_errno(ret);
5615                 goto out;
5616         }
5617
5618         ret = ocfs2_add_new_xattr_bucket(inode,
5619                                          xs->xattr_bh,
5620                                          xs->bucket,
5621                                          ctxt);
5622         if (ret) {
5623                 mlog_errno(ret);
5624                 goto out;
5625         }
5626
5627         /*
5628          * ocfs2_add_new_xattr_bucket() will have updated
5629          * xs->bucket if it moved, but it will not have updated
5630          * any of the other search fields.  Thus, we drop it and
5631          * re-search.  Everything should be cached, so it'll be
5632          * quick.
5633          */
5634         ocfs2_xattr_bucket_relse(xs->bucket);
5635         ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5636                                            xi->xi_name_index,
5637                                            xi->xi_name, xs);
5638         if (ret && ret != -ENODATA)
5639                 goto out;
5640         xs->not_found = ret;
5641
5642         /* Ok, we have a new bucket, let's try again */
5643         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5644         if (ret && (ret != -ENOSPC))
5645                 mlog_errno(ret);
5646
5647 out:
5648         return ret;
5649 }
5650
5651 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5652                                         struct ocfs2_xattr_bucket *bucket,
5653                                         void *para)
5654 {
5655         int ret = 0, ref_credits;
5656         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5657         u16 i;
5658         struct ocfs2_xattr_entry *xe;
5659         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5660         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5661         int credits = ocfs2_remove_extent_credits(osb->sb) +
5662                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5663         struct ocfs2_xattr_value_root *xv;
5664         struct ocfs2_rm_xattr_bucket_para *args =
5665                         (struct ocfs2_rm_xattr_bucket_para *)para;
5666
5667         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5668
5669         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5670                 xe = &xh->xh_entries[i];
5671                 if (ocfs2_xattr_is_local(xe))
5672                         continue;
5673
5674                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5675                                                       i, &xv, NULL);
5676                 if (ret) {
5677                         mlog_errno(ret);
5678                         break;
5679                 }
5680
5681                 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5682                                                          args->ref_ci,
5683                                                          args->ref_root_bh,
5684                                                          &ctxt.meta_ac,
5685                                                          &ref_credits);
5686
5687                 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5688                 if (IS_ERR(ctxt.handle)) {
5689                         ret = PTR_ERR(ctxt.handle);
5690                         mlog_errno(ret);
5691                         break;
5692                 }
5693
5694                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5695                                                         i, 0, &ctxt);
5696
5697                 ocfs2_commit_trans(osb, ctxt.handle);
5698                 if (ctxt.meta_ac) {
5699                         ocfs2_free_alloc_context(ctxt.meta_ac);
5700                         ctxt.meta_ac = NULL;
5701                 }
5702                 if (ret) {
5703                         mlog_errno(ret);
5704                         break;
5705                 }
5706         }
5707
5708         if (ctxt.meta_ac)
5709                 ocfs2_free_alloc_context(ctxt.meta_ac);
5710         ocfs2_schedule_truncate_log_flush(osb, 1);
5711         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5712         return ret;
5713 }
5714
5715 /*
5716  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5717  * or change the extent record flag), we need to recalculate
5718  * the metaecc for the whole bucket. So it is done here.
5719  *
5720  * Note:
5721  * We have to give the extra credits for the caller.
5722  */
5723 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5724                                             handle_t *handle,
5725                                             void *para)
5726 {
5727         int ret;
5728         struct ocfs2_xattr_bucket *bucket =
5729                         (struct ocfs2_xattr_bucket *)para;
5730
5731         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5732                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5733         if (ret) {
5734                 mlog_errno(ret);
5735                 return ret;
5736         }
5737
5738         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5739
5740         return 0;
5741 }
5742
5743 /*
5744  * Special action we need if the xattr value is refcounted.
5745  *
5746  * 1. If the xattr is refcounted, lock the tree.
5747  * 2. CoW the xattr if we are setting the new value and the value
5748  *    will be stored outside.
5749  * 3. In other case, decrease_refcount will work for us, so just
5750  *    lock the refcount tree, calculate the meta and credits is OK.
5751  *
5752  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5753  * currently CoW is a completed transaction, while this function
5754  * will also lock the allocators and let us deadlock. So we will
5755  * CoW the whole xattr value.
5756  */
5757 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5758                                         struct ocfs2_dinode *di,
5759                                         struct ocfs2_xattr_info *xi,
5760                                         struct ocfs2_xattr_search *xis,
5761                                         struct ocfs2_xattr_search *xbs,
5762                                         struct ocfs2_refcount_tree **ref_tree,
5763                                         int *meta_add,
5764                                         int *credits)
5765 {
5766         int ret = 0;
5767         struct ocfs2_xattr_block *xb;
5768         struct ocfs2_xattr_entry *xe;
5769         char *base;
5770         u32 p_cluster, num_clusters;
5771         unsigned int ext_flags;
5772         int name_offset, name_len;
5773         struct ocfs2_xattr_value_buf vb;
5774         struct ocfs2_xattr_bucket *bucket = NULL;
5775         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5776         struct ocfs2_post_refcount refcount;
5777         struct ocfs2_post_refcount *p = NULL;
5778         struct buffer_head *ref_root_bh = NULL;
5779
5780         if (!xis->not_found) {
5781                 xe = xis->here;
5782                 name_offset = le16_to_cpu(xe->xe_name_offset);
5783                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5784                 base = xis->base;
5785                 vb.vb_bh = xis->inode_bh;
5786                 vb.vb_access = ocfs2_journal_access_di;
5787         } else {
5788                 int i, block_off = 0;
5789                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5790                 xe = xbs->here;
5791                 name_offset = le16_to_cpu(xe->xe_name_offset);
5792                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5793                 i = xbs->here - xbs->header->xh_entries;
5794
5795                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5796                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5797                                                         bucket_xh(xbs->bucket),
5798                                                         i, &block_off,
5799                                                         &name_offset);
5800                         if (ret) {
5801                                 mlog_errno(ret);
5802                                 goto out;
5803                         }
5804                         base = bucket_block(xbs->bucket, block_off);
5805                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5806                         vb.vb_access = ocfs2_journal_access;
5807
5808                         if (ocfs2_meta_ecc(osb)) {
5809                                 /*create parameters for ocfs2_post_refcount. */
5810                                 bucket = xbs->bucket;
5811                                 refcount.credits = bucket->bu_blocks;
5812                                 refcount.para = bucket;
5813                                 refcount.func =
5814                                         ocfs2_xattr_bucket_post_refcount;
5815                                 p = &refcount;
5816                         }
5817                 } else {
5818                         base = xbs->base;
5819                         vb.vb_bh = xbs->xattr_bh;
5820                         vb.vb_access = ocfs2_journal_access_xb;
5821                 }
5822         }
5823
5824         if (ocfs2_xattr_is_local(xe))
5825                 goto out;
5826
5827         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5828                                 (base + name_offset + name_len);
5829
5830         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5831                                        &num_clusters, &vb.vb_xv->xr_list,
5832                                        &ext_flags);
5833         if (ret) {
5834                 mlog_errno(ret);
5835                 goto out;
5836         }
5837
5838         /*
5839          * We just need to check the 1st extent record, since we always
5840          * CoW the whole xattr. So there shouldn't be a xattr with
5841          * some REFCOUNT extent recs after the 1st one.
5842          */
5843         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5844                 goto out;
5845
5846         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5847                                        1, ref_tree, &ref_root_bh);
5848         if (ret) {
5849                 mlog_errno(ret);
5850                 goto out;
5851         }
5852
5853         /*
5854          * If we are deleting the xattr or the new size will be stored inside,
5855          * cool, leave it there, the xattr truncate process will remove them
5856          * for us(it still needs the refcount tree lock and the meta, credits).
5857          * And the worse case is that every cluster truncate will split the
5858          * refcount tree, and make the original extent become 3. So we will need
5859          * 2 * cluster more extent recs at most.
5860          */
5861         if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5862
5863                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5864                                                          &(*ref_tree)->rf_ci,
5865                                                          ref_root_bh, vb.vb_xv,
5866                                                          meta_add, credits);
5867                 if (ret)
5868                         mlog_errno(ret);
5869                 goto out;
5870         }
5871
5872         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5873                                        *ref_tree, ref_root_bh, 0,
5874                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5875         if (ret)
5876                 mlog_errno(ret);
5877
5878 out:
5879         brelse(ref_root_bh);
5880         return ret;
5881 }
5882
5883 /*
5884  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5885  * The physical clusters will be added to refcount tree.
5886  */
5887 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5888                                 struct ocfs2_xattr_value_root *xv,
5889                                 struct ocfs2_extent_tree *value_et,
5890                                 struct ocfs2_caching_info *ref_ci,
5891                                 struct buffer_head *ref_root_bh,
5892                                 struct ocfs2_cached_dealloc_ctxt *dealloc,
5893                                 struct ocfs2_post_refcount *refcount)
5894 {
5895         int ret = 0;
5896         u32 clusters = le32_to_cpu(xv->xr_clusters);
5897         u32 cpos, p_cluster, num_clusters;
5898         struct ocfs2_extent_list *el = &xv->xr_list;
5899         unsigned int ext_flags;
5900
5901         cpos = 0;
5902         while (cpos < clusters) {
5903                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5904                                                &num_clusters, el, &ext_flags);
5905                 if (ret) {
5906                         mlog_errno(ret);
5907                         break;
5908                 }
5909
5910                 cpos += num_clusters;
5911                 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5912                         continue;
5913
5914                 BUG_ON(!p_cluster);
5915
5916                 ret = ocfs2_add_refcount_flag(inode, value_et,
5917                                               ref_ci, ref_root_bh,
5918                                               cpos - num_clusters,
5919                                               p_cluster, num_clusters,
5920                                               dealloc, refcount);
5921                 if (ret) {
5922                         mlog_errno(ret);
5923                         break;
5924                 }
5925         }
5926
5927         return ret;
5928 }
5929
5930 /*
5931  * Given a normal ocfs2_xattr_header, refcount all the entries which
5932  * have value stored outside.
5933  * Used for xattrs stored in inode and ocfs2_xattr_block.
5934  */
5935 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5936                                 struct ocfs2_xattr_value_buf *vb,
5937                                 struct ocfs2_xattr_header *header,
5938                                 struct ocfs2_caching_info *ref_ci,
5939                                 struct buffer_head *ref_root_bh,
5940                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5941 {
5942
5943         struct ocfs2_xattr_entry *xe;
5944         struct ocfs2_xattr_value_root *xv;
5945         struct ocfs2_extent_tree et;
5946         int i, ret = 0;
5947
5948         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5949                 xe = &header->xh_entries[i];
5950
5951                 if (ocfs2_xattr_is_local(xe))
5952                         continue;
5953
5954                 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5955                         le16_to_cpu(xe->xe_name_offset) +
5956                         OCFS2_XATTR_SIZE(xe->xe_name_len));
5957
5958                 vb->vb_xv = xv;
5959                 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5960
5961                 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5962                                                         ref_ci, ref_root_bh,
5963                                                         dealloc, NULL);
5964                 if (ret) {
5965                         mlog_errno(ret);
5966                         break;
5967                 }
5968         }
5969
5970         return ret;
5971 }
5972
5973 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5974                                 struct buffer_head *fe_bh,
5975                                 struct ocfs2_caching_info *ref_ci,
5976                                 struct buffer_head *ref_root_bh,
5977                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5978 {
5979         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5980         struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5981                                 (fe_bh->b_data + inode->i_sb->s_blocksize -
5982                                 le16_to_cpu(di->i_xattr_inline_size));
5983         struct ocfs2_xattr_value_buf vb = {
5984                 .vb_bh = fe_bh,
5985                 .vb_access = ocfs2_journal_access_di,
5986         };
5987
5988         return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5989                                                   ref_ci, ref_root_bh, dealloc);
5990 }
5991
5992 struct ocfs2_xattr_tree_value_refcount_para {
5993         struct ocfs2_caching_info *ref_ci;
5994         struct buffer_head *ref_root_bh;
5995         struct ocfs2_cached_dealloc_ctxt *dealloc;
5996 };
5997
5998 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5999                                            struct ocfs2_xattr_bucket *bucket,
6000                                            int offset,
6001                                            struct ocfs2_xattr_value_root **xv,
6002                                            struct buffer_head **bh)
6003 {
6004         int ret, block_off, name_offset;
6005         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6006         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6007         void *base;
6008
6009         ret = ocfs2_xattr_bucket_get_name_value(sb,
6010                                                 bucket_xh(bucket),
6011                                                 offset,
6012                                                 &block_off,
6013                                                 &name_offset);
6014         if (ret) {
6015                 mlog_errno(ret);
6016                 goto out;
6017         }
6018
6019         base = bucket_block(bucket, block_off);
6020
6021         *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6022                          OCFS2_XATTR_SIZE(xe->xe_name_len));
6023
6024         if (bh)
6025                 *bh = bucket->bu_bhs[block_off];
6026 out:
6027         return ret;
6028 }
6029
6030 /*
6031  * For a given xattr bucket, refcount all the entries which
6032  * have value stored outside.
6033  */
6034 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6035                                              struct ocfs2_xattr_bucket *bucket,
6036                                              void *para)
6037 {
6038         int i, ret = 0;
6039         struct ocfs2_extent_tree et;
6040         struct ocfs2_xattr_tree_value_refcount_para *ref =
6041                         (struct ocfs2_xattr_tree_value_refcount_para *)para;
6042         struct ocfs2_xattr_header *xh =
6043                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6044         struct ocfs2_xattr_entry *xe;
6045         struct ocfs2_xattr_value_buf vb = {
6046                 .vb_access = ocfs2_journal_access,
6047         };
6048         struct ocfs2_post_refcount refcount = {
6049                 .credits = bucket->bu_blocks,
6050                 .para = bucket,
6051                 .func = ocfs2_xattr_bucket_post_refcount,
6052         };
6053         struct ocfs2_post_refcount *p = NULL;
6054
6055         /* We only need post_refcount if we support metaecc. */
6056         if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6057                 p = &refcount;
6058
6059         trace_ocfs2_xattr_bucket_value_refcount(
6060                                 (unsigned long long)bucket_blkno(bucket),
6061                                 le16_to_cpu(xh->xh_count));
6062         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6063                 xe = &xh->xh_entries[i];
6064
6065                 if (ocfs2_xattr_is_local(xe))
6066                         continue;
6067
6068                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6069                                                       &vb.vb_xv, &vb.vb_bh);
6070                 if (ret) {
6071                         mlog_errno(ret);
6072                         break;
6073                 }
6074
6075                 ocfs2_init_xattr_value_extent_tree(&et,
6076                                                    INODE_CACHE(inode), &vb);
6077
6078                 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6079                                                         &et, ref->ref_ci,
6080                                                         ref->ref_root_bh,
6081                                                         ref->dealloc, p);
6082                 if (ret) {
6083                         mlog_errno(ret);
6084                         break;
6085                 }
6086         }
6087
6088         return ret;
6089
6090 }
6091
6092 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6093                                      struct buffer_head *root_bh,
6094                                      u64 blkno, u32 cpos, u32 len, void *para)
6095 {
6096         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6097                                            ocfs2_xattr_bucket_value_refcount,
6098                                            para);
6099 }
6100
6101 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6102                                 struct buffer_head *blk_bh,
6103                                 struct ocfs2_caching_info *ref_ci,
6104                                 struct buffer_head *ref_root_bh,
6105                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
6106 {
6107         int ret = 0;
6108         struct ocfs2_xattr_block *xb =
6109                                 (struct ocfs2_xattr_block *)blk_bh->b_data;
6110
6111         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6112                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6113                 struct ocfs2_xattr_value_buf vb = {
6114                         .vb_bh = blk_bh,
6115                         .vb_access = ocfs2_journal_access_xb,
6116                 };
6117
6118                 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6119                                                          ref_ci, ref_root_bh,
6120                                                          dealloc);
6121         } else {
6122                 struct ocfs2_xattr_tree_value_refcount_para para = {
6123                         .ref_ci = ref_ci,
6124                         .ref_root_bh = ref_root_bh,
6125                         .dealloc = dealloc,
6126                 };
6127
6128                 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6129                                                 ocfs2_refcount_xattr_tree_rec,
6130                                                 &para);
6131         }
6132
6133         return ret;
6134 }
6135
6136 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6137                                      struct buffer_head *fe_bh,
6138                                      struct ocfs2_caching_info *ref_ci,
6139                                      struct buffer_head *ref_root_bh,
6140                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
6141 {
6142         int ret = 0;
6143         struct ocfs2_inode_info *oi = OCFS2_I(inode);
6144         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6145         struct buffer_head *blk_bh = NULL;
6146
6147         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6148                 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6149                                                          ref_ci, ref_root_bh,
6150                                                          dealloc);
6151                 if (ret) {
6152                         mlog_errno(ret);
6153                         goto out;
6154                 }
6155         }
6156
6157         if (!di->i_xattr_loc)
6158                 goto out;
6159
6160         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6161                                      &blk_bh);
6162         if (ret < 0) {
6163                 mlog_errno(ret);
6164                 goto out;
6165         }
6166
6167         ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6168                                                 ref_root_bh, dealloc);
6169         if (ret)
6170                 mlog_errno(ret);
6171
6172         brelse(blk_bh);
6173 out:
6174
6175         return ret;
6176 }
6177
6178 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6179 /*
6180  * Store the information we need in xattr reflink.
6181  * old_bh and new_bh are inode bh for the old and new inode.
6182  */
6183 struct ocfs2_xattr_reflink {
6184         struct inode *old_inode;
6185         struct inode *new_inode;
6186         struct buffer_head *old_bh;
6187         struct buffer_head *new_bh;
6188         struct ocfs2_caching_info *ref_ci;
6189         struct buffer_head *ref_root_bh;
6190         struct ocfs2_cached_dealloc_ctxt *dealloc;
6191         should_xattr_reflinked *xattr_reflinked;
6192 };
6193
6194 /*
6195  * Given a xattr header and xe offset,
6196  * return the proper xv and the corresponding bh.
6197  * xattr in inode, block and xattr tree have different implementaions.
6198  */
6199 typedef int (get_xattr_value_root)(struct super_block *sb,
6200                                    struct buffer_head *bh,
6201                                    struct ocfs2_xattr_header *xh,
6202                                    int offset,
6203                                    struct ocfs2_xattr_value_root **xv,
6204                                    struct buffer_head **ret_bh,
6205                                    void *para);
6206
6207 /*
6208  * Calculate all the xattr value root metadata stored in this xattr header and
6209  * credits we need if we create them from the scratch.
6210  * We use get_xattr_value_root so that all types of xattr container can use it.
6211  */
6212 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6213                                              struct buffer_head *bh,
6214                                              struct ocfs2_xattr_header *xh,
6215                                              int *metas, int *credits,
6216                                              int *num_recs,
6217                                              get_xattr_value_root *func,
6218                                              void *para)
6219 {
6220         int i, ret = 0;
6221         struct ocfs2_xattr_value_root *xv;
6222         struct ocfs2_xattr_entry *xe;
6223
6224         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6225                 xe = &xh->xh_entries[i];
6226                 if (ocfs2_xattr_is_local(xe))
6227                         continue;
6228
6229                 ret = func(sb, bh, xh, i, &xv, NULL, para);
6230                 if (ret) {
6231                         mlog_errno(ret);
6232                         break;
6233                 }
6234
6235                 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6236                           le16_to_cpu(xv->xr_list.l_next_free_rec);
6237
6238                 *credits += ocfs2_calc_extend_credits(sb,
6239                                                 &def_xv.xv.xr_list);
6240
6241                 /*
6242                  * If the value is a tree with depth > 1, We don't go deep
6243                  * to the extent block, so just calculate a maximum record num.
6244                  */
6245                 if (!xv->xr_list.l_tree_depth)
6246                         *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6247                 else
6248                         *num_recs += ocfs2_clusters_for_bytes(sb,
6249                                                               XATTR_SIZE_MAX);
6250         }
6251
6252         return ret;
6253 }
6254
6255 /* Used by xattr inode and block to return the right xv and buffer_head. */
6256 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6257                                       struct buffer_head *bh,
6258                                       struct ocfs2_xattr_header *xh,
6259                                       int offset,
6260                                       struct ocfs2_xattr_value_root **xv,
6261                                       struct buffer_head **ret_bh,
6262                                       void *para)
6263 {
6264         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6265
6266         *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6267                 le16_to_cpu(xe->xe_name_offset) +
6268                 OCFS2_XATTR_SIZE(xe->xe_name_len));
6269
6270         if (ret_bh)
6271                 *ret_bh = bh;
6272
6273         return 0;
6274 }
6275
6276 /*
6277  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6278  * It is only used for inline xattr and xattr block.
6279  */
6280 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6281                                         struct ocfs2_xattr_header *xh,
6282                                         struct buffer_head *ref_root_bh,
6283                                         int *credits,
6284                                         struct ocfs2_alloc_context **meta_ac)
6285 {
6286         int ret, meta_add = 0, num_recs = 0;
6287         struct ocfs2_refcount_block *rb =
6288                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6289
6290         *credits = 0;
6291
6292         ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6293                                                 &meta_add, credits, &num_recs,
6294                                                 ocfs2_get_xattr_value_root,
6295                                                 NULL);
6296         if (ret) {
6297                 mlog_errno(ret);
6298                 goto out;
6299         }
6300
6301         /*
6302          * We need to add/modify num_recs in refcount tree, so just calculate
6303          * an approximate number we need for refcount tree change.
6304          * Sometimes we need to split the tree, and after split,  half recs
6305          * will be moved to the new block, and a new block can only provide
6306          * half number of recs. So we multiple new blocks by 2.
6307          */
6308         num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6309         meta_add += num_recs;
6310         *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6311         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6312                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6313                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6314         else
6315                 *credits += 1;
6316
6317         ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6318         if (ret)
6319                 mlog_errno(ret);
6320
6321 out:
6322         return ret;
6323 }
6324
6325 /*
6326  * Given a xattr header, reflink all the xattrs in this container.
6327  * It can be used for inode, block and bucket.
6328  *
6329  * NOTE:
6330  * Before we call this function, the caller has memcpy the xattr in
6331  * old_xh to the new_xh.
6332  *
6333  * If args.xattr_reflinked is set, call it to decide whether the xe should
6334  * be reflinked or not. If not, remove it from the new xattr header.
6335  */
6336 static int ocfs2_reflink_xattr_header(handle_t *handle,
6337                                       struct ocfs2_xattr_reflink *args,
6338                                       struct buffer_head *old_bh,
6339                                       struct ocfs2_xattr_header *xh,
6340                                       struct buffer_head *new_bh,
6341                                       struct ocfs2_xattr_header *new_xh,
6342                                       struct ocfs2_xattr_value_buf *vb,
6343                                       struct ocfs2_alloc_context *meta_ac,
6344                                       get_xattr_value_root *func,
6345                                       void *para)
6346 {
6347         int ret = 0, i, j;
6348         struct super_block *sb = args->old_inode->i_sb;
6349         struct buffer_head *value_bh;
6350         struct ocfs2_xattr_entry *xe, *last;
6351         struct ocfs2_xattr_value_root *xv, *new_xv;
6352         struct ocfs2_extent_tree data_et;
6353         u32 clusters, cpos, p_cluster, num_clusters;
6354         unsigned int ext_flags = 0;
6355
6356         trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6357                                          le16_to_cpu(xh->xh_count));
6358
6359         last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6360         for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6361                 xe = &xh->xh_entries[i];
6362
6363                 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6364                         xe = &new_xh->xh_entries[j];
6365
6366                         le16_add_cpu(&new_xh->xh_count, -1);
6367                         if (new_xh->xh_count) {
6368                                 memmove(xe, xe + 1,
6369                                         (void *)last - (void *)xe);
6370                                 memset(last, 0,
6371                                        sizeof(struct ocfs2_xattr_entry));
6372                         }
6373
6374                         /*
6375                          * We don't want j to increase in the next round since
6376                          * it is already moved ahead.
6377                          */
6378                         j--;
6379                         continue;
6380                 }
6381
6382                 if (ocfs2_xattr_is_local(xe))
6383                         continue;
6384
6385                 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6386                 if (ret) {
6387                         mlog_errno(ret);
6388                         break;
6389                 }
6390
6391                 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6392                 if (ret) {
6393                         mlog_errno(ret);
6394                         break;
6395                 }
6396
6397                 /*
6398                  * For the xattr which has l_tree_depth = 0, all the extent
6399                  * recs have already be copied to the new xh with the
6400                  * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6401                  * increase the refount count int the refcount tree.
6402                  *
6403                  * For the xattr which has l_tree_depth > 0, we need
6404                  * to initialize it to the empty default value root,
6405                  * and then insert the extents one by one.
6406                  */
6407                 if (xv->xr_list.l_tree_depth) {
6408                         memcpy(new_xv, &def_xv, OCFS2_XATTR_ROOT_SIZE);
6409                         vb->vb_xv = new_xv;
6410                         vb->vb_bh = value_bh;
6411                         ocfs2_init_xattr_value_extent_tree(&data_et,
6412                                         INODE_CACHE(args->new_inode), vb);
6413                 }
6414
6415                 clusters = le32_to_cpu(xv->xr_clusters);
6416                 cpos = 0;
6417                 while (cpos < clusters) {
6418                         ret = ocfs2_xattr_get_clusters(args->old_inode,
6419                                                        cpos,
6420                                                        &p_cluster,
6421                                                        &num_clusters,
6422                                                        &xv->xr_list,
6423                                                        &ext_flags);
6424                         if (ret) {
6425                                 mlog_errno(ret);
6426                                 goto out;
6427                         }
6428
6429                         BUG_ON(!p_cluster);
6430
6431                         if (xv->xr_list.l_tree_depth) {
6432                                 ret = ocfs2_insert_extent(handle,
6433                                                 &data_et, cpos,
6434                                                 ocfs2_clusters_to_blocks(
6435                                                         args->old_inode->i_sb,
6436                                                         p_cluster),
6437                                                 num_clusters, ext_flags,
6438                                                 meta_ac);
6439                                 if (ret) {
6440                                         mlog_errno(ret);
6441                                         goto out;
6442                                 }
6443                         }
6444
6445                         ret = ocfs2_increase_refcount(handle, args->ref_ci,
6446                                                       args->ref_root_bh,
6447                                                       p_cluster, num_clusters,
6448                                                       meta_ac, args->dealloc);
6449                         if (ret) {
6450                                 mlog_errno(ret);
6451                                 goto out;
6452                         }
6453
6454                         cpos += num_clusters;
6455                 }
6456         }
6457
6458 out:
6459         return ret;
6460 }
6461
6462 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6463 {
6464         int ret = 0, credits = 0;
6465         handle_t *handle;
6466         struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6467         struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6468         int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6469         int header_off = osb->sb->s_blocksize - inline_size;
6470         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6471                                         (args->old_bh->b_data + header_off);
6472         struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6473                                         (args->new_bh->b_data + header_off);
6474         struct ocfs2_alloc_context *meta_ac = NULL;
6475         struct ocfs2_inode_info *new_oi;
6476         struct ocfs2_dinode *new_di;
6477         struct ocfs2_xattr_value_buf vb = {
6478                 .vb_bh = args->new_bh,
6479                 .vb_access = ocfs2_journal_access_di,
6480         };
6481
6482         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6483                                                   &credits, &meta_ac);
6484         if (ret) {
6485                 mlog_errno(ret);
6486                 goto out;
6487         }
6488
6489         handle = ocfs2_start_trans(osb, credits);
6490         if (IS_ERR(handle)) {
6491                 ret = PTR_ERR(handle);
6492                 mlog_errno(ret);
6493                 goto out;
6494         }
6495
6496         ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6497                                       args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6498         if (ret) {
6499                 mlog_errno(ret);
6500                 goto out_commit;
6501         }
6502
6503         memcpy(args->new_bh->b_data + header_off,
6504                args->old_bh->b_data + header_off, inline_size);
6505
6506         new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6507         new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6508
6509         ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6510                                          args->new_bh, new_xh, &vb, meta_ac,
6511                                          ocfs2_get_xattr_value_root, NULL);
6512         if (ret) {
6513                 mlog_errno(ret);
6514                 goto out_commit;
6515         }
6516
6517         new_oi = OCFS2_I(args->new_inode);
6518         /*
6519          * Adjust extent record count to reserve space for extended attribute.
6520          * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6521          */
6522         if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6523             !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6524                 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6525                 le16_add_cpu(&el->l_count, -(inline_size /
6526                                         sizeof(struct ocfs2_extent_rec)));
6527         }
6528         spin_lock(&new_oi->ip_lock);
6529         new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6530         new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6531         spin_unlock(&new_oi->ip_lock);
6532
6533         ocfs2_journal_dirty(handle, args->new_bh);
6534
6535 out_commit:
6536         ocfs2_commit_trans(osb, handle);
6537
6538 out:
6539         if (meta_ac)
6540                 ocfs2_free_alloc_context(meta_ac);
6541         return ret;
6542 }
6543
6544 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6545                                           struct buffer_head *fe_bh,
6546                                           struct buffer_head **ret_bh,
6547                                           int indexed)
6548 {
6549         int ret;
6550         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6551         struct ocfs2_xattr_set_ctxt ctxt;
6552
6553         memset(&ctxt, 0, sizeof(ctxt));
6554         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6555         if (ret < 0) {
6556                 mlog_errno(ret);
6557                 return ret;
6558         }
6559
6560         ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6561         if (IS_ERR(ctxt.handle)) {
6562                 ret = PTR_ERR(ctxt.handle);
6563                 mlog_errno(ret);
6564                 goto out;
6565         }
6566
6567         trace_ocfs2_create_empty_xattr_block(
6568                                 (unsigned long long)fe_bh->b_blocknr, indexed);
6569         ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6570                                        ret_bh);
6571         if (ret)
6572                 mlog_errno(ret);
6573
6574         ocfs2_commit_trans(osb, ctxt.handle);
6575 out:
6576         ocfs2_free_alloc_context(ctxt.meta_ac);
6577         return ret;
6578 }
6579
6580 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6581                                      struct buffer_head *blk_bh,
6582                                      struct buffer_head *new_blk_bh)
6583 {
6584         int ret = 0, credits = 0;
6585         handle_t *handle;
6586         struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6587         struct ocfs2_dinode *new_di;
6588         struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6589         int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6590         struct ocfs2_xattr_block *xb =
6591                         (struct ocfs2_xattr_block *)blk_bh->b_data;
6592         struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6593         struct ocfs2_xattr_block *new_xb =
6594                         (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6595         struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6596         struct ocfs2_alloc_context *meta_ac;
6597         struct ocfs2_xattr_value_buf vb = {
6598                 .vb_bh = new_blk_bh,
6599                 .vb_access = ocfs2_journal_access_xb,
6600         };
6601
6602         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6603                                                   &credits, &meta_ac);
6604         if (ret) {
6605                 mlog_errno(ret);
6606                 return ret;
6607         }
6608
6609         /* One more credits in case we need to add xattr flags in new inode. */
6610         handle = ocfs2_start_trans(osb, credits + 1);
6611         if (IS_ERR(handle)) {
6612                 ret = PTR_ERR(handle);
6613                 mlog_errno(ret);
6614                 goto out;
6615         }
6616
6617         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6618                 ret = ocfs2_journal_access_di(handle,
6619                                               INODE_CACHE(args->new_inode),
6620                                               args->new_bh,
6621                                               OCFS2_JOURNAL_ACCESS_WRITE);
6622                 if (ret) {
6623                         mlog_errno(ret);
6624                         goto out_commit;
6625                 }
6626         }
6627
6628         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6629                                       new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6630         if (ret) {
6631                 mlog_errno(ret);
6632                 goto out_commit;
6633         }
6634
6635         memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6636                osb->sb->s_blocksize - header_off);
6637
6638         ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6639                                          new_blk_bh, new_xh, &vb, meta_ac,
6640                                          ocfs2_get_xattr_value_root, NULL);
6641         if (ret) {
6642                 mlog_errno(ret);
6643                 goto out_commit;
6644         }
6645
6646         ocfs2_journal_dirty(handle, new_blk_bh);
6647
6648         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6649                 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6650                 spin_lock(&new_oi->ip_lock);
6651                 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6652                 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6653                 spin_unlock(&new_oi->ip_lock);
6654
6655                 ocfs2_journal_dirty(handle, args->new_bh);
6656         }
6657
6658 out_commit:
6659         ocfs2_commit_trans(osb, handle);
6660
6661 out:
6662         ocfs2_free_alloc_context(meta_ac);
6663         return ret;
6664 }
6665
6666 struct ocfs2_reflink_xattr_tree_args {
6667         struct ocfs2_xattr_reflink *reflink;
6668         struct buffer_head *old_blk_bh;
6669         struct buffer_head *new_blk_bh;
6670         struct ocfs2_xattr_bucket *old_bucket;
6671         struct ocfs2_xattr_bucket *new_bucket;
6672 };
6673
6674 /*
6675  * NOTE:
6676  * We have to handle the case that both old bucket and new bucket
6677  * will call this function to get the right ret_bh.
6678  * So The caller must give us the right bh.
6679  */
6680 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6681                                         struct buffer_head *bh,
6682                                         struct ocfs2_xattr_header *xh,
6683                                         int offset,
6684                                         struct ocfs2_xattr_value_root **xv,
6685                                         struct buffer_head **ret_bh,
6686                                         void *para)
6687 {
6688         struct ocfs2_reflink_xattr_tree_args *args =
6689                         (struct ocfs2_reflink_xattr_tree_args *)para;
6690         struct ocfs2_xattr_bucket *bucket;
6691
6692         if (bh == args->old_bucket->bu_bhs[0])
6693                 bucket = args->old_bucket;
6694         else
6695                 bucket = args->new_bucket;
6696
6697         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6698                                                xv, ret_bh);
6699 }
6700
6701 struct ocfs2_value_tree_metas {
6702         int num_metas;
6703         int credits;
6704         int num_recs;
6705 };
6706
6707 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6708                                         struct buffer_head *bh,
6709                                         struct ocfs2_xattr_header *xh,
6710                                         int offset,
6711                                         struct ocfs2_xattr_value_root **xv,
6712                                         struct buffer_head **ret_bh,
6713                                         void *para)
6714 {
6715         struct ocfs2_xattr_bucket *bucket =
6716                                 (struct ocfs2_xattr_bucket *)para;
6717
6718         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6719                                                xv, ret_bh);
6720 }
6721
6722 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6723                                       struct ocfs2_xattr_bucket *bucket,
6724                                       void *para)
6725 {
6726         struct ocfs2_value_tree_metas *metas =
6727                         (struct ocfs2_value_tree_metas *)para;
6728         struct ocfs2_xattr_header *xh =
6729                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6730
6731         /* Add the credits for this bucket first. */
6732         metas->credits += bucket->bu_blocks;
6733         return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6734                                         xh, &metas->num_metas,
6735                                         &metas->credits, &metas->num_recs,
6736                                         ocfs2_value_tree_metas_in_bucket,
6737                                         bucket);
6738 }
6739
6740 /*
6741  * Given a xattr extent rec starting from blkno and having len clusters,
6742  * iterate all the buckets calculate how much metadata we need for reflinking
6743  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6744  */
6745 static int ocfs2_lock_reflink_xattr_rec_allocators(
6746                                 struct ocfs2_reflink_xattr_tree_args *args,
6747                                 struct ocfs2_extent_tree *xt_et,
6748                                 u64 blkno, u32 len, int *credits,
6749                                 struct ocfs2_alloc_context **meta_ac,
6750                                 struct ocfs2_alloc_context **data_ac)
6751 {
6752         int ret, num_free_extents;
6753         struct ocfs2_value_tree_metas metas;
6754         struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6755         struct ocfs2_refcount_block *rb;
6756
6757         memset(&metas, 0, sizeof(metas));
6758
6759         ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6760                                           ocfs2_calc_value_tree_metas, &metas);
6761         if (ret) {
6762                 mlog_errno(ret);
6763                 goto out;
6764         }
6765
6766         *credits = metas.credits;
6767
6768         /*
6769          * Calculate we need for refcount tree change.
6770          *
6771          * We need to add/modify num_recs in refcount tree, so just calculate
6772          * an approximate number we need for refcount tree change.
6773          * Sometimes we need to split the tree, and after split,  half recs
6774          * will be moved to the new block, and a new block can only provide
6775          * half number of recs. So we multiple new blocks by 2.
6776          * In the end, we have to add credits for modifying the already
6777          * existed refcount block.
6778          */
6779         rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6780         metas.num_recs =
6781                 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6782                  ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6783         metas.num_metas += metas.num_recs;
6784         *credits += metas.num_recs +
6785                     metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6786         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6787                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6788                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6789         else
6790                 *credits += 1;
6791
6792         /* count in the xattr tree change. */
6793         num_free_extents = ocfs2_num_free_extents(xt_et);
6794         if (num_free_extents < 0) {
6795                 ret = num_free_extents;
6796                 mlog_errno(ret);
6797                 goto out;
6798         }
6799
6800         if (num_free_extents < len)
6801                 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6802
6803         *credits += ocfs2_calc_extend_credits(osb->sb,
6804                                               xt_et->et_root_el);
6805
6806         if (metas.num_metas) {
6807                 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6808                                                         meta_ac);
6809                 if (ret) {
6810                         mlog_errno(ret);
6811                         goto out;
6812                 }
6813         }
6814
6815         if (len) {
6816                 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6817                 if (ret)
6818                         mlog_errno(ret);
6819         }
6820 out:
6821         if (ret) {
6822                 if (*meta_ac) {
6823                         ocfs2_free_alloc_context(*meta_ac);
6824                         *meta_ac = NULL;
6825                 }
6826         }
6827
6828         return ret;
6829 }
6830
6831 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6832                                 u64 blkno, u64 new_blkno, u32 clusters,
6833                                 u32 *cpos, int num_buckets,
6834                                 struct ocfs2_alloc_context *meta_ac,
6835                                 struct ocfs2_alloc_context *data_ac,
6836                                 struct ocfs2_reflink_xattr_tree_args *args)
6837 {
6838         int i, j, ret = 0;
6839         struct super_block *sb = args->reflink->old_inode->i_sb;
6840         int bpb = args->old_bucket->bu_blocks;
6841         struct ocfs2_xattr_value_buf vb = {
6842                 .vb_access = ocfs2_journal_access,
6843         };
6844
6845         for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6846                 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6847                 if (ret) {
6848                         mlog_errno(ret);
6849                         break;
6850                 }
6851
6852                 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6853                 if (ret) {
6854                         mlog_errno(ret);
6855                         break;
6856                 }
6857
6858                 ret = ocfs2_xattr_bucket_journal_access(handle,
6859                                                 args->new_bucket,
6860                                                 OCFS2_JOURNAL_ACCESS_CREATE);
6861                 if (ret) {
6862                         mlog_errno(ret);
6863                         break;
6864                 }
6865
6866                 for (j = 0; j < bpb; j++)
6867                         memcpy(bucket_block(args->new_bucket, j),
6868                                bucket_block(args->old_bucket, j),
6869                                sb->s_blocksize);
6870
6871                 /*
6872                  * Record the start cpos so that we can use it to initialize
6873                  * our xattr tree we also set the xh_num_bucket for the new
6874                  * bucket.
6875                  */
6876                 if (i == 0) {
6877                         *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6878                                             xh_entries[0].xe_name_hash);
6879                         bucket_xh(args->new_bucket)->xh_num_buckets =
6880                                 cpu_to_le16(num_buckets);
6881                 }
6882
6883                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6884
6885                 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6886                                         args->old_bucket->bu_bhs[0],
6887                                         bucket_xh(args->old_bucket),
6888                                         args->new_bucket->bu_bhs[0],
6889                                         bucket_xh(args->new_bucket),
6890                                         &vb, meta_ac,
6891                                         ocfs2_get_reflink_xattr_value_root,
6892                                         args);
6893                 if (ret) {
6894                         mlog_errno(ret);
6895                         break;
6896                 }
6897
6898                 /*
6899                  * Re-access and dirty the bucket to calculate metaecc.
6900                  * Because we may extend the transaction in reflink_xattr_header
6901                  * which will let the already accessed block gone.
6902                  */
6903                 ret = ocfs2_xattr_bucket_journal_access(handle,
6904                                                 args->new_bucket,
6905                                                 OCFS2_JOURNAL_ACCESS_WRITE);
6906                 if (ret) {
6907                         mlog_errno(ret);
6908                         break;
6909                 }
6910
6911                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6912
6913                 ocfs2_xattr_bucket_relse(args->old_bucket);
6914                 ocfs2_xattr_bucket_relse(args->new_bucket);
6915         }
6916
6917         ocfs2_xattr_bucket_relse(args->old_bucket);
6918         ocfs2_xattr_bucket_relse(args->new_bucket);
6919         return ret;
6920 }
6921
6922 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6923                                 struct inode *inode,
6924                                 struct ocfs2_reflink_xattr_tree_args *args,
6925                                 struct ocfs2_extent_tree *et,
6926                                 struct ocfs2_alloc_context *meta_ac,
6927                                 struct ocfs2_alloc_context *data_ac,
6928                                 u64 blkno, u32 cpos, u32 len)
6929 {
6930         int ret, first_inserted = 0;
6931         u32 p_cluster, num_clusters, reflink_cpos = 0;
6932         u64 new_blkno;
6933         unsigned int num_buckets, reflink_buckets;
6934         unsigned int bpc =
6935                 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6936
6937         ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6938         if (ret) {
6939                 mlog_errno(ret);
6940                 goto out;
6941         }
6942         num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6943         ocfs2_xattr_bucket_relse(args->old_bucket);
6944
6945         while (len && num_buckets) {
6946                 ret = ocfs2_claim_clusters(handle, data_ac,
6947                                            1, &p_cluster, &num_clusters);
6948                 if (ret) {
6949                         mlog_errno(ret);
6950                         goto out;
6951                 }
6952
6953                 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6954                 reflink_buckets = min(num_buckets, bpc * num_clusters);
6955
6956                 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6957                                                  new_blkno, num_clusters,
6958                                                  &reflink_cpos, reflink_buckets,
6959                                                  meta_ac, data_ac, args);
6960                 if (ret) {
6961                         mlog_errno(ret);
6962                         goto out;
6963                 }
6964
6965                 /*
6966                  * For the 1st allocated cluster, we make it use the same cpos
6967                  * so that the xattr tree looks the same as the original one
6968                  * in the most case.
6969                  */
6970                 if (!first_inserted) {
6971                         reflink_cpos = cpos;
6972                         first_inserted = 1;
6973                 }
6974                 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6975                                           num_clusters, 0, meta_ac);
6976                 if (ret)
6977                         mlog_errno(ret);
6978
6979                 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6980                                                   num_clusters, reflink_cpos);
6981
6982                 len -= num_clusters;
6983                 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6984                 num_buckets -= reflink_buckets;
6985         }
6986 out:
6987         return ret;
6988 }
6989
6990 /*
6991  * Create the same xattr extent record in the new inode's xattr tree.
6992  */
6993 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6994                                    struct buffer_head *root_bh,
6995                                    u64 blkno,
6996                                    u32 cpos,
6997                                    u32 len,
6998                                    void *para)
6999 {
7000         int ret, credits = 0;
7001         handle_t *handle;
7002         struct ocfs2_reflink_xattr_tree_args *args =
7003                         (struct ocfs2_reflink_xattr_tree_args *)para;
7004         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7005         struct ocfs2_alloc_context *meta_ac = NULL;
7006         struct ocfs2_alloc_context *data_ac = NULL;
7007         struct ocfs2_extent_tree et;
7008
7009         trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
7010
7011         ocfs2_init_xattr_tree_extent_tree(&et,
7012                                           INODE_CACHE(args->reflink->new_inode),
7013                                           args->new_blk_bh);
7014
7015         ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7016                                                       len, &credits,
7017                                                       &meta_ac, &data_ac);
7018         if (ret) {
7019                 mlog_errno(ret);
7020                 goto out;
7021         }
7022
7023         handle = ocfs2_start_trans(osb, credits);
7024         if (IS_ERR(handle)) {
7025                 ret = PTR_ERR(handle);
7026                 mlog_errno(ret);
7027                 goto out;
7028         }
7029
7030         ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7031                                           meta_ac, data_ac,
7032                                           blkno, cpos, len);
7033         if (ret)
7034                 mlog_errno(ret);
7035
7036         ocfs2_commit_trans(osb, handle);
7037
7038 out:
7039         if (meta_ac)
7040                 ocfs2_free_alloc_context(meta_ac);
7041         if (data_ac)
7042                 ocfs2_free_alloc_context(data_ac);
7043         return ret;
7044 }
7045
7046 /*
7047  * Create reflinked xattr buckets.
7048  * We will add bucket one by one, and refcount all the xattrs in the bucket
7049  * if they are stored outside.
7050  */
7051 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7052                                     struct buffer_head *blk_bh,
7053                                     struct buffer_head *new_blk_bh)
7054 {
7055         int ret;
7056         struct ocfs2_reflink_xattr_tree_args para;
7057
7058         memset(&para, 0, sizeof(para));
7059         para.reflink = args;
7060         para.old_blk_bh = blk_bh;
7061         para.new_blk_bh = new_blk_bh;
7062
7063         para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7064         if (!para.old_bucket) {
7065                 mlog_errno(-ENOMEM);
7066                 return -ENOMEM;
7067         }
7068
7069         para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7070         if (!para.new_bucket) {
7071                 ret = -ENOMEM;
7072                 mlog_errno(ret);
7073                 goto out;
7074         }
7075
7076         ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7077                                               ocfs2_reflink_xattr_rec,
7078                                               &para);
7079         if (ret)
7080                 mlog_errno(ret);
7081
7082 out:
7083         ocfs2_xattr_bucket_free(para.old_bucket);
7084         ocfs2_xattr_bucket_free(para.new_bucket);
7085         return ret;
7086 }
7087
7088 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7089                                         struct buffer_head *blk_bh)
7090 {
7091         int ret, indexed = 0;
7092         struct buffer_head *new_blk_bh = NULL;
7093         struct ocfs2_xattr_block *xb =
7094                         (struct ocfs2_xattr_block *)blk_bh->b_data;
7095
7096
7097         if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7098                 indexed = 1;
7099
7100         ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7101                                              &new_blk_bh, indexed);
7102         if (ret) {
7103                 mlog_errno(ret);
7104                 goto out;
7105         }
7106
7107         if (!indexed)
7108                 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7109         else
7110                 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7111         if (ret)
7112                 mlog_errno(ret);
7113
7114 out:
7115         brelse(new_blk_bh);
7116         return ret;
7117 }
7118
7119 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7120 {
7121         int type = ocfs2_xattr_get_type(xe);
7122
7123         return type != OCFS2_XATTR_INDEX_SECURITY &&
7124                type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7125                type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7126 }
7127
7128 int ocfs2_reflink_xattrs(struct inode *old_inode,
7129                          struct buffer_head *old_bh,
7130                          struct inode *new_inode,
7131                          struct buffer_head *new_bh,
7132                          bool preserve_security)
7133 {
7134         int ret;
7135         struct ocfs2_xattr_reflink args;
7136         struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7137         struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7138         struct buffer_head *blk_bh = NULL;
7139         struct ocfs2_cached_dealloc_ctxt dealloc;
7140         struct ocfs2_refcount_tree *ref_tree;
7141         struct buffer_head *ref_root_bh = NULL;
7142
7143         ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7144                                        le64_to_cpu(di->i_refcount_loc),
7145                                        1, &ref_tree, &ref_root_bh);
7146         if (ret) {
7147                 mlog_errno(ret);
7148                 goto out;
7149         }
7150
7151         ocfs2_init_dealloc_ctxt(&dealloc);
7152
7153         args.old_inode = old_inode;
7154         args.new_inode = new_inode;
7155         args.old_bh = old_bh;
7156         args.new_bh = new_bh;
7157         args.ref_ci = &ref_tree->rf_ci;
7158         args.ref_root_bh = ref_root_bh;
7159         args.dealloc = &dealloc;
7160         if (preserve_security)
7161                 args.xattr_reflinked = NULL;
7162         else
7163                 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7164
7165         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7166                 ret = ocfs2_reflink_xattr_inline(&args);
7167                 if (ret) {
7168                         mlog_errno(ret);
7169                         goto out_unlock;
7170                 }
7171         }
7172
7173         if (!di->i_xattr_loc)
7174                 goto out_unlock;
7175
7176         ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7177                                      &blk_bh);
7178         if (ret < 0) {
7179                 mlog_errno(ret);
7180                 goto out_unlock;
7181         }
7182
7183         ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7184         if (ret)
7185                 mlog_errno(ret);
7186
7187         brelse(blk_bh);
7188
7189 out_unlock:
7190         ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7191                                    ref_tree, 1);
7192         brelse(ref_root_bh);
7193
7194         if (ocfs2_dealloc_has_cluster(&dealloc)) {
7195                 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7196                 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7197         }
7198
7199 out:
7200         return ret;
7201 }
7202
7203 /*
7204  * Initialize security and acl for a already created inode.
7205  * Used for reflink a non-preserve-security file.
7206  *
7207  * It uses common api like ocfs2_xattr_set, so the caller
7208  * must not hold any lock expect i_mutex.
7209  */
7210 int ocfs2_init_security_and_acl(struct inode *dir,
7211                                 struct inode *inode,
7212                                 const struct qstr *qstr)
7213 {
7214         int ret = 0;
7215         struct buffer_head *dir_bh = NULL;
7216
7217         ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7218         if (ret) {
7219                 mlog_errno(ret);
7220                 goto leave;
7221         }
7222
7223         ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7224         if (ret) {
7225                 mlog_errno(ret);
7226                 goto leave;
7227         }
7228         ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7229         if (ret)
7230                 mlog_errno(ret);
7231
7232         ocfs2_inode_unlock(dir, 0);
7233         brelse(dir_bh);
7234 leave:
7235         return ret;
7236 }
7237
7238 /*
7239  * 'security' attributes support
7240  */
7241 static int ocfs2_xattr_security_get(const struct xattr_handler *handler,
7242                                     struct dentry *unused, struct inode *inode,
7243                                     const char *name, void *buffer, size_t size)
7244 {
7245         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY,
7246                                name, buffer, size);
7247 }
7248
7249 static int ocfs2_xattr_security_set(const struct xattr_handler *handler,
7250                                     struct user_namespace *mnt_userns,
7251                                     struct dentry *unused, struct inode *inode,
7252                                     const char *name, const void *value,
7253                                     size_t size, int flags)
7254 {
7255         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7256                                name, value, size, flags);
7257 }
7258
7259 static 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         .get    = ocfs2_xattr_security_get,
7308         .set    = ocfs2_xattr_security_set,
7309 };
7310
7311 /*
7312  * 'trusted' attributes support
7313  */
7314 static int ocfs2_xattr_trusted_get(const struct xattr_handler *handler,
7315                                    struct dentry *unused, struct inode *inode,
7316                                    const char *name, void *buffer, size_t size)
7317 {
7318         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED,
7319                                name, buffer, size);
7320 }
7321
7322 static int ocfs2_xattr_trusted_set(const struct xattr_handler *handler,
7323                                    struct user_namespace *mnt_userns,
7324                                    struct dentry *unused, struct inode *inode,
7325                                    const char *name, const void *value,
7326                                    size_t size, int flags)
7327 {
7328         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED,
7329                                name, value, size, flags);
7330 }
7331
7332 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7333         .prefix = XATTR_TRUSTED_PREFIX,
7334         .get    = ocfs2_xattr_trusted_get,
7335         .set    = ocfs2_xattr_trusted_set,
7336 };
7337
7338 /*
7339  * 'user' attributes support
7340  */
7341 static int ocfs2_xattr_user_get(const struct xattr_handler *handler,
7342                                 struct dentry *unused, struct inode *inode,
7343                                 const char *name, void *buffer, size_t size)
7344 {
7345         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7346
7347         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7348                 return -EOPNOTSUPP;
7349         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
7350                                buffer, size);
7351 }
7352
7353 static int ocfs2_xattr_user_set(const struct xattr_handler *handler,
7354                                 struct user_namespace *mnt_userns,
7355                                 struct dentry *unused, struct inode *inode,
7356                                 const char *name, const void *value,
7357                                 size_t size, int flags)
7358 {
7359         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7360
7361         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7362                 return -EOPNOTSUPP;
7363
7364         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER,
7365                                name, value, size, flags);
7366 }
7367
7368 const struct xattr_handler ocfs2_xattr_user_handler = {
7369         .prefix = XATTR_USER_PREFIX,
7370         .get    = ocfs2_xattr_user_get,
7371         .set    = ocfs2_xattr_user_set,
7372 };