fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  11  *
  12  * This program is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU General Public
  14  * License version 2 as published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  */
  21
  22 #include <linux/capability.h>
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/uio.h>
  29 #include <linux/sched.h>
  30 #include <linux/splice.h>
  31 #include <linux/mount.h>
  32 #include <linux/writeback.h>
  33 #include <linux/falloc.h>
  34 #include <linux/sort.h>
  35 #include <linux/init.h>
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/security.h>
  39
  40 #include <cluster/masklog.h>
  41
  42 #include "ocfs2.h"
  43 #include "alloc.h"
  44 #include "blockcheck.h"
  45 #include "dlmglue.h"
  46 #include "file.h"
  47 #include "symlink.h"
  48 #include "sysfile.h"
  49 #include "inode.h"
  50 #include "journal.h"
  51 #include "ocfs2_fs.h"
  52 #include "suballoc.h"
  53 #include "uptodate.h"
  54 #include "buffer_head_io.h"
  55 #include "super.h"
  56 #include "xattr.h"
  57 #include "refcounttree.h"
  58 #include "acl.h"
  59 #include "ocfs2_trace.h"
  60
  61 struct ocfs2_xattr_def_value_root {
  62         struct ocfs2_xattr_value_root   xv;
  63         struct ocfs2_extent_rec         er;
  64 };
  65
  66 struct ocfs2_xattr_bucket {
  67         /* The inode these xattrs are associated with */
  68         struct inode *bu_inode;
  69
  70         /* The actual buffers that make up the bucket */
  71         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  72
  73         /* How many blocks make up one bucket for this filesystem */
  74         int bu_blocks;
  75 };
  76
  77 struct ocfs2_xattr_set_ctxt {
  78         handle_t *handle;
  79         struct ocfs2_alloc_context *meta_ac;
  80         struct ocfs2_alloc_context *data_ac;
  81         struct ocfs2_cached_dealloc_ctxt dealloc;
  82         int set_abort;
  83 };
  84
  85 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  86 #define OCFS2_XATTR_INLINE_SIZE 80
  87 #define OCFS2_XATTR_HEADER_GAP  4
  88 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  89                                          - sizeof(struct ocfs2_xattr_header) \
  90                                          - OCFS2_XATTR_HEADER_GAP)
  91 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  92                                          - sizeof(struct ocfs2_xattr_block) \
  93                                          - sizeof(struct ocfs2_xattr_header) \
  94                                          - OCFS2_XATTR_HEADER_GAP)
  95
  96 static struct ocfs2_xattr_def_value_root def_xv = {
  97         .xv.xr_list.l_count = cpu_to_le16(1),
  98 };
  99
 100 const struct xattr_handler *ocfs2_xattr_handlers[] = {
 101         &ocfs2_xattr_user_handler,
 102         &ocfs2_xattr_acl_access_handler,
 103         &ocfs2_xattr_acl_default_handler,
 104         &ocfs2_xattr_trusted_handler,
 105         &ocfs2_xattr_security_handler,
 106         NULL
 107 };
 108
 109 static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 110         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 111         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 112                                         = &ocfs2_xattr_acl_access_handler,
 113         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 114                                         = &ocfs2_xattr_acl_default_handler,
 115         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 116         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 117 };
 118
 119 struct ocfs2_xattr_info {
 120         int             xi_name_index;
 121         const char      *xi_name;
 122         int             xi_name_len;
 123         const void      *xi_value;
 124         size_t          xi_value_len;
 125 };
 126
 127 struct ocfs2_xattr_search {
 128         struct buffer_head *inode_bh;
 129         /*
 130          * xattr_bh point to the block buffer head which has extended attribute
 131          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 132          */
 133         struct buffer_head *xattr_bh;
 134         struct ocfs2_xattr_header *header;
 135         struct ocfs2_xattr_bucket *bucket;
 136         void *base;
 137         void *end;
 138         struct ocfs2_xattr_entry *here;
 139         int not_found;
 140 };
 141
 142 /* Operations on struct ocfs2_xa_entry */
 143 struct ocfs2_xa_loc;
 144 struct ocfs2_xa_loc_operations {
 145         /*
 146          * Journal functions
 147          */
 148         int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
 149                                   int type);
 150         void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
 151
 152         /*
 153          * Return a pointer to the appropriate buffer in loc->xl_storage
 154          * at the given offset from loc->xl_header.
 155          */
 156         void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
 157
 158         /* Can we reuse the existing entry for the new value? */
 159         int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
 160                              struct ocfs2_xattr_info *xi);
 161
 162         /* How much space is needed for the new value? */
 163         int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
 164                                struct ocfs2_xattr_info *xi);
 165
 166         /*
 167          * Return the offset of the first name+value pair.  This is
 168          * the start of our downward-filling free space.
 169          */
 170         int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
 171
 172         /*
 173          * Remove the name+value at this location.  Do whatever is
 174          * appropriate with the remaining name+value pairs.
 175          */
 176         void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
 177
 178         /* Fill xl_entry with a new entry */
 179         void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
 180
 181         /* Add name+value storage to an entry */
 182         void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
 183
 184         /*
 185          * Initialize the value buf's access and bh fields for this entry.
 186          * ocfs2_xa_fill_value_buf() will handle the xv pointer.
 187          */
 188         void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
 189                                    struct ocfs2_xattr_value_buf *vb);
 190 };
 191
 192 /*
 193  * Describes an xattr entry location.  This is a memory structure
 194  * tracking the on-disk structure.
 195  */
 196 struct ocfs2_xa_loc {
 197         /* This xattr belongs to this inode */
 198         struct inode *xl_inode;
 199
 200         /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
 201         struct ocfs2_xattr_header *xl_header;
 202
 203         /* Bytes from xl_header to the end of the storage */
 204         int xl_size;
 205
 206         /*
 207          * The ocfs2_xattr_entry this location describes.  If this is
 208          * NULL, this location describes the on-disk structure where it
 209          * would have been.
 210          */
 211         struct ocfs2_xattr_entry *xl_entry;
 212
 213         /*
 214          * Internal housekeeping
 215          */
 216
 217         /* Buffer(s) containing this entry */
 218         void *xl_storage;
 219
 220         /* Operations on the storage backing this location */
 221         const struct ocfs2_xa_loc_operations *xl_ops;
 222 };
 223
 224 /*
 225  * Convenience functions to calculate how much space is needed for a
 226  * given name+value pair
 227  */
 228 static int namevalue_size(int name_len, uint64_t value_len)
 229 {
 230         if (value_len > OCFS2_XATTR_INLINE_SIZE)
 231                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 232         else
 233                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 234 }
 235
 236 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
 237 {
 238         return namevalue_size(xi->xi_name_len, xi->xi_value_len);
 239 }
 240
 241 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
 242 {
 243         u64 value_len = le64_to_cpu(xe->xe_value_size);
 244
 245         BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
 246                ocfs2_xattr_is_local(xe));
 247         return namevalue_size(xe->xe_name_len, value_len);
 248 }
 249
 250
 251 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 252                                              struct ocfs2_xattr_header *xh,
 253                                              int index,
 254                                              int *block_off,
 255                                              int *new_offset);
 256
 257 static int ocfs2_xattr_block_find(struct inode *inode,
 258                                   int name_index,
 259                                   const char *name,
 260                                   struct ocfs2_xattr_search *xs);
 261 static int ocfs2_xattr_index_block_find(struct inode *inode,
 262                                         struct buffer_head *root_bh,
 263                                         int name_index,
 264                                         const char *name,
 265                                         struct ocfs2_xattr_search *xs);
 266
 267 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 268                                         struct buffer_head *blk_bh,
 269                                         char *buffer,
 270                                         size_t buffer_size);
 271
 272 static int ocfs2_xattr_create_index_block(struct inode *inode,
 273                                           struct ocfs2_xattr_search *xs,
 274                                           struct ocfs2_xattr_set_ctxt *ctxt);
 275
 276 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 277                                              struct ocfs2_xattr_info *xi,
 278                                              struct ocfs2_xattr_search *xs,
 279                                              struct ocfs2_xattr_set_ctxt *ctxt);
 280
 281 typedef int (xattr_tree_rec_func)(struct inode *inode,
 282                                   struct buffer_head *root_bh,
 283                                   u64 blkno, u32 cpos, u32 len, void *para);
 284 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
 285                                            struct buffer_head *root_bh,
 286                                            xattr_tree_rec_func *rec_func,
 287                                            void *para);
 288 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
 289                                         struct ocfs2_xattr_bucket *bucket,
 290                                         void *para);
 291 static int ocfs2_rm_xattr_cluster(struct inode *inode,
 292                                   struct buffer_head *root_bh,
 293                                   u64 blkno,
 294                                   u32 cpos,
 295                                   u32 len,
 296                                   void *para);
 297
 298 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 299                                   u64 src_blk, u64 last_blk, u64 to_blk,
 300                                   unsigned int start_bucket,
 301                                   u32 *first_hash);
 302 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
 303                                         struct ocfs2_dinode *di,
 304                                         struct ocfs2_xattr_info *xi,
 305                                         struct ocfs2_xattr_search *xis,
 306                                         struct ocfs2_xattr_search *xbs,
 307                                         struct ocfs2_refcount_tree **ref_tree,
 308                                         int *meta_need,
 309                                         int *credits);
 310 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
 311                                            struct ocfs2_xattr_bucket *bucket,
 312                                            int offset,
 313                                            struct ocfs2_xattr_value_root **xv,
 314                                            struct buffer_head **bh);
 315
 316 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 317 {
 318         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 319 }
 320
 321 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 322 {
 323         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 324 }
 325
 326 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 327 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 328 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 329
 330 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 331 {
 332         struct ocfs2_xattr_bucket *bucket;
 333         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 334
 335         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 336
 337         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 338         if (bucket) {
 339                 bucket->bu_inode = inode;
 340                 bucket->bu_blocks = blks;
 341         }
 342
 343         return bucket;
 344 }
 345
 346 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 347 {
 348         int i;
 349
 350         for (i = 0; i < bucket->bu_blocks; i++) {
 351                 brelse(bucket->bu_bhs[i]);
 352                 bucket->bu_bhs[i] = NULL;
 353         }
 354 }
 355
 356 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 357 {
 358         if (bucket) {
 359                 ocfs2_xattr_bucket_relse(bucket);
 360                 bucket->bu_inode = NULL;
 361                 kfree(bucket);
 362         }
 363 }
 364
 365 /*
 366  * A bucket that has never been written to disk doesn't need to be
 367  * read.  We just need the buffer_heads.  Don't call this for
 368  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 369  * them fully.
 370  */
 371 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 372                                    u64 xb_blkno)
 373 {
 374         int i, rc = 0;
 375
 376         for (i = 0; i < bucket->bu_blocks; i++) {
 377                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 378                                               xb_blkno + i);
 379                 if (!bucket->bu_bhs[i]) {
 380                         rc = -ENOMEM;
 381                         mlog_errno(rc);
 382                         break;
 383                 }
 384
 385                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 386                                            bucket->bu_bhs[i]))
 387                         ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 388                                                       bucket->bu_bhs[i]);
 389         }
 390
 391         if (rc)
 392                 ocfs2_xattr_bucket_relse(bucket);
 393         return rc;
 394 }
 395
 396 /* Read the xattr bucket at xb_blkno */
 397 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 398                                    u64 xb_blkno)
 399 {
 400         int rc;
 401
 402         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 403                                bucket->bu_blocks, bucket->bu_bhs, 0,
 404                                NULL);
 405         if (!rc) {
 406                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 407                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 408                                                  bucket->bu_bhs,
 409                                                  bucket->bu_blocks,
 410                                                  &bucket_xh(bucket)->xh_check);
 411                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 412                 if (rc)
 413                         mlog_errno(rc);
 414         }
 415
 416         if (rc)
 417                 ocfs2_xattr_bucket_relse(bucket);
 418         return rc;
 419 }
 420
 421 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 422                                              struct ocfs2_xattr_bucket *bucket,
 423                                              int type)
 424 {
 425         int i, rc = 0;
 426
 427         for (i = 0; i < bucket->bu_blocks; i++) {
 428                 rc = ocfs2_journal_access(handle,
 429                                           INODE_CACHE(bucket->bu_inode),
 430                                           bucket->bu_bhs[i], type);
 431                 if (rc) {
 432                         mlog_errno(rc);
 433                         break;
 434                 }
 435         }
 436
 437         return rc;
 438 }
 439
 440 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 441                                              struct ocfs2_xattr_bucket *bucket)
 442 {
 443         int i;
 444
 445         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 446         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 447                                    bucket->bu_bhs, bucket->bu_blocks,
 448                                    &bucket_xh(bucket)->xh_check);
 449         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 450
 451         for (i = 0; i < bucket->bu_blocks; i++)
 452                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 453 }
 454
 455 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 456                                          struct ocfs2_xattr_bucket *src)
 457 {
 458         int i;
 459         int blocksize = src->bu_inode->i_sb->s_blocksize;
 460
 461         BUG_ON(dest->bu_blocks != src->bu_blocks);
 462         BUG_ON(dest->bu_inode != src->bu_inode);
 463
 464         for (i = 0; i < src->bu_blocks; i++) {
 465                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 466                        blocksize);
 467         }
 468 }
 469
 470 static int ocfs2_validate_xattr_block(struct super_block *sb,
 471                                       struct buffer_head *bh)
 472 {
 473         int rc;
 474         struct ocfs2_xattr_block *xb =
 475                 (struct ocfs2_xattr_block *)bh->b_data;
 476
 477         trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
 478
 479         BUG_ON(!buffer_uptodate(bh));
 480
 481         /*
 482          * If the ecc fails, we return the error but otherwise
 483          * leave the filesystem running.  We know any error is
 484          * local to this block.
 485          */
 486         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 487         if (rc)
 488                 return rc;
 489
 490         /*
 491          * Errors after here are fatal
 492          */
 493
 494         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 495                 ocfs2_error(sb,
 496                             "Extended attribute block #%llu has bad "
 497                             "signature %.*s",
 498                             (unsigned long long)bh->b_blocknr, 7,
 499                             xb->xb_signature);
 500                 return -EINVAL;
 501         }
 502
 503         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 504                 ocfs2_error(sb,
 505                             "Extended attribute block #%llu has an "
 506                             "invalid xb_blkno of %llu",
 507                             (unsigned long long)bh->b_blocknr,
 508                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 509                 return -EINVAL;
 510         }
 511
 512         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 513                 ocfs2_error(sb,
 514                             "Extended attribute block #%llu has an invalid "
 515                             "xb_fs_generation of #%u",
 516                             (unsigned long long)bh->b_blocknr,
 517                             le32_to_cpu(xb->xb_fs_generation));
 518                 return -EINVAL;
 519         }
 520
 521         return 0;
 522 }
 523
 524 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 525                                   struct buffer_head **bh)
 526 {
 527         int rc;
 528         struct buffer_head *tmp = *bh;
 529
 530         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 531                               ocfs2_validate_xattr_block);
 532
 533         /* If ocfs2_read_block() got us a new bh, pass it up. */
 534         if (!rc && !*bh)
 535                 *bh = tmp;
 536
 537         return rc;
 538 }
 539
 540 static inline const char *ocfs2_xattr_prefix(int name_index)
 541 {
 542         const struct xattr_handler *handler = NULL;
 543
 544         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 545                 handler = ocfs2_xattr_handler_map[name_index];
 546
 547         return handler ? handler->prefix : NULL;
 548 }
 549
 550 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 551                                  const char *name,
 552                                  int name_len)
 553 {
 554         /* Get hash value of uuid from super block */
 555         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 556         int i;
 557
 558         /* hash extended attribute name */
 559         for (i = 0; i < name_len; i++) {
 560                 hash = (hash << OCFS2_HASH_SHIFT) ^
 561                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 562                        *name++;
 563         }
 564
 565         return hash;
 566 }
 567
 568 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 569 {
 570         return namevalue_size(name_len, value_len) +
 571                 sizeof(struct ocfs2_xattr_entry);
 572 }
 573
 574 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 575 {
 576         return namevalue_size_xi(xi) +
 577                 sizeof(struct ocfs2_xattr_entry);
 578 }
 579
 580 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 581 {
 582         return namevalue_size_xe(xe) +
 583                 sizeof(struct ocfs2_xattr_entry);
 584 }
 585
 586 int ocfs2_calc_security_init(struct inode *dir,
 587                              struct ocfs2_security_xattr_info *si,
 588                              int *want_clusters,
 589                              int *xattr_credits,
 590                              struct ocfs2_alloc_context **xattr_ac)
 591 {
 592         int ret = 0;
 593         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 594         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 595                                                  si->value_len);
 596
 597         /*
 598          * The max space of security xattr taken inline is
 599          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 600          * So reserve one metadata block for it is ok.
 601          */
 602         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 603             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 604                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 605                 if (ret) {
 606                         mlog_errno(ret);
 607                         return ret;
 608                 }
 609                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 610         }
 611
 612         /* reserve clusters for xattr value which will be set in B tree*/
 613         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 614                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 615                                                             si->value_len);
 616
 617                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 618                                                            new_clusters);
 619                 *want_clusters += new_clusters;
 620         }
 621         return ret;
 622 }
 623
 624 int ocfs2_calc_xattr_init(struct inode *dir,
 625                           struct buffer_head *dir_bh,
 626                           umode_t mode,
 627                           struct ocfs2_security_xattr_info *si,
 628                           int *want_clusters,
 629                           int *xattr_credits,
 630                           int *want_meta)
 631 {
 632         int ret = 0;
 633         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 634         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 635
 636         if (si->enable)
 637                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 638                                                      si->value_len);
 639
 640         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 641                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 642                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 643                                         "", NULL, 0);
 644                 if (acl_len > 0) {
 645                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 646                         if (S_ISDIR(mode))
 647                                 a_size <<= 1;
 648                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 649                         mlog_errno(ret);
 650                         return ret;
 651                 }
 652         }
 653
 654         if (!(s_size + a_size))
 655                 return ret;
 656
 657         /*
 658          * The max space of security xattr taken inline is
 659          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 660          * The max space of acl xattr taken inline is
 661          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 662          * when blocksize = 512, may reserve one more cluser for
 663          * xattr bucket, otherwise reserve one metadata block
 664          * for them is ok.
 665          * If this is a new directory with inline data,
 666          * we choose to reserve the entire inline area for
 667          * directory contents and force an external xattr block.
 668          */
 669         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 670             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 671             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 672                 *want_meta = *want_meta + 1;
 673                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 674         }
 675
 676         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 677             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 678                 *want_clusters += 1;
 679                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 680         }
 681
 682         /*
 683          * reserve credits and clusters for xattrs which has large value
 684          * and have to be set outside
 685          */
 686         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 687                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 688                                                         si->value_len);
 689                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 690                                                            new_clusters);
 691                 *want_clusters += new_clusters;
 692         }
 693         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 694             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 695                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 696                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 697                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 698                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 699                                                            new_clusters);
 700                 *want_clusters += new_clusters;
 701         }
 702
 703         return ret;
 704 }
 705
 706 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 707                                          u32 clusters_to_add,
 708                                          struct ocfs2_xattr_value_buf *vb,
 709                                          struct ocfs2_xattr_set_ctxt *ctxt)
 710 {
 711         int status = 0, credits;
 712         handle_t *handle = ctxt->handle;
 713         enum ocfs2_alloc_restarted why;
 714         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 715         struct ocfs2_extent_tree et;
 716
 717         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 718
 719         while (clusters_to_add) {
 720                 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 721
 722                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 723                                        OCFS2_JOURNAL_ACCESS_WRITE);
 724                 if (status < 0) {
 725                         mlog_errno(status);
 726                         break;
 727                 }
 728
 729                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 730                 status = ocfs2_add_clusters_in_btree(handle,
 731                                                      &et,
 732                                                      &logical_start,
 733                                                      clusters_to_add,
 734                                                      0,
 735                                                      ctxt->data_ac,
 736                                                      ctxt->meta_ac,
 737                                                      &why);
 738                 if ((status < 0) && (status != -EAGAIN)) {
 739                         if (status != -ENOSPC)
 740                                 mlog_errno(status);
 741                         break;
 742                 }
 743
 744                 ocfs2_journal_dirty(handle, vb->vb_bh);
 745
 746                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 747                                          prev_clusters;
 748
 749                 if (why != RESTART_NONE && clusters_to_add) {
 750                         /*
 751                          * We can only fail in case the alloc file doesn't give
 752                          * up enough clusters.
 753                          */
 754                         BUG_ON(why == RESTART_META);
 755
 756                         credits = ocfs2_calc_extend_credits(inode->i_sb,
 757                                                             &vb->vb_xv->xr_list);
 758                         status = ocfs2_extend_trans(handle, credits);
 759                         if (status < 0) {
 760                                 status = -ENOMEM;
 761                                 mlog_errno(status);
 762                                 break;
 763                         }
 764                 }
 765         }
 766
 767         return status;
 768 }
 769
 770 static int __ocfs2_remove_xattr_range(struct inode *inode,
 771                                       struct ocfs2_xattr_value_buf *vb,
 772                                       u32 cpos, u32 phys_cpos, u32 len,
 773                                       unsigned int ext_flags,
 774                                       struct ocfs2_xattr_set_ctxt *ctxt)
 775 {
 776         int ret;
 777         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 778         handle_t *handle = ctxt->handle;
 779         struct ocfs2_extent_tree et;
 780
 781         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 782
 783         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 784                             OCFS2_JOURNAL_ACCESS_WRITE);
 785         if (ret) {
 786                 mlog_errno(ret);
 787                 goto out;
 788         }
 789
 790         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 791                                   &ctxt->dealloc);
 792         if (ret) {
 793                 mlog_errno(ret);
 794                 goto out;
 795         }
 796
 797         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 798         ocfs2_journal_dirty(handle, vb->vb_bh);
 799
 800         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 801                 ret = ocfs2_decrease_refcount(inode, handle,
 802                                         ocfs2_blocks_to_clusters(inode->i_sb,
 803                                                                  phys_blkno),
 804                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 805         else
 806                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 807                                                   phys_blkno, len);
 808         if (ret)
 809                 mlog_errno(ret);
 810
 811 out:
 812         return ret;
 813 }
 814
 815 static int ocfs2_xattr_shrink_size(struct inode *inode,
 816                                    u32 old_clusters,
 817                                    u32 new_clusters,
 818                                    struct ocfs2_xattr_value_buf *vb,
 819                                    struct ocfs2_xattr_set_ctxt *ctxt)
 820 {
 821         int ret = 0;
 822         unsigned int ext_flags;
 823         u32 trunc_len, cpos, phys_cpos, alloc_size;
 824         u64 block;
 825
 826         if (old_clusters <= new_clusters)
 827                 return 0;
 828
 829         cpos = new_clusters;
 830         trunc_len = old_clusters - new_clusters;
 831         while (trunc_len) {
 832                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 833                                                &alloc_size,
 834                                                &vb->vb_xv->xr_list, &ext_flags);
 835                 if (ret) {
 836                         mlog_errno(ret);
 837                         goto out;
 838                 }
 839
 840                 if (alloc_size > trunc_len)
 841                         alloc_size = trunc_len;
 842
 843                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 844                                                  phys_cpos, alloc_size,
 845                                                  ext_flags, ctxt);
 846                 if (ret) {
 847                         mlog_errno(ret);
 848                         goto out;
 849                 }
 850
 851                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 852                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 853                                                        block, alloc_size);
 854                 cpos += alloc_size;
 855                 trunc_len -= alloc_size;
 856         }
 857
 858 out:
 859         return ret;
 860 }
 861
 862 static int ocfs2_xattr_value_truncate(struct inode *inode,
 863                                       struct ocfs2_xattr_value_buf *vb,
 864                                       int len,
 865                                       struct ocfs2_xattr_set_ctxt *ctxt)
 866 {
 867         int ret;
 868         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 869         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 870
 871         if (new_clusters == old_clusters)
 872                 return 0;
 873
 874         if (new_clusters > old_clusters)
 875                 ret = ocfs2_xattr_extend_allocation(inode,
 876                                                     new_clusters - old_clusters,
 877                                                     vb, ctxt);
 878         else
 879                 ret = ocfs2_xattr_shrink_size(inode,
 880                                               old_clusters, new_clusters,
 881                                               vb, ctxt);
 882
 883         return ret;
 884 }
 885
 886 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 887                                   size_t *result, const char *prefix,
 888                                   const char *name, int name_len)
 889 {
 890         char *p = buffer + *result;
 891         int prefix_len = strlen(prefix);
 892         int total_len = prefix_len + name_len + 1;
 893
 894         *result += total_len;
 895
 896         /* we are just looking for how big our buffer needs to be */
 897         if (!size)
 898                 return 0;
 899
 900         if (*result > size)
 901                 return -ERANGE;
 902
 903         memcpy(p, prefix, prefix_len);
 904         memcpy(p + prefix_len, name, name_len);
 905         p[prefix_len + name_len] = '\0';
 906
 907         return 0;
 908 }
 909
 910 static int ocfs2_xattr_list_entries(struct inode *inode,
 911                                     struct ocfs2_xattr_header *header,
 912                                     char *buffer, size_t buffer_size)
 913 {
 914         size_t result = 0;
 915         int i, type, ret;
 916         const char *prefix, *name;
 917
 918         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 919                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 920                 type = ocfs2_xattr_get_type(entry);
 921                 prefix = ocfs2_xattr_prefix(type);
 922
 923                 if (prefix) {
 924                         name = (const char *)header +
 925                                 le16_to_cpu(entry->xe_name_offset);
 926
 927                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 928                                                      &result, prefix, name,
 929                                                      entry->xe_name_len);
 930                         if (ret)
 931                                 return ret;
 932                 }
 933         }
 934
 935         return result;
 936 }
 937
 938 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 939                                          struct ocfs2_dinode *di)
 940 {
 941         struct ocfs2_xattr_header *xh;
 942         int i;
 943
 944         xh = (struct ocfs2_xattr_header *)
 945                  ((void *)di + inode->i_sb->s_blocksize -
 946                  le16_to_cpu(di->i_xattr_inline_size));
 947
 948         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 949                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 950                         return 1;
 951
 952         return 0;
 953 }
 954
 955 static int ocfs2_xattr_ibody_list(struct inode *inode,
 956                                   struct ocfs2_dinode *di,
 957                                   char *buffer,
 958                                   size_t buffer_size)
 959 {
 960         struct ocfs2_xattr_header *header = NULL;
 961         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 962         int ret = 0;
 963
 964         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 965                 return ret;
 966
 967         header = (struct ocfs2_xattr_header *)
 968                  ((void *)di + inode->i_sb->s_blocksize -
 969                  le16_to_cpu(di->i_xattr_inline_size));
 970
 971         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 972
 973         return ret;
 974 }
 975
 976 static int ocfs2_xattr_block_list(struct inode *inode,
 977                                   struct ocfs2_dinode *di,
 978                                   char *buffer,
 979                                   size_t buffer_size)
 980 {
 981         struct buffer_head *blk_bh = NULL;
 982         struct ocfs2_xattr_block *xb;
 983         int ret = 0;
 984
 985         if (!di->i_xattr_loc)
 986                 return ret;
 987
 988         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 989                                      &blk_bh);
 990         if (ret < 0) {
 991                 mlog_errno(ret);
 992                 return ret;
 993         }
 994
 995         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 996         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 997                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 998                 ret = ocfs2_xattr_list_entries(inode, header,
 999                                                buffer, buffer_size);
1000         } else
1001                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1002                                                    buffer, buffer_size);
1003
1004         brelse(blk_bh);
1005
1006         return ret;
1007 }
1008
1009 ssize_t ocfs2_listxattr(struct dentry *dentry,
1010                         char *buffer,
1011                         size_t size)
1012 {
1013         int ret = 0, i_ret = 0, b_ret = 0;
1014         struct buffer_head *di_bh = NULL;
1015         struct ocfs2_dinode *di = NULL;
1016         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1017
1018         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1019                 return -EOPNOTSUPP;
1020
1021         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1022                 return ret;
1023
1024         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1025         if (ret < 0) {
1026                 mlog_errno(ret);
1027                 return ret;
1028         }
1029
1030         di = (struct ocfs2_dinode *)di_bh->b_data;
1031
1032         down_read(&oi->ip_xattr_sem);
1033         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1034         if (i_ret < 0)
1035                 b_ret = 0;
1036         else {
1037                 if (buffer) {
1038                         buffer += i_ret;
1039                         size -= i_ret;
1040                 }
1041                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1042                                                buffer, size);
1043                 if (b_ret < 0)
1044                         i_ret = 0;
1045         }
1046         up_read(&oi->ip_xattr_sem);
1047         ocfs2_inode_unlock(dentry->d_inode, 0);
1048
1049         brelse(di_bh);
1050
1051         return i_ret + b_ret;
1052 }
1053
1054 static int ocfs2_xattr_find_entry(int name_index,
1055                                   const char *name,
1056                                   struct ocfs2_xattr_search *xs)
1057 {
1058         struct ocfs2_xattr_entry *entry;
1059         size_t name_len;
1060         int i, cmp = 1;
1061
1062         if (name == NULL)
1063                 return -EINVAL;
1064
1065         name_len = strlen(name);
1066         entry = xs->here;
1067         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1068                 cmp = name_index - ocfs2_xattr_get_type(entry);
1069                 if (!cmp)
1070                         cmp = name_len - entry->xe_name_len;
1071                 if (!cmp)
1072                         cmp = memcmp(name, (xs->base +
1073                                      le16_to_cpu(entry->xe_name_offset)),
1074                                      name_len);
1075                 if (cmp == 0)
1076                         break;
1077                 entry += 1;
1078         }
1079         xs->here = entry;
1080
1081         return cmp ? -ENODATA : 0;
1082 }
1083
1084 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1085                                          struct ocfs2_xattr_value_root *xv,
1086                                          void *buffer,
1087                                          size_t len)
1088 {
1089         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1090         u64 blkno;
1091         int i, ret = 0;
1092         size_t cplen, blocksize;
1093         struct buffer_head *bh = NULL;
1094         struct ocfs2_extent_list *el;
1095
1096         el = &xv->xr_list;
1097         clusters = le32_to_cpu(xv->xr_clusters);
1098         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1099         blocksize = inode->i_sb->s_blocksize;
1100
1101         cpos = 0;
1102         while (cpos < clusters) {
1103                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1104                                                &num_clusters, el, NULL);
1105                 if (ret) {
1106                         mlog_errno(ret);
1107                         goto out;
1108                 }
1109
1110                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1111                 /* Copy ocfs2_xattr_value */
1112                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1113                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1114                                                &bh, NULL);
1115                         if (ret) {
1116                                 mlog_errno(ret);
1117                                 goto out;
1118                         }
1119
1120                         cplen = len >= blocksize ? blocksize : len;
1121                         memcpy(buffer, bh->b_data, cplen);
1122                         len -= cplen;
1123                         buffer += cplen;
1124
1125                         brelse(bh);
1126                         bh = NULL;
1127                         if (len == 0)
1128                                 break;
1129                 }
1130                 cpos += num_clusters;
1131         }
1132 out:
1133         return ret;
1134 }
1135
1136 static int ocfs2_xattr_ibody_get(struct inode *inode,
1137                                  int name_index,
1138                                  const char *name,
1139                                  void *buffer,
1140                                  size_t buffer_size,
1141                                  struct ocfs2_xattr_search *xs)
1142 {
1143         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1144         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1145         struct ocfs2_xattr_value_root *xv;
1146         size_t size;
1147         int ret = 0;
1148
1149         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1150                 return -ENODATA;
1151
1152         xs->end = (void *)di + inode->i_sb->s_blocksize;
1153         xs->header = (struct ocfs2_xattr_header *)
1154                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1155         xs->base = (void *)xs->header;
1156         xs->here = xs->header->xh_entries;
1157
1158         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1159         if (ret)
1160                 return ret;
1161         size = le64_to_cpu(xs->here->xe_value_size);
1162         if (buffer) {
1163                 if (size > buffer_size)
1164                         return -ERANGE;
1165                 if (ocfs2_xattr_is_local(xs->here)) {
1166                         memcpy(buffer, (void *)xs->base +
1167                                le16_to_cpu(xs->here->xe_name_offset) +
1168                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1169                 } else {
1170                         xv = (struct ocfs2_xattr_value_root *)
1171                                 (xs->base + le16_to_cpu(
1172                                  xs->here->xe_name_offset) +
1173                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1174                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1175                                                             buffer, size);
1176                         if (ret < 0) {
1177                                 mlog_errno(ret);
1178                                 return ret;
1179                         }
1180                 }
1181         }
1182
1183         return size;
1184 }
1185
1186 static int ocfs2_xattr_block_get(struct inode *inode,
1187                                  int name_index,
1188                                  const char *name,
1189                                  void *buffer,
1190                                  size_t buffer_size,
1191                                  struct ocfs2_xattr_search *xs)
1192 {
1193         struct ocfs2_xattr_block *xb;
1194         struct ocfs2_xattr_value_root *xv;
1195         size_t size;
1196         int ret = -ENODATA, name_offset, name_len, i;
1197         int uninitialized_var(block_off);
1198
1199         xs->bucket = ocfs2_xattr_bucket_new(inode);
1200         if (!xs->bucket) {
1201                 ret = -ENOMEM;
1202                 mlog_errno(ret);
1203                 goto cleanup;
1204         }
1205
1206         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1207         if (ret) {
1208                 mlog_errno(ret);
1209                 goto cleanup;
1210         }
1211
1212         if (xs->not_found) {
1213                 ret = -ENODATA;
1214                 goto cleanup;
1215         }
1216
1217         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1218         size = le64_to_cpu(xs->here->xe_value_size);
1219         if (buffer) {
1220                 ret = -ERANGE;
1221                 if (size > buffer_size)
1222                         goto cleanup;
1223
1224                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1225                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1226                 i = xs->here - xs->header->xh_entries;
1227
1228                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1229                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1230                                                                 bucket_xh(xs->bucket),
1231                                                                 i,
1232                                                                 &block_off,
1233                                                                 &name_offset);
1234                         xs->base = bucket_block(xs->bucket, block_off);
1235                 }
1236                 if (ocfs2_xattr_is_local(xs->here)) {
1237                         memcpy(buffer, (void *)xs->base +
1238                                name_offset + name_len, size);
1239                 } else {
1240                         xv = (struct ocfs2_xattr_value_root *)
1241                                 (xs->base + name_offset + name_len);
1242                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1243                                                             buffer, size);
1244                         if (ret < 0) {
1245                                 mlog_errno(ret);
1246                                 goto cleanup;
1247                         }
1248                 }
1249         }
1250         ret = size;
1251 cleanup:
1252         ocfs2_xattr_bucket_free(xs->bucket);
1253
1254         brelse(xs->xattr_bh);
1255         xs->xattr_bh = NULL;
1256         return ret;
1257 }
1258
1259 int ocfs2_xattr_get_nolock(struct inode *inode,
1260                            struct buffer_head *di_bh,
1261                            int name_index,
1262                            const char *name,
1263                            void *buffer,
1264                            size_t buffer_size)
1265 {
1266         int ret;
1267         struct ocfs2_dinode *di = NULL;
1268         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1269         struct ocfs2_xattr_search xis = {
1270                 .not_found = -ENODATA,
1271         };
1272         struct ocfs2_xattr_search xbs = {
1273                 .not_found = -ENODATA,
1274         };
1275
1276         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1277                 return -EOPNOTSUPP;
1278
1279         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1280                 ret = -ENODATA;
1281
1282         xis.inode_bh = xbs.inode_bh = di_bh;
1283         di = (struct ocfs2_dinode *)di_bh->b_data;
1284
1285         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1286                                     buffer_size, &xis);
1287         if (ret == -ENODATA && di->i_xattr_loc)
1288                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1289                                             buffer_size, &xbs);
1290
1291         return ret;
1292 }
1293
1294 /* ocfs2_xattr_get()
1295  *
1296  * Copy an extended attribute into the buffer provided.
1297  * Buffer is NULL to compute the size of buffer required.
1298  */
1299 static int ocfs2_xattr_get(struct inode *inode,
1300                            int name_index,
1301                            const char *name,
1302                            void *buffer,
1303                            size_t buffer_size)
1304 {
1305         int ret;
1306         struct buffer_head *di_bh = NULL;
1307
1308         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1309         if (ret < 0) {
1310                 mlog_errno(ret);
1311                 return ret;
1312         }
1313         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1314         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1315                                      name, buffer, buffer_size);
1316         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1317
1318         ocfs2_inode_unlock(inode, 0);
1319
1320         brelse(di_bh);
1321
1322         return ret;
1323 }
1324
1325 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1326                                            handle_t *handle,
1327                                            struct ocfs2_xattr_value_buf *vb,
1328                                            const void *value,
1329                                            int value_len)
1330 {
1331         int ret = 0, i, cp_len;
1332         u16 blocksize = inode->i_sb->s_blocksize;
1333         u32 p_cluster, num_clusters;
1334         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1335         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1336         u64 blkno;
1337         struct buffer_head *bh = NULL;
1338         unsigned int ext_flags;
1339         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1340
1341         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1342
1343         while (cpos < clusters) {
1344                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1345                                                &num_clusters, &xv->xr_list,
1346                                                &ext_flags);
1347                 if (ret) {
1348                         mlog_errno(ret);
1349                         goto out;
1350                 }
1351
1352                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1353
1354                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1355
1356                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1357                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1358                                                &bh, NULL);
1359                         if (ret) {
1360                                 mlog_errno(ret);
1361                                 goto out;
1362                         }
1363
1364                         ret = ocfs2_journal_access(handle,
1365                                                    INODE_CACHE(inode),
1366                                                    bh,
1367                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1368                         if (ret < 0) {
1369                                 mlog_errno(ret);
1370                                 goto out;
1371                         }
1372
1373                         cp_len = value_len > blocksize ? blocksize : value_len;
1374                         memcpy(bh->b_data, value, cp_len);
1375                         value_len -= cp_len;
1376                         value += cp_len;
1377                         if (cp_len < blocksize)
1378                                 memset(bh->b_data + cp_len, 0,
1379                                        blocksize - cp_len);
1380
1381                         ocfs2_journal_dirty(handle, bh);
1382                         brelse(bh);
1383                         bh = NULL;
1384
1385                         /*
1386                          * XXX: do we need to empty all the following
1387                          * blocks in this cluster?
1388                          */
1389                         if (!value_len)
1390                                 break;
1391                 }
1392                 cpos += num_clusters;
1393         }
1394 out:
1395         brelse(bh);
1396
1397         return ret;
1398 }
1399
1400 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1401                                        int num_entries)
1402 {
1403         int free_space;
1404
1405         if (!needed_space)
1406                 return 0;
1407
1408         free_space = free_start -
1409                 sizeof(struct ocfs2_xattr_header) -
1410                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1411                 OCFS2_XATTR_HEADER_GAP;
1412         if (free_space < 0)
1413                 return -EIO;
1414         if (free_space < needed_space)
1415                 return -ENOSPC;
1416
1417         return 0;
1418 }
1419
1420 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1421                                    int type)
1422 {
1423         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1424 }
1425
1426 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1427 {
1428         loc->xl_ops->xlo_journal_dirty(handle, loc);
1429 }
1430
1431 /* Give a pointer into the storage for the given offset */
1432 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1433 {
1434         BUG_ON(offset >= loc->xl_size);
1435         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1436 }
1437
1438 /*
1439  * Wipe the name+value pair and allow the storage to reclaim it.  This
1440  * must be followed by either removal of the entry or a call to
1441  * ocfs2_xa_add_namevalue().
1442  */
1443 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1444 {
1445         loc->xl_ops->xlo_wipe_namevalue(loc);
1446 }
1447
1448 /*
1449  * Find lowest offset to a name+value pair.  This is the start of our
1450  * downward-growing free space.
1451  */
1452 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1453 {
1454         return loc->xl_ops->xlo_get_free_start(loc);
1455 }
1456
1457 /* Can we reuse loc->xl_entry for xi? */
1458 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1459                                     struct ocfs2_xattr_info *xi)
1460 {
1461         return loc->xl_ops->xlo_can_reuse(loc, xi);
1462 }
1463
1464 /* How much free space is needed to set the new value */
1465 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1466                                 struct ocfs2_xattr_info *xi)
1467 {
1468         return loc->xl_ops->xlo_check_space(loc, xi);
1469 }
1470
1471 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1472 {
1473         loc->xl_ops->xlo_add_entry(loc, name_hash);
1474         loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1475         /*
1476          * We can't leave the new entry's xe_name_offset at zero or
1477          * add_namevalue() will go nuts.  We set it to the size of our
1478          * storage so that it can never be less than any other entry.
1479          */
1480         loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1481 }
1482
1483 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1484                                    struct ocfs2_xattr_info *xi)
1485 {
1486         int size = namevalue_size_xi(xi);
1487         int nameval_offset;
1488         char *nameval_buf;
1489
1490         loc->xl_ops->xlo_add_namevalue(loc, size);
1491         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1492         loc->xl_entry->xe_name_len = xi->xi_name_len;
1493         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1494         ocfs2_xattr_set_local(loc->xl_entry,
1495                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1496
1497         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1498         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1499         memset(nameval_buf, 0, size);
1500         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1501 }
1502
1503 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1504                                     struct ocfs2_xattr_value_buf *vb)
1505 {
1506         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1507         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1508
1509         /* Value bufs are for value trees */
1510         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1511         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1512                (name_size + OCFS2_XATTR_ROOT_SIZE));
1513
1514         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1515         vb->vb_xv =
1516                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1517                                                         nameval_offset +
1518                                                         name_size);
1519 }
1520
1521 static int ocfs2_xa_block_journal_access(handle_t *handle,
1522                                          struct ocfs2_xa_loc *loc, int type)
1523 {
1524         struct buffer_head *bh = loc->xl_storage;
1525         ocfs2_journal_access_func access;
1526
1527         if (loc->xl_size == (bh->b_size -
1528                              offsetof(struct ocfs2_xattr_block,
1529                                       xb_attrs.xb_header)))
1530                 access = ocfs2_journal_access_xb;
1531         else
1532                 access = ocfs2_journal_access_di;
1533         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1534 }
1535
1536 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1537                                          struct ocfs2_xa_loc *loc)
1538 {
1539         struct buffer_head *bh = loc->xl_storage;
1540
1541         ocfs2_journal_dirty(handle, bh);
1542 }
1543
1544 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1545                                            int offset)
1546 {
1547         return (char *)loc->xl_header + offset;
1548 }
1549
1550 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1551                                     struct ocfs2_xattr_info *xi)
1552 {
1553         /*
1554          * Block storage is strict.  If the sizes aren't exact, we will
1555          * remove the old one and reinsert the new.
1556          */
1557         return namevalue_size_xe(loc->xl_entry) ==
1558                 namevalue_size_xi(xi);
1559 }
1560
1561 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1562 {
1563         struct ocfs2_xattr_header *xh = loc->xl_header;
1564         int i, count = le16_to_cpu(xh->xh_count);
1565         int offset, free_start = loc->xl_size;
1566
1567         for (i = 0; i < count; i++) {
1568                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1569                 if (offset < free_start)
1570                         free_start = offset;
1571         }
1572
1573         return free_start;
1574 }
1575
1576 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1577                                       struct ocfs2_xattr_info *xi)
1578 {
1579         int count = le16_to_cpu(loc->xl_header->xh_count);
1580         int free_start = ocfs2_xa_get_free_start(loc);
1581         int needed_space = ocfs2_xi_entry_usage(xi);
1582
1583         /*
1584          * Block storage will reclaim the original entry before inserting
1585          * the new value, so we only need the difference.  If the new
1586          * entry is smaller than the old one, we don't need anything.
1587          */
1588         if (loc->xl_entry) {
1589                 /* Don't need space if we're reusing! */
1590                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1591                         needed_space = 0;
1592                 else
1593                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1594         }
1595         if (needed_space < 0)
1596                 needed_space = 0;
1597         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1598 }
1599
1600 /*
1601  * Block storage for xattrs keeps the name+value pairs compacted.  When
1602  * we remove one, we have to shift any that preceded it towards the end.
1603  */
1604 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1605 {
1606         int i, offset;
1607         int namevalue_offset, first_namevalue_offset, namevalue_size;
1608         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1609         struct ocfs2_xattr_header *xh = loc->xl_header;
1610         int count = le16_to_cpu(xh->xh_count);
1611
1612         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1613         namevalue_size = namevalue_size_xe(entry);
1614         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1615
1616         /* Shift the name+value pairs */
1617         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1618                 (char *)xh + first_namevalue_offset,
1619                 namevalue_offset - first_namevalue_offset);
1620         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1621
1622         /* Now tell xh->xh_entries about it */
1623         for (i = 0; i < count; i++) {
1624                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1625                 if (offset <= namevalue_offset)
1626                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1627                                      namevalue_size);
1628         }
1629
1630         /*
1631          * Note that we don't update xh_free_start or xh_name_value_len
1632          * because they're not used in block-stored xattrs.
1633          */
1634 }
1635
1636 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1637 {
1638         int count = le16_to_cpu(loc->xl_header->xh_count);
1639         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1640         le16_add_cpu(&loc->xl_header->xh_count, 1);
1641         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1642 }
1643
1644 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1645 {
1646         int free_start = ocfs2_xa_get_free_start(loc);
1647
1648         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1649 }
1650
1651 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1652                                           struct ocfs2_xattr_value_buf *vb)
1653 {
1654         struct buffer_head *bh = loc->xl_storage;
1655
1656         if (loc->xl_size == (bh->b_size -
1657                              offsetof(struct ocfs2_xattr_block,
1658                                       xb_attrs.xb_header)))
1659                 vb->vb_access = ocfs2_journal_access_xb;
1660         else
1661                 vb->vb_access = ocfs2_journal_access_di;
1662         vb->vb_bh = bh;
1663 }
1664
1665 /*
1666  * Operations for xattrs stored in blocks.  This includes inline inode
1667  * storage and unindexed ocfs2_xattr_blocks.
1668  */
1669 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1670         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1671         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1672         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1673         .xlo_check_space        = ocfs2_xa_block_check_space,
1674         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1675         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1676         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1677         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1678         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1679         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1680 };
1681
1682 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1683                                           struct ocfs2_xa_loc *loc, int type)
1684 {
1685         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1686
1687         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1688 }
1689
1690 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1691                                           struct ocfs2_xa_loc *loc)
1692 {
1693         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1694
1695         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1696 }
1697
1698 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1699                                             int offset)
1700 {
1701         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1702         int block, block_offset;
1703
1704         /* The header is at the front of the bucket */
1705         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1706         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1707
1708         return bucket_block(bucket, block) + block_offset;
1709 }
1710
1711 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1712                                      struct ocfs2_xattr_info *xi)
1713 {
1714         return namevalue_size_xe(loc->xl_entry) >=
1715                 namevalue_size_xi(xi);
1716 }
1717
1718 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1719 {
1720         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1721         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1722 }
1723
1724 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1725                                          int free_start, int size)
1726 {
1727         /*
1728          * We need to make sure that the name+value pair fits within
1729          * one block.
1730          */
1731         if (((free_start - size) >> sb->s_blocksize_bits) !=
1732             ((free_start - 1) >> sb->s_blocksize_bits))
1733                 free_start -= free_start % sb->s_blocksize;
1734
1735         return free_start;
1736 }
1737
1738 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1739                                        struct ocfs2_xattr_info *xi)
1740 {
1741         int rc;
1742         int count = le16_to_cpu(loc->xl_header->xh_count);
1743         int free_start = ocfs2_xa_get_free_start(loc);
1744         int needed_space = ocfs2_xi_entry_usage(xi);
1745         int size = namevalue_size_xi(xi);
1746         struct super_block *sb = loc->xl_inode->i_sb;
1747
1748         /*
1749          * Bucket storage does not reclaim name+value pairs it cannot
1750          * reuse.  They live as holes until the bucket fills, and then
1751          * the bucket is defragmented.  However, the bucket can reclaim
1752          * the ocfs2_xattr_entry.
1753          */
1754         if (loc->xl_entry) {
1755                 /* Don't need space if we're reusing! */
1756                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1757                         needed_space = 0;
1758                 else
1759                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1760         }
1761         BUG_ON(needed_space < 0);
1762
1763         if (free_start < size) {
1764                 if (needed_space)
1765                         return -ENOSPC;
1766         } else {
1767                 /*
1768                  * First we check if it would fit in the first place.
1769                  * Below, we align the free start to a block.  This may
1770                  * slide us below the minimum gap.  By checking unaligned
1771                  * first, we avoid that error.
1772                  */
1773                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1774                                                  count);
1775                 if (rc)
1776                         return rc;
1777                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1778                                                            size);
1779         }
1780         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1781 }
1782
1783 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1784 {
1785         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1786                      -namevalue_size_xe(loc->xl_entry));
1787 }
1788
1789 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1790 {
1791         struct ocfs2_xattr_header *xh = loc->xl_header;
1792         int count = le16_to_cpu(xh->xh_count);
1793         int low = 0, high = count - 1, tmp;
1794         struct ocfs2_xattr_entry *tmp_xe;
1795
1796         /*
1797          * We keep buckets sorted by name_hash, so we need to find
1798          * our insert place.
1799          */
1800         while (low <= high && count) {
1801                 tmp = (low + high) / 2;
1802                 tmp_xe = &xh->xh_entries[tmp];
1803
1804                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1805                         low = tmp + 1;
1806                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1807                         high = tmp - 1;
1808                 else {
1809                         low = tmp;
1810                         break;
1811                 }
1812         }
1813
1814         if (low != count)
1815                 memmove(&xh->xh_entries[low + 1],
1816                         &xh->xh_entries[low],
1817                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1818
1819         le16_add_cpu(&xh->xh_count, 1);
1820         loc->xl_entry = &xh->xh_entries[low];
1821         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1822 }
1823
1824 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1825 {
1826         int free_start = ocfs2_xa_get_free_start(loc);
1827         struct ocfs2_xattr_header *xh = loc->xl_header;
1828         struct super_block *sb = loc->xl_inode->i_sb;
1829         int nameval_offset;
1830
1831         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1832         nameval_offset = free_start - size;
1833         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1834         xh->xh_free_start = cpu_to_le16(nameval_offset);
1835         le16_add_cpu(&xh->xh_name_value_len, size);
1836
1837 }
1838
1839 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1840                                            struct ocfs2_xattr_value_buf *vb)
1841 {
1842         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1843         struct super_block *sb = loc->xl_inode->i_sb;
1844         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1845         int size = namevalue_size_xe(loc->xl_entry);
1846         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1847
1848         /* Values are not allowed to straddle block boundaries */
1849         BUG_ON(block_offset !=
1850                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1851         /* We expect the bucket to be filled in */
1852         BUG_ON(!bucket->bu_bhs[block_offset]);
1853
1854         vb->vb_access = ocfs2_journal_access;
1855         vb->vb_bh = bucket->bu_bhs[block_offset];
1856 }
1857
1858 /* Operations for xattrs stored in buckets. */
1859 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1860         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1861         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1862         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1863         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1864         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1865         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1866         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1867         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1868         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1869         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1870 };
1871
1872 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1873 {
1874         struct ocfs2_xattr_value_buf vb;
1875
1876         if (ocfs2_xattr_is_local(loc->xl_entry))
1877                 return 0;
1878
1879         ocfs2_xa_fill_value_buf(loc, &vb);
1880         return le32_to_cpu(vb.vb_xv->xr_clusters);
1881 }
1882
1883 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1884                                    struct ocfs2_xattr_set_ctxt *ctxt)
1885 {
1886         int trunc_rc, access_rc;
1887         struct ocfs2_xattr_value_buf vb;
1888
1889         ocfs2_xa_fill_value_buf(loc, &vb);
1890         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1891                                               ctxt);
1892
1893         /*
1894          * The caller of ocfs2_xa_value_truncate() has already called
1895          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1896          * calls ocfs2_extend_trans().  This may commit the previous
1897          * transaction and open a new one.  If this is a bucket, truncate
1898          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1899          * the caller is expecting to dirty the entire bucket.  So we must
1900          * reset the journal work.  We do this even if truncate has failed,
1901          * as it could have failed after committing the extend.
1902          */
1903         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1904                                             OCFS2_JOURNAL_ACCESS_WRITE);
1905
1906         /* Errors in truncate take precedence */
1907         return trunc_rc ? trunc_rc : access_rc;
1908 }
1909
1910 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1911 {
1912         int index, count;
1913         struct ocfs2_xattr_header *xh = loc->xl_header;
1914         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1915
1916         ocfs2_xa_wipe_namevalue(loc);
1917         loc->xl_entry = NULL;
1918
1919         le16_add_cpu(&xh->xh_count, -1);
1920         count = le16_to_cpu(xh->xh_count);
1921
1922         /*
1923          * Only zero out the entry if there are more remaining.  This is
1924          * important for an empty bucket, as it keeps track of the
1925          * bucket's hash value.  It doesn't hurt empty block storage.
1926          */
1927         if (count) {
1928                 index = ((char *)entry - (char *)&xh->xh_entries) /
1929                         sizeof(struct ocfs2_xattr_entry);
1930                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1931                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1932                 memset(&xh->xh_entries[count], 0,
1933                        sizeof(struct ocfs2_xattr_entry));
1934         }
1935 }
1936
1937 /*
1938  * If we have a problem adjusting the size of an external value during
1939  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1940  * in an intermediate state.  For example, the value may be partially
1941  * truncated.
1942  *
1943  * If the value tree hasn't changed, the extend/truncate went nowhere.
1944  * We have nothing to do.  The caller can treat it as a straight error.
1945  *
1946  * If the value tree got partially truncated, we now have a corrupted
1947  * extended attribute.  We're going to wipe its entry and leak the
1948  * clusters.  Better to leak some storage than leave a corrupt entry.
1949  *
1950  * If the value tree grew, it obviously didn't grow enough for the
1951  * new entry.  We're not going to try and reclaim those clusters either.
1952  * If there was already an external value there (orig_clusters != 0),
1953  * the new clusters are attached safely and we can just leave the old
1954  * value in place.  If there was no external value there, we remove
1955  * the entry.
1956  *
1957  * This way, the xattr block we store in the journal will be consistent.
1958  * If the size change broke because of the journal, no changes will hit
1959  * disk anyway.
1960  */
1961 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1962                                             const char *what,
1963                                             unsigned int orig_clusters)
1964 {
1965         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1966         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1967                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1968
1969         if (new_clusters < orig_clusters) {
1970                 mlog(ML_ERROR,
1971                      "Partial truncate while %s xattr %.*s.  Leaking "
1972                      "%u clusters and removing the entry\n",
1973                      what, loc->xl_entry->xe_name_len, nameval_buf,
1974                      orig_clusters - new_clusters);
1975                 ocfs2_xa_remove_entry(loc);
1976         } else if (!orig_clusters) {
1977                 mlog(ML_ERROR,
1978                      "Unable to allocate an external value for xattr "
1979                      "%.*s safely.  Leaking %u clusters and removing the "
1980                      "entry\n",
1981                      loc->xl_entry->xe_name_len, nameval_buf,
1982                      new_clusters - orig_clusters);
1983                 ocfs2_xa_remove_entry(loc);
1984         } else if (new_clusters > orig_clusters)
1985                 mlog(ML_ERROR,
1986                      "Unable to grow xattr %.*s safely.  %u new clusters "
1987                      "have been added, but the value will not be "
1988                      "modified\n",
1989                      loc->xl_entry->xe_name_len, nameval_buf,
1990                      new_clusters - orig_clusters);
1991 }
1992
1993 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
1994                            struct ocfs2_xattr_set_ctxt *ctxt)
1995 {
1996         int rc = 0;
1997         unsigned int orig_clusters;
1998
1999         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2000                 orig_clusters = ocfs2_xa_value_clusters(loc);
2001                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2002                 if (rc) {
2003                         mlog_errno(rc);
2004                         /*
2005                          * Since this is remove, we can return 0 if
2006                          * ocfs2_xa_cleanup_value_truncate() is going to
2007                          * wipe the entry anyway.  So we check the
2008                          * cluster count as well.
2009                          */
2010                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2011                                 rc = 0;
2012                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2013                                                         orig_clusters);
2014                         if (rc)
2015                                 goto out;
2016                 }
2017         }
2018
2019         ocfs2_xa_remove_entry(loc);
2020
2021 out:
2022         return rc;
2023 }
2024
2025 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2026 {
2027         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2028         char *nameval_buf;
2029
2030         nameval_buf = ocfs2_xa_offset_pointer(loc,
2031                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2032         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2033 }
2034
2035 /*
2036  * Take an existing entry and make it ready for the new value.  This
2037  * won't allocate space, but it may free space.  It should be ready for
2038  * ocfs2_xa_prepare_entry() to finish the work.
2039  */
2040 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2041                                 struct ocfs2_xattr_info *xi,
2042                                 struct ocfs2_xattr_set_ctxt *ctxt)
2043 {
2044         int rc = 0;
2045         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2046         unsigned int orig_clusters;
2047         char *nameval_buf;
2048         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2049         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2050
2051         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2052                name_size);
2053
2054         nameval_buf = ocfs2_xa_offset_pointer(loc,
2055                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2056         if (xe_local) {
2057                 memset(nameval_buf + name_size, 0,
2058                        namevalue_size_xe(loc->xl_entry) - name_size);
2059                 if (!xi_local)
2060                         ocfs2_xa_install_value_root(loc);
2061         } else {
2062                 orig_clusters = ocfs2_xa_value_clusters(loc);
2063                 if (xi_local) {
2064                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2065                         if (rc < 0)
2066                                 mlog_errno(rc);
2067                         else
2068                                 memset(nameval_buf + name_size, 0,
2069                                        namevalue_size_xe(loc->xl_entry) -
2070                                        name_size);
2071                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2072                            xi->xi_value_len) {
2073                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2074                                                      ctxt);
2075                         if (rc < 0)
2076                                 mlog_errno(rc);
2077                 }
2078
2079                 if (rc) {
2080                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2081                                                         orig_clusters);
2082                         goto out;
2083                 }
2084         }
2085
2086         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2087         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2088
2089 out:
2090         return rc;
2091 }
2092
2093 /*
2094  * Prepares loc->xl_entry to receive the new xattr.  This includes
2095  * properly setting up the name+value pair region.  If loc->xl_entry
2096  * already exists, it will take care of modifying it appropriately.
2097  *
2098  * Note that this modifies the data.  You did journal_access already,
2099  * right?
2100  */
2101 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2102                                   struct ocfs2_xattr_info *xi,
2103                                   u32 name_hash,
2104                                   struct ocfs2_xattr_set_ctxt *ctxt)
2105 {
2106         int rc = 0;
2107         unsigned int orig_clusters;
2108         __le64 orig_value_size = 0;
2109
2110         rc = ocfs2_xa_check_space(loc, xi);
2111         if (rc)
2112                 goto out;
2113
2114         if (loc->xl_entry) {
2115                 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2116                         orig_value_size = loc->xl_entry->xe_value_size;
2117                         rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2118                         if (rc)
2119                                 goto out;
2120                         goto alloc_value;
2121                 }
2122
2123                 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2124                         orig_clusters = ocfs2_xa_value_clusters(loc);
2125                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2126                         if (rc) {
2127                                 mlog_errno(rc);
2128                                 ocfs2_xa_cleanup_value_truncate(loc,
2129                                                                 "overwriting",
2130                                                                 orig_clusters);
2131                                 goto out;
2132                         }
2133                 }
2134                 ocfs2_xa_wipe_namevalue(loc);
2135         } else
2136                 ocfs2_xa_add_entry(loc, name_hash);
2137
2138         /*
2139          * If we get here, we have a blank entry.  Fill it.  We grow our
2140          * name+value pair back from the end.
2141          */
2142         ocfs2_xa_add_namevalue(loc, xi);
2143         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2144                 ocfs2_xa_install_value_root(loc);
2145
2146 alloc_value:
2147         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2148                 orig_clusters = ocfs2_xa_value_clusters(loc);
2149                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2150                 if (rc < 0) {
2151                         ctxt->set_abort = 1;
2152                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2153                                                         orig_clusters);
2154                         /*
2155                          * If we were growing an existing value,
2156                          * ocfs2_xa_cleanup_value_truncate() won't remove
2157                          * the entry. We need to restore the original value
2158                          * size.
2159                          */
2160                         if (loc->xl_entry) {
2161                                 BUG_ON(!orig_value_size);
2162                                 loc->xl_entry->xe_value_size = orig_value_size;
2163                         }
2164                         mlog_errno(rc);
2165                 }
2166         }
2167
2168 out:
2169         return rc;
2170 }
2171
2172 /*
2173  * Store the value portion of the name+value pair.  This will skip
2174  * values that are stored externally.  Their tree roots were set up
2175  * by ocfs2_xa_prepare_entry().
2176  */
2177 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2178                                 struct ocfs2_xattr_info *xi,
2179                                 struct ocfs2_xattr_set_ctxt *ctxt)
2180 {
2181         int rc = 0;
2182         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2183         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2184         char *nameval_buf;
2185         struct ocfs2_xattr_value_buf vb;
2186
2187         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2188         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2189                 ocfs2_xa_fill_value_buf(loc, &vb);
2190                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2191                                                      ctxt->handle, &vb,
2192                                                      xi->xi_value,
2193                                                      xi->xi_value_len);
2194         } else
2195                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2196
2197         return rc;
2198 }
2199
2200 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2201                         struct ocfs2_xattr_info *xi,
2202                         struct ocfs2_xattr_set_ctxt *ctxt)
2203 {
2204         int ret;
2205         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2206                                               xi->xi_name_len);
2207
2208         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2209                                       OCFS2_JOURNAL_ACCESS_WRITE);
2210         if (ret) {
2211                 mlog_errno(ret);
2212                 goto out;
2213         }
2214
2215         /*
2216          * From here on out, everything is going to modify the buffer a
2217          * little.  Errors are going to leave the xattr header in a
2218          * sane state.  Thus, even with errors we dirty the sucker.
2219          */
2220
2221         /* Don't worry, we are never called with !xi_value and !xl_entry */
2222         if (!xi->xi_value) {
2223                 ret = ocfs2_xa_remove(loc, ctxt);
2224                 goto out_dirty;
2225         }
2226
2227         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2228         if (ret) {
2229                 if (ret != -ENOSPC)
2230                         mlog_errno(ret);
2231                 goto out_dirty;
2232         }
2233
2234         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2235         if (ret)
2236                 mlog_errno(ret);
2237
2238 out_dirty:
2239         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2240
2241 out:
2242         return ret;
2243 }
2244
2245 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2246                                      struct inode *inode,
2247                                      struct buffer_head *bh,
2248                                      struct ocfs2_xattr_entry *entry)
2249 {
2250         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2251
2252         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2253
2254         loc->xl_inode = inode;
2255         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2256         loc->xl_storage = bh;
2257         loc->xl_entry = entry;
2258         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2259         loc->xl_header =
2260                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2261                                               loc->xl_size);
2262 }
2263
2264 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2265                                           struct inode *inode,
2266                                           struct buffer_head *bh,
2267                                           struct ocfs2_xattr_entry *entry)
2268 {
2269         struct ocfs2_xattr_block *xb =
2270                 (struct ocfs2_xattr_block *)bh->b_data;
2271
2272         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2273
2274         loc->xl_inode = inode;
2275         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2276         loc->xl_storage = bh;
2277         loc->xl_header = &(xb->xb_attrs.xb_header);
2278         loc->xl_entry = entry;
2279         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2280                                              xb_attrs.xb_header);
2281 }
2282
2283 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2284                                            struct ocfs2_xattr_bucket *bucket,
2285                                            struct ocfs2_xattr_entry *entry)
2286 {
2287         loc->xl_inode = bucket->bu_inode;
2288         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2289         loc->xl_storage = bucket;
2290         loc->xl_header = bucket_xh(bucket);
2291         loc->xl_entry = entry;
2292         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2293 }
2294
2295 /*
2296  * In xattr remove, if it is stored outside and refcounted, we may have
2297  * the chance to split the refcount tree. So need the allocators.
2298  */
2299 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2300                                         struct ocfs2_xattr_value_root *xv,
2301                                         struct ocfs2_caching_info *ref_ci,
2302                                         struct buffer_head *ref_root_bh,
2303                                         struct ocfs2_alloc_context **meta_ac,
2304                                         int *ref_credits)
2305 {
2306         int ret, meta_add = 0;
2307         u32 p_cluster, num_clusters;
2308         unsigned int ext_flags;
2309
2310         *ref_credits = 0;
2311         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2312                                        &num_clusters,
2313                                        &xv->xr_list,
2314                                        &ext_flags);
2315         if (ret) {
2316                 mlog_errno(ret);
2317                 goto out;
2318         }
2319
2320         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2321                 goto out;
2322
2323         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2324                                                  ref_root_bh, xv,
2325                                                  &meta_add, ref_credits);
2326         if (ret) {
2327                 mlog_errno(ret);
2328                 goto out;
2329         }
2330
2331         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2332                                                 meta_add, meta_ac);
2333         if (ret)
2334                 mlog_errno(ret);
2335
2336 out:
2337         return ret;
2338 }
2339
2340 static int ocfs2_remove_value_outside(struct inode*inode,
2341                                       struct ocfs2_xattr_value_buf *vb,
2342                                       struct ocfs2_xattr_header *header,
2343                                       struct ocfs2_caching_info *ref_ci,
2344                                       struct buffer_head *ref_root_bh)
2345 {
2346         int ret = 0, i, ref_credits;
2347         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2348         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2349         void *val;
2350
2351         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2352
2353         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2354                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2355
2356                 if (ocfs2_xattr_is_local(entry))
2357                         continue;
2358
2359                 val = (void *)header +
2360                         le16_to_cpu(entry->xe_name_offset);
2361                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2362                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2363
2364                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2365                                                          ref_ci, ref_root_bh,
2366                                                          &ctxt.meta_ac,
2367                                                          &ref_credits);
2368
2369                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2370                                         ocfs2_remove_extent_credits(osb->sb));
2371                 if (IS_ERR(ctxt.handle)) {
2372                         ret = PTR_ERR(ctxt.handle);
2373                         mlog_errno(ret);
2374                         break;
2375                 }
2376
2377                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2378
2379                 ocfs2_commit_trans(osb, ctxt.handle);
2380                 if (ctxt.meta_ac) {
2381                         ocfs2_free_alloc_context(ctxt.meta_ac);
2382                         ctxt.meta_ac = NULL;
2383                 }
2384
2385                 if (ret < 0) {
2386                         mlog_errno(ret);
2387                         break;
2388                 }
2389
2390         }
2391
2392         if (ctxt.meta_ac)
2393                 ocfs2_free_alloc_context(ctxt.meta_ac);
2394         ocfs2_schedule_truncate_log_flush(osb, 1);
2395         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2396         return ret;
2397 }
2398
2399 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2400                                     struct buffer_head *di_bh,
2401                                     struct ocfs2_caching_info *ref_ci,
2402                                     struct buffer_head *ref_root_bh)
2403 {
2404
2405         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2406         struct ocfs2_xattr_header *header;
2407         int ret;
2408         struct ocfs2_xattr_value_buf vb = {
2409                 .vb_bh = di_bh,
2410                 .vb_access = ocfs2_journal_access_di,
2411         };
2412
2413         header = (struct ocfs2_xattr_header *)
2414                  ((void *)di + inode->i_sb->s_blocksize -
2415                  le16_to_cpu(di->i_xattr_inline_size));
2416
2417         ret = ocfs2_remove_value_outside(inode, &vb, header,
2418                                          ref_ci, ref_root_bh);
2419
2420         return ret;
2421 }
2422
2423 struct ocfs2_rm_xattr_bucket_para {
2424         struct ocfs2_caching_info *ref_ci;
2425         struct buffer_head *ref_root_bh;
2426 };
2427
2428 static int ocfs2_xattr_block_remove(struct inode *inode,
2429                                     struct buffer_head *blk_bh,
2430                                     struct ocfs2_caching_info *ref_ci,
2431                                     struct buffer_head *ref_root_bh)
2432 {
2433         struct ocfs2_xattr_block *xb;
2434         int ret = 0;
2435         struct ocfs2_xattr_value_buf vb = {
2436                 .vb_bh = blk_bh,
2437                 .vb_access = ocfs2_journal_access_xb,
2438         };
2439         struct ocfs2_rm_xattr_bucket_para args = {
2440                 .ref_ci = ref_ci,
2441                 .ref_root_bh = ref_root_bh,
2442         };
2443
2444         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2445         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2446                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2447                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2448                                                  ref_ci, ref_root_bh);
2449         } else
2450                 ret = ocfs2_iterate_xattr_index_block(inode,
2451                                                 blk_bh,
2452                                                 ocfs2_rm_xattr_cluster,
2453                                                 &args);
2454
2455         return ret;
2456 }
2457
2458 static int ocfs2_xattr_free_block(struct inode *inode,
2459                                   u64 block,
2460                                   struct ocfs2_caching_info *ref_ci,
2461                                   struct buffer_head *ref_root_bh)
2462 {
2463         struct inode *xb_alloc_inode;
2464         struct buffer_head *xb_alloc_bh = NULL;
2465         struct buffer_head *blk_bh = NULL;
2466         struct ocfs2_xattr_block *xb;
2467         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2468         handle_t *handle;
2469         int ret = 0;
2470         u64 blk, bg_blkno;
2471         u16 bit;
2472
2473         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2474         if (ret < 0) {
2475                 mlog_errno(ret);
2476                 goto out;
2477         }
2478
2479         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2480         if (ret < 0) {
2481                 mlog_errno(ret);
2482                 goto out;
2483         }
2484
2485         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2486         blk = le64_to_cpu(xb->xb_blkno);
2487         bit = le16_to_cpu(xb->xb_suballoc_bit);
2488         if (xb->xb_suballoc_loc)
2489                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2490         else
2491                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2492
2493         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2494                                 EXTENT_ALLOC_SYSTEM_INODE,
2495                                 le16_to_cpu(xb->xb_suballoc_slot));
2496         if (!xb_alloc_inode) {
2497                 ret = -ENOMEM;
2498                 mlog_errno(ret);
2499                 goto out;
2500         }
2501         mutex_lock(&xb_alloc_inode->i_mutex);
2502
2503         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2504         if (ret < 0) {
2505                 mlog_errno(ret);
2506                 goto out_mutex;
2507         }
2508
2509         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2510         if (IS_ERR(handle)) {
2511                 ret = PTR_ERR(handle);
2512                 mlog_errno(ret);
2513                 goto out_unlock;
2514         }
2515
2516         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2517                                        bit, bg_blkno, 1);
2518         if (ret < 0)
2519                 mlog_errno(ret);
2520
2521         ocfs2_commit_trans(osb, handle);
2522 out_unlock:
2523         ocfs2_inode_unlock(xb_alloc_inode, 1);
2524         brelse(xb_alloc_bh);
2525 out_mutex:
2526         mutex_unlock(&xb_alloc_inode->i_mutex);
2527         iput(xb_alloc_inode);
2528 out:
2529         brelse(blk_bh);
2530         return ret;
2531 }
2532
2533 /*
2534  * ocfs2_xattr_remove()
2535  *
2536  * Free extended attribute resources associated with this inode.
2537  */
2538 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2539 {
2540         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2541         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2542         struct ocfs2_refcount_tree *ref_tree = NULL;
2543         struct buffer_head *ref_root_bh = NULL;
2544         struct ocfs2_caching_info *ref_ci = NULL;
2545         handle_t *handle;
2546         int ret;
2547
2548         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2549                 return 0;
2550
2551         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2552                 return 0;
2553
2554         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2555                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2556                                                le64_to_cpu(di->i_refcount_loc),
2557                                                1, &ref_tree, &ref_root_bh);
2558                 if (ret) {
2559                         mlog_errno(ret);
2560                         goto out;
2561                 }
2562                 ref_ci = &ref_tree->rf_ci;
2563
2564         }
2565
2566         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2567                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2568                                                ref_ci, ref_root_bh);
2569                 if (ret < 0) {
2570                         mlog_errno(ret);
2571                         goto out;
2572                 }
2573         }
2574
2575         if (di->i_xattr_loc) {
2576                 ret = ocfs2_xattr_free_block(inode,
2577                                              le64_to_cpu(di->i_xattr_loc),
2578                                              ref_ci, ref_root_bh);
2579                 if (ret < 0) {
2580                         mlog_errno(ret);
2581                         goto out;
2582                 }
2583         }
2584
2585         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2586                                    OCFS2_INODE_UPDATE_CREDITS);
2587         if (IS_ERR(handle)) {
2588                 ret = PTR_ERR(handle);
2589                 mlog_errno(ret);
2590                 goto out;
2591         }
2592         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2593                                       OCFS2_JOURNAL_ACCESS_WRITE);
2594         if (ret) {
2595                 mlog_errno(ret);
2596                 goto out_commit;
2597         }
2598
2599         di->i_xattr_loc = 0;
2600
2601         spin_lock(&oi->ip_lock);
2602         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2603         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2604         spin_unlock(&oi->ip_lock);
2605
2606         ocfs2_journal_dirty(handle, di_bh);
2607 out_commit:
2608         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2609 out:
2610         if (ref_tree)
2611                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2612         brelse(ref_root_bh);
2613         return ret;
2614 }
2615
2616 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2617                                         struct ocfs2_dinode *di)
2618 {
2619         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2620         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2621         int free;
2622
2623         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2624                 return 0;
2625
2626         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2627                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2628                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2629         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2630                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2631                         le64_to_cpu(di->i_size);
2632         } else {
2633                 struct ocfs2_extent_list *el = &di->id2.i_list;
2634                 free = (le16_to_cpu(el->l_count) -
2635                         le16_to_cpu(el->l_next_free_rec)) *
2636                         sizeof(struct ocfs2_extent_rec);
2637         }
2638         if (free >= xattrsize)
2639                 return 1;
2640
2641         return 0;
2642 }
2643
2644 /*
2645  * ocfs2_xattr_ibody_find()
2646  *
2647  * Find extended attribute in inode block and
2648  * fill search info into struct ocfs2_xattr_search.
2649  */
2650 static int ocfs2_xattr_ibody_find(struct inode *inode,
2651                                   int name_index,
2652                                   const char *name,
2653                                   struct ocfs2_xattr_search *xs)
2654 {
2655         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2656         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2657         int ret;
2658         int has_space = 0;
2659
2660         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2661                 return 0;
2662
2663         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2664                 down_read(&oi->ip_alloc_sem);
2665                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2666                 up_read(&oi->ip_alloc_sem);
2667                 if (!has_space)
2668                         return 0;
2669         }
2670
2671         xs->xattr_bh = xs->inode_bh;
2672         xs->end = (void *)di + inode->i_sb->s_blocksize;
2673         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2674                 xs->header = (struct ocfs2_xattr_header *)
2675                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2676         else
2677                 xs->header = (struct ocfs2_xattr_header *)
2678                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2679         xs->base = (void *)xs->header;
2680         xs->here = xs->header->xh_entries;
2681
2682         /* Find the named attribute. */
2683         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2684                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2685                 if (ret && ret != -ENODATA)
2686                         return ret;
2687                 xs->not_found = ret;
2688         }
2689
2690         return 0;
2691 }
2692
2693 static int ocfs2_xattr_ibody_init(struct inode *inode,
2694                                   struct buffer_head *di_bh,
2695                                   struct ocfs2_xattr_set_ctxt *ctxt)
2696 {
2697         int ret;
2698         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2699         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2700         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2701         unsigned int xattrsize = osb->s_xattr_inline_size;
2702
2703         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2704                 ret = -ENOSPC;
2705                 goto out;
2706         }
2707
2708         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2709                                       OCFS2_JOURNAL_ACCESS_WRITE);
2710         if (ret) {
2711                 mlog_errno(ret);
2712                 goto out;
2713         }
2714
2715         /*
2716          * Adjust extent record count or inline data size
2717          * to reserve space for extended attribute.
2718          */
2719         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2720                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2721                 le16_add_cpu(&idata->id_count, -xattrsize);
2722         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2723                 struct ocfs2_extent_list *el = &di->id2.i_list;
2724                 le16_add_cpu(&el->l_count, -(xattrsize /
2725                                              sizeof(struct ocfs2_extent_rec)));
2726         }
2727         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2728
2729         spin_lock(&oi->ip_lock);
2730         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2731         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2732         spin_unlock(&oi->ip_lock);
2733
2734         ocfs2_journal_dirty(ctxt->handle, di_bh);
2735
2736 out:
2737         return ret;
2738 }
2739
2740 /*
2741  * ocfs2_xattr_ibody_set()
2742  *
2743  * Set, replace or remove an extended attribute into inode block.
2744  *
2745  */
2746 static int ocfs2_xattr_ibody_set(struct inode *inode,
2747                                  struct ocfs2_xattr_info *xi,
2748                                  struct ocfs2_xattr_search *xs,
2749                                  struct ocfs2_xattr_set_ctxt *ctxt)
2750 {
2751         int ret;
2752         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2753         struct ocfs2_xa_loc loc;
2754
2755         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2756                 return -ENOSPC;
2757
2758         down_write(&oi->ip_alloc_sem);
2759         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2760                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2761                 if (ret) {
2762                         if (ret != -ENOSPC)
2763                                 mlog_errno(ret);
2764                         goto out;
2765                 }
2766         }
2767
2768         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2769                                  xs->not_found ? NULL : xs->here);
2770         ret = ocfs2_xa_set(&loc, xi, ctxt);
2771         if (ret) {
2772                 if (ret != -ENOSPC)
2773                         mlog_errno(ret);
2774                 goto out;
2775         }
2776         xs->here = loc.xl_entry;
2777
2778 out:
2779         up_write(&oi->ip_alloc_sem);
2780
2781         return ret;
2782 }
2783
2784 /*
2785  * ocfs2_xattr_block_find()
2786  *
2787  * Find extended attribute in external block and
2788  * fill search info into struct ocfs2_xattr_search.
2789  */
2790 static int ocfs2_xattr_block_find(struct inode *inode,
2791                                   int name_index,
2792                                   const char *name,
2793                                   struct ocfs2_xattr_search *xs)
2794 {
2795         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2796         struct buffer_head *blk_bh = NULL;
2797         struct ocfs2_xattr_block *xb;
2798         int ret = 0;
2799
2800         if (!di->i_xattr_loc)
2801                 return ret;
2802
2803         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2804                                      &blk_bh);
2805         if (ret < 0) {
2806                 mlog_errno(ret);
2807                 return ret;
2808         }
2809
2810         xs->xattr_bh = blk_bh;
2811         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2812
2813         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2814                 xs->header = &xb->xb_attrs.xb_header;
2815                 xs->base = (void *)xs->header;
2816                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2817                 xs->here = xs->header->xh_entries;
2818
2819                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2820         } else
2821                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2822                                                    name_index,
2823                                                    name, xs);
2824
2825         if (ret && ret != -ENODATA) {
2826                 xs->xattr_bh = NULL;
2827                 goto cleanup;
2828         }
2829         xs->not_found = ret;
2830         return 0;
2831 cleanup:
2832         brelse(blk_bh);
2833
2834         return ret;
2835 }
2836
2837 static int ocfs2_create_xattr_block(struct inode *inode,
2838                                     struct buffer_head *inode_bh,
2839                                     struct ocfs2_xattr_set_ctxt *ctxt,
2840                                     int indexed,
2841                                     struct buffer_head **ret_bh)
2842 {
2843         int ret;
2844         u16 suballoc_bit_start;
2845         u32 num_got;
2846         u64 suballoc_loc, first_blkno;
2847         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2848         struct buffer_head *new_bh = NULL;
2849         struct ocfs2_xattr_block *xblk;
2850
2851         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2852                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2853         if (ret < 0) {
2854                 mlog_errno(ret);
2855                 goto end;
2856         }
2857
2858         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2859                                    &suballoc_loc, &suballoc_bit_start,
2860                                    &num_got, &first_blkno);
2861         if (ret < 0) {
2862                 mlog_errno(ret);
2863                 goto end;
2864         }
2865
2866         new_bh = sb_getblk(inode->i_sb, first_blkno);
2867         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2868
2869         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2870                                       new_bh,
2871                                       OCFS2_JOURNAL_ACCESS_CREATE);
2872         if (ret < 0) {
2873                 mlog_errno(ret);
2874                 goto end;
2875         }
2876
2877         /* Initialize ocfs2_xattr_block */
2878         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2879         memset(xblk, 0, inode->i_sb->s_blocksize);
2880         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2881         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2882         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2883         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2884         xblk->xb_fs_generation =
2885                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2886         xblk->xb_blkno = cpu_to_le64(first_blkno);
2887         if (indexed) {
2888                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2889                 xr->xt_clusters = cpu_to_le32(1);
2890                 xr->xt_last_eb_blk = 0;
2891                 xr->xt_list.l_tree_depth = 0;
2892                 xr->xt_list.l_count = cpu_to_le16(
2893                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2894                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2895                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2896         }
2897         ocfs2_journal_dirty(ctxt->handle, new_bh);
2898
2899         /* Add it to the inode */
2900         di->i_xattr_loc = cpu_to_le64(first_blkno);
2901
2902         spin_lock(&OCFS2_I(inode)->ip_lock);
2903         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2904         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2905         spin_unlock(&OCFS2_I(inode)->ip_lock);
2906
2907         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2908
2909         *ret_bh = new_bh;
2910         new_bh = NULL;
2911
2912 end:
2913         brelse(new_bh);
2914         return ret;
2915 }
2916
2917 /*
2918  * ocfs2_xattr_block_set()
2919  *
2920  * Set, replace or remove an extended attribute into external block.
2921  *
2922  */
2923 static int ocfs2_xattr_block_set(struct inode *inode,
2924                                  struct ocfs2_xattr_info *xi,
2925                                  struct ocfs2_xattr_search *xs,
2926                                  struct ocfs2_xattr_set_ctxt *ctxt)
2927 {
2928         struct buffer_head *new_bh = NULL;
2929         struct ocfs2_xattr_block *xblk = NULL;
2930         int ret;
2931         struct ocfs2_xa_loc loc;
2932
2933         if (!xs->xattr_bh) {
2934                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2935                                                0, &new_bh);
2936                 if (ret) {
2937                         mlog_errno(ret);
2938                         goto end;
2939                 }
2940
2941                 xs->xattr_bh = new_bh;
2942                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2943                 xs->header = &xblk->xb_attrs.xb_header;
2944                 xs->base = (void *)xs->header;
2945                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2946                 xs->here = xs->header->xh_entries;
2947         } else
2948                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2949
2950         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2951                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2952                                               xs->not_found ? NULL : xs->here);
2953
2954                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2955                 if (!ret)
2956                         xs->here = loc.xl_entry;
2957                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2958                         goto end;
2959                 else {
2960                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2961                         if (ret)
2962                                 goto end;
2963                 }
2964         }
2965
2966         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2967                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2968
2969 end:
2970         return ret;
2971 }
2972
2973 /* Check whether the new xattr can be inserted into the inode. */
2974 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2975                                        struct ocfs2_xattr_info *xi,
2976                                        struct ocfs2_xattr_search *xs)
2977 {
2978         struct ocfs2_xattr_entry *last;
2979         int free, i;
2980         size_t min_offs = xs->end - xs->base;
2981
2982         if (!xs->header)
2983                 return 0;
2984
2985         last = xs->header->xh_entries;
2986
2987         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2988                 size_t offs = le16_to_cpu(last->xe_name_offset);
2989                 if (offs < min_offs)
2990                         min_offs = offs;
2991                 last += 1;
2992         }
2993
2994         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2995         if (free < 0)
2996                 return 0;
2997
2998         BUG_ON(!xs->not_found);
2999
3000         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3001                 return 1;
3002
3003         return 0;
3004 }
3005
3006 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3007                                      struct ocfs2_dinode *di,
3008                                      struct ocfs2_xattr_info *xi,
3009                                      struct ocfs2_xattr_search *xis,
3010                                      struct ocfs2_xattr_search *xbs,
3011                                      int *clusters_need,
3012                                      int *meta_need,
3013                                      int *credits_need)
3014 {
3015         int ret = 0, old_in_xb = 0;
3016         int clusters_add = 0, meta_add = 0, credits = 0;
3017         struct buffer_head *bh = NULL;
3018         struct ocfs2_xattr_block *xb = NULL;
3019         struct ocfs2_xattr_entry *xe = NULL;
3020         struct ocfs2_xattr_value_root *xv = NULL;
3021         char *base = NULL;
3022         int name_offset, name_len = 0;
3023         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3024                                                     xi->xi_value_len);
3025         u64 value_size;
3026
3027         /*
3028          * Calculate the clusters we need to write.
3029          * No matter whether we replace an old one or add a new one,
3030          * we need this for writing.
3031          */
3032         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3033                 credits += new_clusters *
3034                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3035
3036         if (xis->not_found && xbs->not_found) {
3037                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3038
3039                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3040                         clusters_add += new_clusters;
3041                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3042                                                         &def_xv.xv.xr_list);
3043                 }
3044
3045                 goto meta_guess;
3046         }
3047
3048         if (!xis->not_found) {
3049                 xe = xis->here;
3050                 name_offset = le16_to_cpu(xe->xe_name_offset);
3051                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3052                 base = xis->base;
3053                 credits += OCFS2_INODE_UPDATE_CREDITS;
3054         } else {
3055                 int i, block_off = 0;
3056                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3057                 xe = xbs->here;
3058                 name_offset = le16_to_cpu(xe->xe_name_offset);
3059                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3060                 i = xbs->here - xbs->header->xh_entries;
3061                 old_in_xb = 1;
3062
3063                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3064                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3065                                                         bucket_xh(xbs->bucket),
3066                                                         i, &block_off,
3067                                                         &name_offset);
3068                         base = bucket_block(xbs->bucket, block_off);
3069                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3070                 } else {
3071                         base = xbs->base;
3072                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3073                 }
3074         }
3075
3076         /*
3077          * delete a xattr doesn't need metadata and cluster allocation.
3078          * so just calculate the credits and return.
3079          *
3080          * The credits for removing the value tree will be extended
3081          * by ocfs2_remove_extent itself.
3082          */
3083         if (!xi->xi_value) {
3084                 if (!ocfs2_xattr_is_local(xe))
3085                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3086
3087                 goto out;
3088         }
3089
3090         /* do cluster allocation guess first. */
3091         value_size = le64_to_cpu(xe->xe_value_size);
3092
3093         if (old_in_xb) {
3094                 /*
3095                  * In xattr set, we always try to set the xe in inode first,
3096                  * so if it can be inserted into inode successfully, the old
3097                  * one will be removed from the xattr block, and this xattr
3098                  * will be inserted into inode as a new xattr in inode.
3099                  */
3100                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3101                         clusters_add += new_clusters;
3102                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3103                                     OCFS2_INODE_UPDATE_CREDITS;
3104                         if (!ocfs2_xattr_is_local(xe))
3105                                 credits += ocfs2_calc_extend_credits(
3106                                                         inode->i_sb,
3107                                                         &def_xv.xv.xr_list);
3108                         goto out;
3109                 }
3110         }
3111
3112         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3113                 /* the new values will be stored outside. */
3114                 u32 old_clusters = 0;
3115
3116                 if (!ocfs2_xattr_is_local(xe)) {
3117                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3118                                                                  value_size);
3119                         xv = (struct ocfs2_xattr_value_root *)
3120                              (base + name_offset + name_len);
3121                         value_size = OCFS2_XATTR_ROOT_SIZE;
3122                 } else
3123                         xv = &def_xv.xv;
3124
3125                 if (old_clusters >= new_clusters) {
3126                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3127                         goto out;
3128                 } else {
3129                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3130                         clusters_add += new_clusters - old_clusters;
3131                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3132                                                              &xv->xr_list);
3133                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3134                                 goto out;
3135                 }
3136         } else {
3137                 /*
3138                  * Now the new value will be stored inside. So if the new
3139                  * value is smaller than the size of value root or the old
3140                  * value, we don't need any allocation, otherwise we have
3141                  * to guess metadata allocation.
3142                  */
3143                 if ((ocfs2_xattr_is_local(xe) &&
3144                      (value_size >= xi->xi_value_len)) ||
3145                     (!ocfs2_xattr_is_local(xe) &&
3146                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3147                         goto out;
3148         }
3149
3150 meta_guess:
3151         /* calculate metadata allocation. */
3152         if (di->i_xattr_loc) {
3153                 if (!xbs->xattr_bh) {
3154                         ret = ocfs2_read_xattr_block(inode,
3155                                                      le64_to_cpu(di->i_xattr_loc),
3156                                                      &bh);
3157                         if (ret) {
3158                                 mlog_errno(ret);
3159                                 goto out;
3160                         }
3161
3162                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3163                 } else
3164                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3165
3166                 /*
3167                  * If there is already an xattr tree, good, we can calculate
3168                  * like other b-trees. Otherwise we may have the chance of
3169                  * create a tree, the credit calculation is borrowed from
3170                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3171                  * new tree will be cluster based, so no meta is needed.
3172                  */
3173                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3174                         struct ocfs2_extent_list *el =
3175                                  &xb->xb_attrs.xb_root.xt_list;
3176                         meta_add += ocfs2_extend_meta_needed(el);
3177                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3178                                                              el);
3179                 } else
3180                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3181
3182                 /*
3183                  * This cluster will be used either for new bucket or for
3184                  * new xattr block.
3185                  * If the cluster size is the same as the bucket size, one
3186                  * more is needed since we may need to extend the bucket
3187                  * also.
3188                  */
3189                 clusters_add += 1;
3190                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3191                 if (OCFS2_XATTR_BUCKET_SIZE ==
3192                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3193                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3194                         clusters_add += 1;
3195                 }
3196         } else {
3197                 meta_add += 1;
3198                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3199         }
3200 out:
3201         if (clusters_need)
3202                 *clusters_need = clusters_add;
3203         if (meta_need)
3204                 *meta_need = meta_add;
3205         if (credits_need)
3206                 *credits_need = credits;
3207         brelse(bh);
3208         return ret;
3209 }
3210
3211 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3212                                      struct ocfs2_dinode *di,
3213                                      struct ocfs2_xattr_info *xi,
3214                                      struct ocfs2_xattr_search *xis,
3215                                      struct ocfs2_xattr_search *xbs,
3216                                      struct ocfs2_xattr_set_ctxt *ctxt,
3217                                      int extra_meta,
3218                                      int *credits)
3219 {
3220         int clusters_add, meta_add, ret;
3221         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3222
3223         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3224
3225         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3226
3227         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3228                                         &clusters_add, &meta_add, credits);
3229         if (ret) {
3230                 mlog_errno(ret);
3231                 return ret;
3232         }
3233
3234         meta_add += extra_meta;
3235         trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3236                                         clusters_add, *credits);
3237
3238         if (meta_add) {
3239                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3240                                                         &ctxt->meta_ac);
3241                 if (ret) {
3242                         mlog_errno(ret);
3243                         goto out;
3244                 }
3245         }
3246
3247         if (clusters_add) {
3248                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3249                 if (ret)
3250                         mlog_errno(ret);
3251         }
3252 out:
3253         if (ret) {
3254                 if (ctxt->meta_ac) {
3255                         ocfs2_free_alloc_context(ctxt->meta_ac);
3256                         ctxt->meta_ac = NULL;
3257                 }
3258
3259                 /*
3260                  * We cannot have an error and a non null ctxt->data_ac.
3261                  */
3262         }
3263
3264         return ret;
3265 }
3266
3267 static int __ocfs2_xattr_set_handle(struct inode *inode,
3268                                     struct ocfs2_dinode *di,
3269                                     struct ocfs2_xattr_info *xi,
3270                                     struct ocfs2_xattr_search *xis,
3271                                     struct ocfs2_xattr_search *xbs,
3272                                     struct ocfs2_xattr_set_ctxt *ctxt)
3273 {
3274         int ret = 0, credits, old_found;
3275
3276         if (!xi->xi_value) {
3277                 /* Remove existing extended attribute */
3278                 if (!xis->not_found)
3279                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3280                 else if (!xbs->not_found)
3281                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3282         } else {
3283                 /* We always try to set extended attribute into inode first*/
3284                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3285                 if (!ret && !xbs->not_found) {
3286                         /*
3287                          * If succeed and that extended attribute existing in
3288                          * external block, then we will remove it.
3289                          */
3290                         xi->xi_value = NULL;
3291                         xi->xi_value_len = 0;
3292
3293                         old_found = xis->not_found;
3294                         xis->not_found = -ENODATA;
3295                         ret = ocfs2_calc_xattr_set_need(inode,
3296                                                         di,
3297                                                         xi,
3298                                                         xis,
3299                                                         xbs,
3300                                                         NULL,
3301                                                         NULL,
3302                                                         &credits);
3303                         xis->not_found = old_found;
3304                         if (ret) {
3305                                 mlog_errno(ret);
3306                                 goto out;
3307                         }
3308
3309                         ret = ocfs2_extend_trans(ctxt->handle, credits);
3310                         if (ret) {
3311                                 mlog_errno(ret);
3312                                 goto out;
3313                         }
3314                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3315                 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3316                         if (di->i_xattr_loc && !xbs->xattr_bh) {
3317                                 ret = ocfs2_xattr_block_find(inode,
3318                                                              xi->xi_name_index,
3319                                                              xi->xi_name, xbs);
3320                                 if (ret)
3321                                         goto out;
3322
3323                                 old_found = xis->not_found;
3324                                 xis->not_found = -ENODATA;
3325                                 ret = ocfs2_calc_xattr_set_need(inode,
3326                                                                 di,
3327                                                                 xi,
3328                                                                 xis,
3329                                                                 xbs,
3330                                                                 NULL,
3331                                                                 NULL,
3332                                                                 &credits);
3333                                 xis->not_found = old_found;
3334                                 if (ret) {
3335                                         mlog_errno(ret);
3336                                         goto out;
3337                                 }
3338
3339                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3340                                 if (ret) {
3341                                         mlog_errno(ret);
3342                                         goto out;
3343                                 }
3344                         }
3345                         /*
3346                          * If no space in inode, we will set extended attribute
3347                          * into external block.
3348                          */
3349                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3350                         if (ret)
3351                                 goto out;
3352                         if (!xis->not_found) {
3353                                 /*
3354                                  * If succeed and that extended attribute
3355                                  * existing in inode, we will remove it.
3356                                  */
3357                                 xi->xi_value = NULL;
3358                                 xi->xi_value_len = 0;
3359                                 xbs->not_found = -ENODATA;
3360                                 ret = ocfs2_calc_xattr_set_need(inode,
3361                                                                 di,
3362                                                                 xi,
3363                                                                 xis,
3364                                                                 xbs,
3365                                                                 NULL,
3366                                                                 NULL,
3367                                                                 &credits);
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                                 ret = ocfs2_xattr_ibody_set(inode, xi,
3379                                                             xis, ctxt);
3380                         }
3381                 }
3382         }
3383
3384         if (!ret) {
3385                 /* Update inode ctime. */
3386                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3387                                               xis->inode_bh,
3388                                               OCFS2_JOURNAL_ACCESS_WRITE);
3389                 if (ret) {
3390                         mlog_errno(ret);
3391                         goto out;
3392                 }
3393
3394                 inode->i_ctime = CURRENT_TIME;
3395                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3396                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3397                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3398         }
3399 out:
3400         return ret;
3401 }
3402
3403 /*
3404  * This function only called duing creating inode
3405  * for init security/acl xattrs of the new inode.
3406  * All transanction credits have been reserved in mknod.
3407  */
3408 int ocfs2_xattr_set_handle(handle_t *handle,
3409                            struct inode *inode,
3410                            struct buffer_head *di_bh,
3411                            int name_index,
3412                            const char *name,
3413                            const void *value,
3414                            size_t value_len,
3415                            int flags,
3416                            struct ocfs2_alloc_context *meta_ac,
3417                            struct ocfs2_alloc_context *data_ac)
3418 {
3419         struct ocfs2_dinode *di;
3420         int ret;
3421
3422         struct ocfs2_xattr_info xi = {
3423                 .xi_name_index = name_index,
3424                 .xi_name = name,
3425                 .xi_name_len = strlen(name),
3426                 .xi_value = value,
3427                 .xi_value_len = value_len,
3428         };
3429
3430         struct ocfs2_xattr_search xis = {
3431                 .not_found = -ENODATA,
3432         };
3433
3434         struct ocfs2_xattr_search xbs = {
3435                 .not_found = -ENODATA,
3436         };
3437
3438         struct ocfs2_xattr_set_ctxt ctxt = {
3439                 .handle = handle,
3440                 .meta_ac = meta_ac,
3441                 .data_ac = data_ac,
3442         };
3443
3444         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3445                 return -EOPNOTSUPP;
3446
3447         /*
3448          * In extreme situation, may need xattr bucket when
3449          * block size is too small. And we have already reserved
3450          * the credits for bucket in mknod.
3451          */
3452         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3453                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3454                 if (!xbs.bucket) {
3455                         mlog_errno(-ENOMEM);
3456                         return -ENOMEM;
3457                 }
3458         }
3459
3460         xis.inode_bh = xbs.inode_bh = di_bh;
3461         di = (struct ocfs2_dinode *)di_bh->b_data;
3462
3463         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3464
3465         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3466         if (ret)
3467                 goto cleanup;
3468         if (xis.not_found) {
3469                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3470                 if (ret)
3471                         goto cleanup;
3472         }
3473
3474         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3475
3476 cleanup:
3477         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3478         brelse(xbs.xattr_bh);
3479         ocfs2_xattr_bucket_free(xbs.bucket);
3480
3481         return ret;
3482 }
3483
3484 /*
3485  * ocfs2_xattr_set()
3486  *
3487  * Set, replace or remove an extended attribute for this inode.
3488  * value is NULL to remove an existing extended attribute, else either
3489  * create or replace an extended attribute.
3490  */
3491 int ocfs2_xattr_set(struct inode *inode,
3492                     int name_index,
3493                     const char *name,
3494                     const void *value,
3495                     size_t value_len,
3496                     int flags)
3497 {
3498         struct buffer_head *di_bh = NULL;
3499         struct ocfs2_dinode *di;
3500         int ret, credits, ref_meta = 0, ref_credits = 0;
3501         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3502         struct inode *tl_inode = osb->osb_tl_inode;
3503         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3504         struct ocfs2_refcount_tree *ref_tree = NULL;
3505
3506         struct ocfs2_xattr_info xi = {
3507                 .xi_name_index = name_index,
3508                 .xi_name = name,
3509                 .xi_name_len = strlen(name),
3510                 .xi_value = value,
3511                 .xi_value_len = value_len,
3512         };
3513
3514         struct ocfs2_xattr_search xis = {
3515                 .not_found = -ENODATA,
3516         };
3517
3518         struct ocfs2_xattr_search xbs = {
3519                 .not_found = -ENODATA,
3520         };
3521
3522         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3523                 return -EOPNOTSUPP;
3524
3525         /*
3526          * Only xbs will be used on indexed trees.  xis doesn't need a
3527          * bucket.
3528          */
3529         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3530         if (!xbs.bucket) {
3531                 mlog_errno(-ENOMEM);
3532                 return -ENOMEM;
3533         }
3534
3535         ret = ocfs2_inode_lock(inode, &di_bh, 1);
3536         if (ret < 0) {
3537                 mlog_errno(ret);
3538                 goto cleanup_nolock;
3539         }
3540         xis.inode_bh = xbs.inode_bh = di_bh;
3541         di = (struct ocfs2_dinode *)di_bh->b_data;
3542
3543         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3544         /*
3545          * Scan inode and external block to find the same name
3546          * extended attribute and collect search information.
3547          */
3548         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3549         if (ret)
3550                 goto cleanup;
3551         if (xis.not_found) {
3552                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3553                 if (ret)
3554                         goto cleanup;
3555         }
3556
3557         if (xis.not_found && xbs.not_found) {
3558                 ret = -ENODATA;
3559                 if (flags & XATTR_REPLACE)
3560                         goto cleanup;
3561                 ret = 0;
3562                 if (!value)
3563                         goto cleanup;
3564         } else {
3565                 ret = -EEXIST;
3566                 if (flags & XATTR_CREATE)
3567                         goto cleanup;
3568         }
3569
3570         /* Check whether the value is refcounted and do some preparation. */
3571         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3572             (!xis.not_found || !xbs.not_found)) {
3573                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3574                                                    &xis, &xbs, &ref_tree,
3575                                                    &ref_meta, &ref_credits);
3576                 if (ret) {
3577                         mlog_errno(ret);
3578                         goto cleanup;
3579                 }
3580         }
3581
3582         mutex_lock(&tl_inode->i_mutex);
3583
3584         if (ocfs2_truncate_log_needs_flush(osb)) {
3585                 ret = __ocfs2_flush_truncate_log(osb);
3586                 if (ret < 0) {
3587                         mutex_unlock(&tl_inode->i_mutex);
3588                         mlog_errno(ret);
3589                         goto cleanup;
3590                 }
3591         }
3592         mutex_unlock(&tl_inode->i_mutex);
3593
3594         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3595                                         &xbs, &ctxt, ref_meta, &credits);
3596         if (ret) {
3597                 mlog_errno(ret);
3598                 goto cleanup;
3599         }
3600
3601         /* we need to update inode's ctime field, so add credit for it. */
3602         credits += OCFS2_INODE_UPDATE_CREDITS;
3603         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3604         if (IS_ERR(ctxt.handle)) {
3605                 ret = PTR_ERR(ctxt.handle);
3606                 mlog_errno(ret);
3607                 goto out_free_ac;
3608         }
3609
3610         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3611
3612         ocfs2_commit_trans(osb, ctxt.handle);
3613
3614 out_free_ac:
3615         if (ctxt.data_ac)
3616                 ocfs2_free_alloc_context(ctxt.data_ac);
3617         if (ctxt.meta_ac)
3618                 ocfs2_free_alloc_context(ctxt.meta_ac);
3619         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3620                 ocfs2_schedule_truncate_log_flush(osb, 1);
3621         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3622
3623 cleanup:
3624         if (ref_tree)
3625                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3626         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3627         if (!value && !ret) {
3628                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3629                 if (ret)
3630                         mlog_errno(ret);
3631         }
3632         ocfs2_inode_unlock(inode, 1);
3633 cleanup_nolock:
3634         brelse(di_bh);
3635         brelse(xbs.xattr_bh);
3636         ocfs2_xattr_bucket_free(xbs.bucket);
3637
3638         return ret;
3639 }
3640
3641 /*
3642  * Find the xattr extent rec which may contains name_hash.
3643  * e_cpos will be the first name hash of the xattr rec.
3644  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3645  */
3646 static int ocfs2_xattr_get_rec(struct inode *inode,
3647                                u32 name_hash,
3648                                u64 *p_blkno,
3649                                u32 *e_cpos,
3650                                u32 *num_clusters,
3651                                struct ocfs2_extent_list *el)
3652 {
3653         int ret = 0, i;
3654         struct buffer_head *eb_bh = NULL;
3655         struct ocfs2_extent_block *eb;
3656         struct ocfs2_extent_rec *rec = NULL;
3657         u64 e_blkno = 0;
3658
3659         if (el->l_tree_depth) {
3660                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3661                                       &eb_bh);
3662                 if (ret) {
3663                         mlog_errno(ret);
3664                         goto out;
3665                 }
3666
3667                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3668                 el = &eb->h_list;
3669
3670                 if (el->l_tree_depth) {
3671                         ocfs2_error(inode->i_sb,
3672                                     "Inode %lu has non zero tree depth in "
3673                                     "xattr tree block %llu\n", inode->i_ino,
3674                                     (unsigned long long)eb_bh->b_blocknr);
3675                         ret = -EROFS;
3676                         goto out;
3677                 }
3678         }
3679
3680         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3681                 rec = &el->l_recs[i];
3682
3683                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3684                         e_blkno = le64_to_cpu(rec->e_blkno);
3685                         break;
3686                 }
3687         }
3688
3689         if (!e_blkno) {
3690                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3691                             "record (%u, %u, 0) in xattr", inode->i_ino,
3692                             le32_to_cpu(rec->e_cpos),
3693                             ocfs2_rec_clusters(el, rec));
3694                 ret = -EROFS;
3695                 goto out;
3696         }
3697
3698         *p_blkno = le64_to_cpu(rec->e_blkno);
3699         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3700         if (e_cpos)
3701                 *e_cpos = le32_to_cpu(rec->e_cpos);
3702 out:
3703         brelse(eb_bh);
3704         return ret;
3705 }
3706
3707 typedef int (xattr_bucket_func)(struct inode *inode,
3708                                 struct ocfs2_xattr_bucket *bucket,
3709                                 void *para);
3710
3711 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3712                                    struct ocfs2_xattr_bucket *bucket,
3713                                    int name_index,
3714                                    const char *name,
3715                                    u32 name_hash,
3716                                    u16 *xe_index,
3717                                    int *found)
3718 {
3719         int i, ret = 0, cmp = 1, block_off, new_offset;
3720         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3721         size_t name_len = strlen(name);
3722         struct ocfs2_xattr_entry *xe = NULL;
3723         char *xe_name;
3724
3725         /*
3726          * We don't use binary search in the bucket because there
3727          * may be multiple entries with the same name hash.
3728          */
3729         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3730                 xe = &xh->xh_entries[i];
3731
3732                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3733                         continue;
3734                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3735                         break;
3736
3737                 cmp = name_index - ocfs2_xattr_get_type(xe);
3738                 if (!cmp)
3739                         cmp = name_len - xe->xe_name_len;
3740                 if (cmp)
3741                         continue;
3742
3743                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3744                                                         xh,
3745                                                         i,
3746                                                         &block_off,
3747                                                         &new_offset);
3748                 if (ret) {
3749                         mlog_errno(ret);
3750                         break;
3751                 }
3752
3753
3754                 xe_name = bucket_block(bucket, block_off) + new_offset;
3755                 if (!memcmp(name, xe_name, name_len)) {
3756                         *xe_index = i;
3757                         *found = 1;
3758                         ret = 0;
3759                         break;
3760                 }
3761         }
3762
3763         return ret;
3764 }
3765
3766 /*
3767  * Find the specified xattr entry in a series of buckets.
3768  * This series start from p_blkno and last for num_clusters.
3769  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3770  * the num of the valid buckets.
3771  *
3772  * Return the buffer_head this xattr should reside in. And if the xattr's
3773  * hash is in the gap of 2 buckets, return the lower bucket.
3774  */
3775 static int ocfs2_xattr_bucket_find(struct inode *inode,
3776                                    int name_index,
3777                                    const char *name,
3778                                    u32 name_hash,
3779                                    u64 p_blkno,
3780                                    u32 first_hash,
3781                                    u32 num_clusters,
3782                                    struct ocfs2_xattr_search *xs)
3783 {
3784         int ret, found = 0;
3785         struct ocfs2_xattr_header *xh = NULL;
3786         struct ocfs2_xattr_entry *xe = NULL;
3787         u16 index = 0;
3788         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3789         int low_bucket = 0, bucket, high_bucket;
3790         struct ocfs2_xattr_bucket *search;
3791         u32 last_hash;
3792         u64 blkno, lower_blkno = 0;
3793
3794         search = ocfs2_xattr_bucket_new(inode);
3795         if (!search) {
3796                 ret = -ENOMEM;
3797                 mlog_errno(ret);
3798                 goto out;
3799         }
3800
3801         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3802         if (ret) {
3803                 mlog_errno(ret);
3804                 goto out;
3805         }
3806
3807         xh = bucket_xh(search);
3808         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3809         while (low_bucket <= high_bucket) {
3810                 ocfs2_xattr_bucket_relse(search);
3811
3812                 bucket = (low_bucket + high_bucket) / 2;
3813                 blkno = p_blkno + bucket * blk_per_bucket;
3814                 ret = ocfs2_read_xattr_bucket(search, blkno);
3815                 if (ret) {
3816                         mlog_errno(ret);
3817                         goto out;
3818                 }
3819
3820                 xh = bucket_xh(search);
3821                 xe = &xh->xh_entries[0];
3822                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3823                         high_bucket = bucket - 1;
3824                         continue;
3825                 }
3826
3827                 /*
3828                  * Check whether the hash of the last entry in our
3829                  * bucket is larger than the search one. for an empty
3830                  * bucket, the last one is also the first one.
3831                  */
3832                 if (xh->xh_count)
3833                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3834
3835                 last_hash = le32_to_cpu(xe->xe_name_hash);
3836
3837                 /* record lower_blkno which may be the insert place. */
3838                 lower_blkno = blkno;
3839
3840                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3841                         low_bucket = bucket + 1;
3842                         continue;
3843                 }
3844
3845                 /* the searched xattr should reside in this bucket if exists. */
3846                 ret = ocfs2_find_xe_in_bucket(inode, search,
3847                                               name_index, name, name_hash,
3848                                               &index, &found);
3849                 if (ret) {
3850                         mlog_errno(ret);
3851                         goto out;
3852                 }
3853                 break;
3854         }
3855
3856         /*
3857          * Record the bucket we have found.
3858          * When the xattr's hash value is in the gap of 2 buckets, we will
3859          * always set it to the previous bucket.
3860          */
3861         if (!lower_blkno)
3862                 lower_blkno = p_blkno;
3863
3864         /* This should be in cache - we just read it during the search */
3865         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3866         if (ret) {
3867                 mlog_errno(ret);
3868                 goto out;
3869         }
3870
3871         xs->header = bucket_xh(xs->bucket);
3872         xs->base = bucket_block(xs->bucket, 0);
3873         xs->end = xs->base + inode->i_sb->s_blocksize;
3874
3875         if (found) {
3876                 xs->here = &xs->header->xh_entries[index];
3877                 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3878                         name, name_index, name_hash,
3879                         (unsigned long long)bucket_blkno(xs->bucket),
3880                         index);
3881         } else
3882                 ret = -ENODATA;
3883
3884 out:
3885         ocfs2_xattr_bucket_free(search);
3886         return ret;
3887 }
3888
3889 static int ocfs2_xattr_index_block_find(struct inode *inode,
3890                                         struct buffer_head *root_bh,
3891                                         int name_index,
3892                                         const char *name,
3893                                         struct ocfs2_xattr_search *xs)
3894 {
3895         int ret;
3896         struct ocfs2_xattr_block *xb =
3897                         (struct ocfs2_xattr_block *)root_bh->b_data;
3898         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3899         struct ocfs2_extent_list *el = &xb_root->xt_list;
3900         u64 p_blkno = 0;
3901         u32 first_hash, num_clusters = 0;
3902         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3903
3904         if (le16_to_cpu(el->l_next_free_rec) == 0)
3905                 return -ENODATA;
3906
3907         trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3908                                         name, name_index, name_hash,
3909                                         (unsigned long long)root_bh->b_blocknr,
3910                                         -1);
3911
3912         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3913                                   &num_clusters, el);
3914         if (ret) {
3915                 mlog_errno(ret);
3916                 goto out;
3917         }
3918
3919         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3920
3921         trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3922                                         name, name_index, first_hash,
3923                                         (unsigned long long)p_blkno,
3924                                         num_clusters);
3925
3926         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3927                                       p_blkno, first_hash, num_clusters, xs);
3928
3929 out:
3930         return ret;
3931 }
3932
3933 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3934                                        u64 blkno,
3935                                        u32 clusters,
3936                                        xattr_bucket_func *func,
3937                                        void *para)
3938 {
3939         int i, ret = 0;
3940         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3941         u32 num_buckets = clusters * bpc;
3942         struct ocfs2_xattr_bucket *bucket;
3943
3944         bucket = ocfs2_xattr_bucket_new(inode);
3945         if (!bucket) {
3946                 mlog_errno(-ENOMEM);
3947                 return -ENOMEM;
3948         }
3949
3950         trace_ocfs2_iterate_xattr_buckets(
3951                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3952                 (unsigned long long)blkno, clusters);
3953
3954         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3955                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3956                 if (ret) {
3957                         mlog_errno(ret);
3958                         break;
3959                 }
3960
3961                 /*
3962                  * The real bucket num in this series of blocks is stored
3963                  * in the 1st bucket.
3964                  */
3965                 if (i == 0)
3966                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3967
3968                 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
3969                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3970                 if (func) {
3971                         ret = func(inode, bucket, para);
3972                         if (ret && ret != -ERANGE)
3973                                 mlog_errno(ret);
3974                         /* Fall through to bucket_relse() */
3975                 }
3976
3977                 ocfs2_xattr_bucket_relse(bucket);
3978                 if (ret)
3979                         break;
3980         }
3981
3982         ocfs2_xattr_bucket_free(bucket);
3983         return ret;
3984 }
3985
3986 struct ocfs2_xattr_tree_list {
3987         char *buffer;
3988         size_t buffer_size;
3989         size_t result;
3990 };
3991
3992 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3993                                              struct ocfs2_xattr_header *xh,
3994                                              int index,
3995                                              int *block_off,
3996                                              int *new_offset)
3997 {
3998         u16 name_offset;
3999
4000         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4001                 return -EINVAL;
4002
4003         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4004
4005         *block_off = name_offset >> sb->s_blocksize_bits;
4006         *new_offset = name_offset % sb->s_blocksize;
4007
4008         return 0;
4009 }
4010
4011 static int ocfs2_list_xattr_bucket(struct inode *inode,
4012                                    struct ocfs2_xattr_bucket *bucket,
4013                                    void *para)
4014 {
4015         int ret = 0, type;
4016         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4017         int i, block_off, new_offset;
4018         const char *prefix, *name;
4019
4020         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4021                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4022                 type = ocfs2_xattr_get_type(entry);
4023                 prefix = ocfs2_xattr_prefix(type);
4024
4025                 if (prefix) {
4026                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4027                                                                 bucket_xh(bucket),
4028                                                                 i,
4029                                                                 &block_off,
4030                                                                 &new_offset);
4031                         if (ret)
4032                                 break;
4033
4034                         name = (const char *)bucket_block(bucket, block_off) +
4035                                 new_offset;
4036                         ret = ocfs2_xattr_list_entry(xl->buffer,
4037                                                      xl->buffer_size,
4038                                                      &xl->result,
4039                                                      prefix, name,
4040                                                      entry->xe_name_len);
4041                         if (ret)
4042                                 break;
4043                 }
4044         }
4045
4046         return ret;
4047 }
4048
4049 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4050                                            struct buffer_head *blk_bh,
4051                                            xattr_tree_rec_func *rec_func,
4052                                            void *para)
4053 {
4054         struct ocfs2_xattr_block *xb =
4055                         (struct ocfs2_xattr_block *)blk_bh->b_data;
4056         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4057         int ret = 0;
4058         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4059         u64 p_blkno = 0;
4060
4061         if (!el->l_next_free_rec || !rec_func)
4062                 return 0;
4063
4064         while (name_hash > 0) {
4065                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4066                                           &e_cpos, &num_clusters, el);
4067                 if (ret) {
4068                         mlog_errno(ret);
4069                         break;
4070                 }
4071
4072                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4073                                num_clusters, para);
4074                 if (ret) {
4075                         if (ret != -ERANGE)
4076                                 mlog_errno(ret);
4077                         break;
4078                 }
4079
4080                 if (e_cpos == 0)
4081                         break;
4082
4083                 name_hash = e_cpos - 1;
4084         }
4085
4086         return ret;
4087
4088 }
4089
4090 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4091                                      struct buffer_head *root_bh,
4092                                      u64 blkno, u32 cpos, u32 len, void *para)
4093 {
4094         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4095                                            ocfs2_list_xattr_bucket, para);
4096 }
4097
4098 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4099                                              struct buffer_head *blk_bh,
4100                                              char *buffer,
4101                                              size_t buffer_size)
4102 {
4103         int ret;
4104         struct ocfs2_xattr_tree_list xl = {
4105                 .buffer = buffer,
4106                 .buffer_size = buffer_size,
4107                 .result = 0,
4108         };
4109
4110         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4111                                               ocfs2_list_xattr_tree_rec, &xl);
4112         if (ret) {
4113                 mlog_errno(ret);
4114                 goto out;
4115         }
4116
4117         ret = xl.result;
4118 out:
4119         return ret;
4120 }
4121
4122 static int cmp_xe(const void *a, const void *b)
4123 {
4124         const struct ocfs2_xattr_entry *l = a, *r = b;
4125         u32 l_hash = le32_to_cpu(l->xe_name_hash);
4126         u32 r_hash = le32_to_cpu(r->xe_name_hash);
4127
4128         if (l_hash > r_hash)
4129                 return 1;
4130         if (l_hash < r_hash)
4131                 return -1;
4132         return 0;
4133 }
4134
4135 static void swap_xe(void *a, void *b, int size)
4136 {
4137         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4138
4139         tmp = *l;
4140         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4141         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4142 }
4143
4144 /*
4145  * When the ocfs2_xattr_block is filled up, new bucket will be created
4146  * and all the xattr entries will be moved to the new bucket.
4147  * The header goes at the start of the bucket, and the names+values are
4148  * filled from the end.  This is why *target starts as the last buffer.
4149  * Note: we need to sort the entries since they are not saved in order
4150  * in the ocfs2_xattr_block.
4151  */
4152 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4153                                            struct buffer_head *xb_bh,
4154                                            struct ocfs2_xattr_bucket *bucket)
4155 {
4156         int i, blocksize = inode->i_sb->s_blocksize;
4157         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4158         u16 offset, size, off_change;
4159         struct ocfs2_xattr_entry *xe;
4160         struct ocfs2_xattr_block *xb =
4161                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
4162         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4163         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4164         u16 count = le16_to_cpu(xb_xh->xh_count);
4165         char *src = xb_bh->b_data;
4166         char *target = bucket_block(bucket, blks - 1);
4167
4168         trace_ocfs2_cp_xattr_block_to_bucket_begin(
4169                                 (unsigned long long)xb_bh->b_blocknr,
4170                                 (unsigned long long)bucket_blkno(bucket));
4171
4172         for (i = 0; i < blks; i++)
4173                 memset(bucket_block(bucket, i), 0, blocksize);
4174
4175         /*
4176          * Since the xe_name_offset is based on ocfs2_xattr_header,
4177          * there is a offset change corresponding to the change of
4178          * ocfs2_xattr_header's position.
4179          */
4180         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4181         xe = &xb_xh->xh_entries[count - 1];
4182         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4183         size = blocksize - offset;
4184
4185         /* copy all the names and values. */
4186         memcpy(target + offset, src + offset, size);
4187
4188         /* Init new header now. */
4189         xh->xh_count = xb_xh->xh_count;
4190         xh->xh_num_buckets = cpu_to_le16(1);
4191         xh->xh_name_value_len = cpu_to_le16(size);
4192         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4193
4194         /* copy all the entries. */
4195         target = bucket_block(bucket, 0);
4196         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4197         size = count * sizeof(struct ocfs2_xattr_entry);
4198         memcpy(target + offset, (char *)xb_xh + offset, size);
4199
4200         /* Change the xe offset for all the xe because of the move. */
4201         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4202                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4203         for (i = 0; i < count; i++)
4204                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4205
4206         trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4207
4208         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4209              cmp_xe, swap_xe);
4210 }
4211
4212 /*
4213  * After we move xattr from block to index btree, we have to
4214  * update ocfs2_xattr_search to the new xe and base.
4215  *
4216  * When the entry is in xattr block, xattr_bh indicates the storage place.
4217  * While if the entry is in index b-tree, "bucket" indicates the
4218  * real place of the xattr.
4219  */
4220 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4221                                             struct ocfs2_xattr_search *xs,
4222                                             struct buffer_head *old_bh)
4223 {
4224         char *buf = old_bh->b_data;
4225         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4226         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4227         int i;
4228
4229         xs->header = bucket_xh(xs->bucket);
4230         xs->base = bucket_block(xs->bucket, 0);
4231         xs->end = xs->base + inode->i_sb->s_blocksize;
4232
4233         if (xs->not_found)
4234                 return;
4235
4236         i = xs->here - old_xh->xh_entries;
4237         xs->here = &xs->header->xh_entries[i];
4238 }
4239
4240 static int ocfs2_xattr_create_index_block(struct inode *inode,
4241                                           struct ocfs2_xattr_search *xs,
4242                                           struct ocfs2_xattr_set_ctxt *ctxt)
4243 {
4244         int ret;
4245         u32 bit_off, len;
4246         u64 blkno;
4247         handle_t *handle = ctxt->handle;
4248         struct ocfs2_inode_info *oi = OCFS2_I(inode);
4249         struct buffer_head *xb_bh = xs->xattr_bh;
4250         struct ocfs2_xattr_block *xb =
4251                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4252         struct ocfs2_xattr_tree_root *xr;
4253         u16 xb_flags = le16_to_cpu(xb->xb_flags);
4254
4255         trace_ocfs2_xattr_create_index_block_begin(
4256                                 (unsigned long long)xb_bh->b_blocknr);
4257
4258         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4259         BUG_ON(!xs->bucket);
4260
4261         /*
4262          * XXX:
4263          * We can use this lock for now, and maybe move to a dedicated mutex
4264          * if performance becomes a problem later.
4265          */
4266         down_write(&oi->ip_alloc_sem);
4267
4268         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4269                                       OCFS2_JOURNAL_ACCESS_WRITE);
4270         if (ret) {
4271                 mlog_errno(ret);
4272                 goto out;
4273         }
4274
4275         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4276                                      1, 1, &bit_off, &len);
4277         if (ret) {
4278                 mlog_errno(ret);
4279                 goto out;
4280         }
4281
4282         /*
4283          * The bucket may spread in many blocks, and
4284          * we will only touch the 1st block and the last block
4285          * in the whole bucket(one for entry and one for data).
4286          */
4287         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4288
4289         trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4290
4291         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
4292         if (ret) {
4293                 mlog_errno(ret);
4294                 goto out;
4295         }
4296
4297         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4298                                                 OCFS2_JOURNAL_ACCESS_CREATE);
4299         if (ret) {
4300                 mlog_errno(ret);
4301                 goto out;
4302         }
4303
4304         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4305         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4306
4307         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4308
4309         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4310         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4311                offsetof(struct ocfs2_xattr_block, xb_attrs));
4312
4313         xr = &xb->xb_attrs.xb_root;
4314         xr->xt_clusters = cpu_to_le32(1);
4315         xr->xt_last_eb_blk = 0;
4316         xr->xt_list.l_tree_depth = 0;
4317         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4318         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4319
4320         xr->xt_list.l_recs[0].e_cpos = 0;
4321         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4322         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4323
4324         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4325
4326         ocfs2_journal_dirty(handle, xb_bh);
4327
4328 out:
4329         up_write(&oi->ip_alloc_sem);
4330
4331         return ret;
4332 }
4333
4334 static int cmp_xe_offset(const void *a, const void *b)
4335 {
4336         const struct ocfs2_xattr_entry *l = a, *r = b;
4337         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4338         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4339
4340         if (l_name_offset < r_name_offset)
4341                 return 1;
4342         if (l_name_offset > r_name_offset)
4343                 return -1;
4344         return 0;
4345 }
4346
4347 /*
4348  * defrag a xattr bucket if we find that the bucket has some
4349  * holes beteen name/value pairs.
4350  * We will move all the name/value pairs to the end of the bucket
4351  * so that we can spare some space for insertion.
4352  */
4353 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4354                                      handle_t *handle,
4355                                      struct ocfs2_xattr_bucket *bucket)
4356 {
4357         int ret, i;
4358         size_t end, offset, len;
4359         struct ocfs2_xattr_header *xh;
4360         char *entries, *buf, *bucket_buf = NULL;
4361         u64 blkno = bucket_blkno(bucket);
4362         u16 xh_free_start;
4363         size_t blocksize = inode->i_sb->s_blocksize;
4364         struct ocfs2_xattr_entry *xe;
4365
4366         /*
4367          * In order to make the operation more efficient and generic,
4368          * we copy all the blocks into a contiguous memory and do the
4369          * defragment there, so if anything is error, we will not touch
4370          * the real block.
4371          */
4372         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4373         if (!bucket_buf) {
4374                 ret = -EIO;
4375                 goto out;
4376         }
4377
4378         buf = bucket_buf;
4379         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4380                 memcpy(buf, bucket_block(bucket, i), blocksize);
4381
4382         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4383                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4384         if (ret < 0) {
4385                 mlog_errno(ret);
4386                 goto out;
4387         }
4388
4389         xh = (struct ocfs2_xattr_header *)bucket_buf;
4390         entries = (char *)xh->xh_entries;
4391         xh_free_start = le16_to_cpu(xh->xh_free_start);
4392
4393         trace_ocfs2_defrag_xattr_bucket(
4394              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4395              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4396
4397         /*
4398          * sort all the entries by their offset.
4399          * the largest will be the first, so that we can
4400          * move them to the end one by one.
4401          */
4402         sort(entries, le16_to_cpu(xh->xh_count),
4403              sizeof(struct ocfs2_xattr_entry),
4404              cmp_xe_offset, swap_xe);
4405
4406         /* Move all name/values to the end of the bucket. */
4407         xe = xh->xh_entries;
4408         end = OCFS2_XATTR_BUCKET_SIZE;
4409         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4410                 offset = le16_to_cpu(xe->xe_name_offset);
4411                 len = namevalue_size_xe(xe);
4412
4413                 /*
4414                  * We must make sure that the name/value pair
4415                  * exist in the same block. So adjust end to
4416                  * the previous block end if needed.
4417                  */
4418                 if (((end - len) / blocksize !=
4419                         (end - 1) / blocksize))
4420                         end = end - end % blocksize;
4421
4422                 if (end > offset + len) {
4423                         memmove(bucket_buf + end - len,
4424                                 bucket_buf + offset, len);
4425                         xe->xe_name_offset = cpu_to_le16(end - len);
4426                 }
4427
4428                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4429                                 "bucket %llu\n", (unsigned long long)blkno);
4430
4431                 end -= len;
4432         }
4433
4434         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4435                         "bucket %llu\n", (unsigned long long)blkno);
4436
4437         if (xh_free_start == end)
4438                 goto out;
4439
4440         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4441         xh->xh_free_start = cpu_to_le16(end);
4442
4443         /* sort the entries by their name_hash. */
4444         sort(entries, le16_to_cpu(xh->xh_count),
4445              sizeof(struct ocfs2_xattr_entry),
4446              cmp_xe, swap_xe);
4447
4448         buf = bucket_buf;
4449         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4450                 memcpy(bucket_block(bucket, i), buf, blocksize);
4451         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4452
4453 out:
4454         kfree(bucket_buf);
4455         return ret;
4456 }
4457
4458 /*
4459  * prev_blkno points to the start of an existing extent.  new_blkno
4460  * points to a newly allocated extent.  Because we know each of our
4461  * clusters contains more than bucket, we can easily split one cluster
4462  * at a bucket boundary.  So we take the last cluster of the existing
4463  * extent and split it down the middle.  We move the last half of the
4464  * buckets in the last cluster of the existing extent over to the new
4465  * extent.
4466  *
4467  * first_bh is the buffer at prev_blkno so we can update the existing
4468  * extent's bucket count.  header_bh is the bucket were we were hoping
4469  * to insert our xattr.  If the bucket move places the target in the new
4470  * extent, we'll update first_bh and header_bh after modifying the old
4471  * extent.
4472  *
4473  * first_hash will be set as the 1st xe's name_hash in the new extent.
4474  */
4475 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4476                                                handle_t *handle,
4477                                                struct ocfs2_xattr_bucket *first,
4478                                                struct ocfs2_xattr_bucket *target,
4479                                                u64 new_blkno,
4480                                                u32 num_clusters,
4481                                                u32 *first_hash)
4482 {
4483         int ret;
4484         struct super_block *sb = inode->i_sb;
4485         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4486         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4487         int to_move = num_buckets / 2;
4488         u64 src_blkno;
4489         u64 last_cluster_blkno = bucket_blkno(first) +
4490                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4491
4492         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4493         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4494
4495         trace_ocfs2_mv_xattr_bucket_cross_cluster(
4496                                 (unsigned long long)last_cluster_blkno,
4497                                 (unsigned long long)new_blkno);
4498
4499         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4500                                      last_cluster_blkno, new_blkno,
4501                                      to_move, first_hash);
4502         if (ret) {
4503                 mlog_errno(ret);
4504                 goto out;
4505         }
4506
4507         /* This is the first bucket that got moved */
4508         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4509
4510         /*
4511          * If the target bucket was part of the moved buckets, we need to
4512          * update first and target.
4513          */
4514         if (bucket_blkno(target) >= src_blkno) {
4515                 /* Find the block for the new target bucket */
4516                 src_blkno = new_blkno +
4517                         (bucket_blkno(target) - src_blkno);
4518
4519                 ocfs2_xattr_bucket_relse(first);
4520                 ocfs2_xattr_bucket_relse(target);
4521
4522                 /*
4523                  * These shouldn't fail - the buffers are in the
4524                  * journal from ocfs2_cp_xattr_bucket().
4525                  */
4526                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4527                 if (ret) {
4528                         mlog_errno(ret);
4529                         goto out;
4530                 }
4531                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4532                 if (ret)
4533                         mlog_errno(ret);
4534
4535         }
4536
4537 out:
4538         return ret;
4539 }
4540
4541 /*
4542  * Find the suitable pos when we divide a bucket into 2.
4543  * We have to make sure the xattrs with the same hash value exist
4544  * in the same bucket.
4545  *
4546  * If this ocfs2_xattr_header covers more than one hash value, find a
4547  * place where the hash value changes.  Try to find the most even split.
4548  * The most common case is that all entries have different hash values,
4549  * and the first check we make will find a place to split.
4550  */
4551 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4552 {
4553         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4554         int count = le16_to_cpu(xh->xh_count);
4555         int delta, middle = count / 2;
4556
4557         /*
4558          * We start at the middle.  Each step gets farther away in both
4559          * directions.  We therefore hit the change in hash value
4560          * nearest to the middle.  Note that this loop does not execute for
4561          * count < 2.
4562          */
4563         for (delta = 0; delta < middle; delta++) {
4564                 /* Let's check delta earlier than middle */
4565                 if (cmp_xe(&entries[middle - delta - 1],
4566                            &entries[middle - delta]))
4567                         return middle - delta;
4568
4569                 /* For even counts, don't walk off the end */
4570                 if ((middle + delta + 1) == count)
4571                         continue;
4572
4573                 /* Now try delta past middle */
4574                 if (cmp_xe(&entries[middle + delta],
4575                            &entries[middle + delta + 1]))
4576                         return middle + delta + 1;
4577         }
4578
4579         /* Every entry had the same hash */
4580         return count;
4581 }
4582
4583 /*
4584  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4585  * first_hash will record the 1st hash of the new bucket.
4586  *
4587  * Normally half of the xattrs will be moved.  But we have to make
4588  * sure that the xattrs with the same hash value are stored in the
4589  * same bucket. If all the xattrs in this bucket have the same hash
4590  * value, the new bucket will be initialized as an empty one and the
4591  * first_hash will be initialized as (hash_value+1).
4592  */
4593 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4594                                     handle_t *handle,
4595                                     u64 blk,
4596                                     u64 new_blk,
4597                                     u32 *first_hash,
4598                                     int new_bucket_head)
4599 {
4600         int ret, i;
4601         int count, start, len, name_value_len = 0, name_offset = 0;
4602         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4603         struct ocfs2_xattr_header *xh;
4604         struct ocfs2_xattr_entry *xe;
4605         int blocksize = inode->i_sb->s_blocksize;
4606
4607         trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4608                                               (unsigned long long)new_blk);
4609
4610         s_bucket = ocfs2_xattr_bucket_new(inode);
4611         t_bucket = ocfs2_xattr_bucket_new(inode);
4612         if (!s_bucket || !t_bucket) {
4613                 ret = -ENOMEM;
4614                 mlog_errno(ret);
4615                 goto out;
4616         }
4617
4618         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4619         if (ret) {
4620                 mlog_errno(ret);
4621                 goto out;
4622         }
4623
4624         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4625                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4626         if (ret) {
4627                 mlog_errno(ret);
4628                 goto out;
4629         }
4630
4631         /*
4632          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4633          * there's no need to read it.
4634          */
4635         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4636         if (ret) {
4637                 mlog_errno(ret);
4638                 goto out;
4639         }
4640
4641         /*
4642          * Hey, if we're overwriting t_bucket, what difference does
4643          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4644          * same part of ocfs2_cp_xattr_bucket().
4645          */
4646         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4647                                                 new_bucket_head ?
4648                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4649                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4650         if (ret) {
4651                 mlog_errno(ret);
4652                 goto out;
4653         }
4654
4655         xh = bucket_xh(s_bucket);
4656         count = le16_to_cpu(xh->xh_count);
4657         start = ocfs2_xattr_find_divide_pos(xh);
4658
4659         if (start == count) {
4660                 xe = &xh->xh_entries[start-1];
4661
4662                 /*
4663                  * initialized a new empty bucket here.
4664                  * The hash value is set as one larger than
4665                  * that of the last entry in the previous bucket.
4666                  */
4667                 for (i = 0; i < t_bucket->bu_blocks; i++)
4668                         memset(bucket_block(t_bucket, i), 0, blocksize);
4669
4670                 xh = bucket_xh(t_bucket);
4671                 xh->xh_free_start = cpu_to_le16(blocksize);
4672                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4673                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4674
4675                 goto set_num_buckets;
4676         }
4677
4678         /* copy the whole bucket to the new first. */
4679         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4680
4681         /* update the new bucket. */
4682         xh = bucket_xh(t_bucket);
4683
4684         /*
4685          * Calculate the total name/value len and xh_free_start for
4686          * the old bucket first.
4687          */
4688         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4689         name_value_len = 0;
4690         for (i = 0; i < start; i++) {
4691                 xe = &xh->xh_entries[i];
4692                 name_value_len += namevalue_size_xe(xe);
4693                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4694                         name_offset = le16_to_cpu(xe->xe_name_offset);
4695         }
4696
4697         /*
4698          * Now begin the modification to the new bucket.
4699          *
4700          * In the new bucket, We just move the xattr entry to the beginning
4701          * and don't touch the name/value. So there will be some holes in the
4702          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4703          * called.
4704          */
4705         xe = &xh->xh_entries[start];
4706         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4707         trace_ocfs2_divide_xattr_bucket_move(len,
4708                         (int)((char *)xe - (char *)xh),
4709                         (int)((char *)xh->xh_entries - (char *)xh));
4710         memmove((char *)xh->xh_entries, (char *)xe, len);
4711         xe = &xh->xh_entries[count - start];
4712         len = sizeof(struct ocfs2_xattr_entry) * start;
4713         memset((char *)xe, 0, len);
4714
4715         le16_add_cpu(&xh->xh_count, -start);
4716         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4717
4718         /* Calculate xh_free_start for the new bucket. */
4719         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4720         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4721                 xe = &xh->xh_entries[i];
4722                 if (le16_to_cpu(xe->xe_name_offset) <
4723                     le16_to_cpu(xh->xh_free_start))
4724                         xh->xh_free_start = xe->xe_name_offset;
4725         }
4726
4727 set_num_buckets:
4728         /* set xh->xh_num_buckets for the new xh. */
4729         if (new_bucket_head)
4730                 xh->xh_num_buckets = cpu_to_le16(1);
4731         else
4732                 xh->xh_num_buckets = 0;
4733
4734         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4735
4736         /* store the first_hash of the new bucket. */
4737         if (first_hash)
4738                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4739
4740         /*
4741          * Now only update the 1st block of the old bucket.  If we
4742          * just added a new empty bucket, there is no need to modify
4743          * it.
4744          */
4745         if (start == count)
4746                 goto out;
4747
4748         xh = bucket_xh(s_bucket);
4749         memset(&xh->xh_entries[start], 0,
4750                sizeof(struct ocfs2_xattr_entry) * (count - start));
4751         xh->xh_count = cpu_to_le16(start);
4752         xh->xh_free_start = cpu_to_le16(name_offset);
4753         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4754
4755         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4756
4757 out:
4758         ocfs2_xattr_bucket_free(s_bucket);
4759         ocfs2_xattr_bucket_free(t_bucket);
4760
4761         return ret;
4762 }
4763
4764 /*
4765  * Copy xattr from one bucket to another bucket.
4766  *
4767  * The caller must make sure that the journal transaction
4768  * has enough space for journaling.
4769  */
4770 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4771                                  handle_t *handle,
4772                                  u64 s_blkno,
4773                                  u64 t_blkno,
4774                                  int t_is_new)
4775 {
4776         int ret;
4777         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4778
4779         BUG_ON(s_blkno == t_blkno);
4780
4781         trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4782                                     (unsigned long long)t_blkno,
4783                                     t_is_new);
4784
4785         s_bucket = ocfs2_xattr_bucket_new(inode);
4786         t_bucket = ocfs2_xattr_bucket_new(inode);
4787         if (!s_bucket || !t_bucket) {
4788                 ret = -ENOMEM;
4789                 mlog_errno(ret);
4790                 goto out;
4791         }
4792
4793         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4794         if (ret)
4795                 goto out;
4796
4797         /*
4798          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4799          * there's no need to read it.
4800          */
4801         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4802         if (ret)
4803                 goto out;
4804
4805         /*
4806          * Hey, if we're overwriting t_bucket, what difference does
4807          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4808          * cluster to fill, we came here from
4809          * ocfs2_mv_xattr_buckets(), and it is really new -
4810          * ACCESS_CREATE is required.  But we also might have moved data
4811          * out of t_bucket before extending back into it.
4812          * ocfs2_add_new_xattr_bucket() can do this - its call to
4813          * ocfs2_add_new_xattr_cluster() may have created a new extent
4814          * and copied out the end of the old extent.  Then it re-extends
4815          * the old extent back to create space for new xattrs.  That's
4816          * how we get here, and the bucket isn't really new.
4817          */
4818         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4819                                                 t_is_new ?
4820                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4821                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4822         if (ret)
4823                 goto out;
4824
4825         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4826         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4827
4828 out:
4829         ocfs2_xattr_bucket_free(t_bucket);
4830         ocfs2_xattr_bucket_free(s_bucket);
4831
4832         return ret;
4833 }
4834
4835 /*
4836  * src_blk points to the start of an existing extent.  last_blk points to
4837  * last cluster in that extent.  to_blk points to a newly allocated
4838  * extent.  We copy the buckets from the cluster at last_blk to the new
4839  * extent.  If start_bucket is non-zero, we skip that many buckets before
4840  * we start copying.  The new extent's xh_num_buckets gets set to the
4841  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4842  * by the same amount.
4843  */
4844 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4845                                   u64 src_blk, u64 last_blk, u64 to_blk,
4846                                   unsigned int start_bucket,
4847                                   u32 *first_hash)
4848 {
4849         int i, ret, credits;
4850         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4851         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4852         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4853         struct ocfs2_xattr_bucket *old_first, *new_first;
4854
4855         trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4856                                      (unsigned long long)to_blk);
4857
4858         BUG_ON(start_bucket >= num_buckets);
4859         if (start_bucket) {
4860                 num_buckets -= start_bucket;
4861                 last_blk += (start_bucket * blks_per_bucket);
4862         }
4863
4864         /* The first bucket of the original extent */
4865         old_first = ocfs2_xattr_bucket_new(inode);
4866         /* The first bucket of the new extent */
4867         new_first = ocfs2_xattr_bucket_new(inode);
4868         if (!old_first || !new_first) {
4869                 ret = -ENOMEM;
4870                 mlog_errno(ret);
4871                 goto out;
4872         }
4873
4874         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4875         if (ret) {
4876                 mlog_errno(ret);
4877                 goto out;
4878         }
4879
4880         /*
4881          * We need to update the first bucket of the old extent and all
4882          * the buckets going to the new extent.
4883          */
4884         credits = ((num_buckets + 1) * blks_per_bucket);
4885         ret = ocfs2_extend_trans(handle, credits);
4886         if (ret) {
4887                 mlog_errno(ret);
4888                 goto out;
4889         }
4890
4891         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4892                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4893         if (ret) {
4894                 mlog_errno(ret);
4895                 goto out;
4896         }
4897
4898         for (i = 0; i < num_buckets; i++) {
4899                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4900                                             last_blk + (i * blks_per_bucket),
4901                                             to_blk + (i * blks_per_bucket),
4902                                             1);
4903                 if (ret) {
4904                         mlog_errno(ret);
4905                         goto out;
4906                 }
4907         }
4908
4909         /*
4910          * Get the new bucket ready before we dirty anything
4911          * (This actually shouldn't fail, because we already dirtied
4912          * it once in ocfs2_cp_xattr_bucket()).
4913          */
4914         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4915         if (ret) {
4916                 mlog_errno(ret);
4917                 goto out;
4918         }
4919         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4920                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4921         if (ret) {
4922                 mlog_errno(ret);
4923                 goto out;
4924         }
4925
4926         /* Now update the headers */
4927         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4928         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4929
4930         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4931         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4932
4933         if (first_hash)
4934                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4935
4936 out:
4937         ocfs2_xattr_bucket_free(new_first);
4938         ocfs2_xattr_bucket_free(old_first);
4939         return ret;
4940 }
4941
4942 /*
4943  * Move some xattrs in this cluster to the new cluster.
4944  * This function should only be called when bucket size == cluster size.
4945  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4946  */
4947 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4948                                       handle_t *handle,
4949                                       u64 prev_blk,
4950                                       u64 new_blk,
4951                                       u32 *first_hash)
4952 {
4953         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4954         int ret, credits = 2 * blk_per_bucket;
4955
4956         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4957
4958         ret = ocfs2_extend_trans(handle, credits);
4959         if (ret) {
4960                 mlog_errno(ret);
4961                 return ret;
4962         }
4963
4964         /* Move half of the xattr in start_blk to the next bucket. */
4965         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4966                                           new_blk, first_hash, 1);
4967 }
4968
4969 /*
4970  * Move some xattrs from the old cluster to the new one since they are not
4971  * contiguous in ocfs2 xattr tree.
4972  *
4973  * new_blk starts a new separate cluster, and we will move some xattrs from
4974  * prev_blk to it. v_start will be set as the first name hash value in this
4975  * new cluster so that it can be used as e_cpos during tree insertion and
4976  * don't collide with our original b-tree operations. first_bh and header_bh
4977  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4978  * to extend the insert bucket.
4979  *
4980  * The problem is how much xattr should we move to the new one and when should
4981  * we update first_bh and header_bh?
4982  * 1. If cluster size > bucket size, that means the previous cluster has more
4983  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4984  *    update the first_bh and header_bh if the insert bucket has been moved
4985  *    to the new cluster.
4986  * 2. If cluster_size == bucket_size:
4987  *    a) If the previous extent rec has more than one cluster and the insert
4988  *       place isn't in the last cluster, copy the entire last cluster to the
4989  *       new one. This time, we don't need to upate the first_bh and header_bh
4990  *       since they will not be moved into the new cluster.
4991  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4992  *       the new one. And we set the extend flag to zero if the insert place is
4993  *       moved into the new allocated cluster since no extend is needed.
4994  */
4995 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4996                                             handle_t *handle,
4997                                             struct ocfs2_xattr_bucket *first,
4998                                             struct ocfs2_xattr_bucket *target,
4999                                             u64 new_blk,
5000                                             u32 prev_clusters,
5001                                             u32 *v_start,
5002                                             int *extend)
5003 {
5004         int ret;
5005
5006         trace_ocfs2_adjust_xattr_cross_cluster(
5007                         (unsigned long long)bucket_blkno(first),
5008                         (unsigned long long)new_blk, prev_clusters);
5009
5010         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5011                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5012                                                           handle,
5013                                                           first, target,
5014                                                           new_blk,
5015                                                           prev_clusters,
5016                                                           v_start);
5017                 if (ret)
5018                         mlog_errno(ret);
5019         } else {
5020                 /* The start of the last cluster in the first extent */
5021                 u64 last_blk = bucket_blkno(first) +
5022                         ((prev_clusters - 1) *
5023                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
5024
5025                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5026                         ret = ocfs2_mv_xattr_buckets(inode, handle,
5027                                                      bucket_blkno(first),
5028                                                      last_blk, new_blk, 0,
5029                                                      v_start);
5030                         if (ret)
5031                                 mlog_errno(ret);
5032                 } else {
5033                         ret = ocfs2_divide_xattr_cluster(inode, handle,
5034                                                          last_blk, new_blk,
5035                                                          v_start);
5036                         if (ret)
5037                                 mlog_errno(ret);
5038
5039                         if ((bucket_blkno(target) == last_blk) && extend)
5040                                 *extend = 0;
5041                 }
5042         }
5043
5044         return ret;
5045 }
5046
5047 /*
5048  * Add a new cluster for xattr storage.
5049  *
5050  * If the new cluster is contiguous with the previous one, it will be
5051  * appended to the same extent record, and num_clusters will be updated.
5052  * If not, we will insert a new extent for it and move some xattrs in
5053  * the last cluster into the new allocated one.
5054  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5055  * lose the benefits of hashing because we'll have to search large leaves.
5056  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5057  * if it's bigger).
5058  *
5059  * first_bh is the first block of the previous extent rec and header_bh
5060  * indicates the bucket we will insert the new xattrs. They will be updated
5061  * when the header_bh is moved into the new cluster.
5062  */
5063 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5064                                        struct buffer_head *root_bh,
5065                                        struct ocfs2_xattr_bucket *first,
5066                                        struct ocfs2_xattr_bucket *target,
5067                                        u32 *num_clusters,
5068                                        u32 prev_cpos,
5069                                        int *extend,
5070                                        struct ocfs2_xattr_set_ctxt *ctxt)
5071 {
5072         int ret;
5073         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5074         u32 prev_clusters = *num_clusters;
5075         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5076         u64 block;
5077         handle_t *handle = ctxt->handle;
5078         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5079         struct ocfs2_extent_tree et;
5080
5081         trace_ocfs2_add_new_xattr_cluster_begin(
5082                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5083                 (unsigned long long)bucket_blkno(first),
5084                 prev_cpos, prev_clusters);
5085
5086         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5087
5088         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5089                                       OCFS2_JOURNAL_ACCESS_WRITE);
5090         if (ret < 0) {
5091                 mlog_errno(ret);
5092                 goto leave;
5093         }
5094
5095         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5096                                      clusters_to_add, &bit_off, &num_bits);
5097         if (ret < 0) {
5098                 if (ret != -ENOSPC)
5099                         mlog_errno(ret);
5100                 goto leave;
5101         }
5102
5103         BUG_ON(num_bits > clusters_to_add);
5104
5105         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5106         trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5107
5108         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5109             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5110              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5111                 /*
5112                  * If this cluster is contiguous with the old one and
5113                  * adding this new cluster, we don't surpass the limit of
5114                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5115                  * initialized and used like other buckets in the previous
5116                  * cluster.
5117                  * So add it as a contiguous one. The caller will handle
5118                  * its init process.
5119                  */
5120                 v_start = prev_cpos + prev_clusters;
5121                 *num_clusters = prev_clusters + num_bits;
5122         } else {
5123                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5124                                                        handle,
5125                                                        first,
5126                                                        target,
5127                                                        block,
5128                                                        prev_clusters,
5129                                                        &v_start,
5130                                                        extend);
5131                 if (ret) {
5132                         mlog_errno(ret);
5133                         goto leave;
5134                 }
5135         }
5136
5137         trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5138                                                  v_start, num_bits);
5139         ret = ocfs2_insert_extent(handle, &et, v_start, block,
5140                                   num_bits, 0, ctxt->meta_ac);
5141         if (ret < 0) {
5142                 mlog_errno(ret);
5143                 goto leave;
5144         }
5145
5146         ocfs2_journal_dirty(handle, root_bh);
5147
5148 leave:
5149         return ret;
5150 }
5151
5152 /*
5153  * We are given an extent.  'first' is the bucket at the very front of
5154  * the extent.  The extent has space for an additional bucket past
5155  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5156  * of the target bucket.  We wish to shift every bucket past the target
5157  * down one, filling in that additional space.  When we get back to the
5158  * target, we split the target between itself and the now-empty bucket
5159  * at target+1 (aka, target_blkno + blks_per_bucket).
5160  */
5161 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5162                                      handle_t *handle,
5163                                      struct ocfs2_xattr_bucket *first,
5164                                      u64 target_blk,
5165                                      u32 num_clusters)
5166 {
5167         int ret, credits;
5168         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5169         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5170         u64 end_blk;
5171         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5172
5173         trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5174                                         (unsigned long long)bucket_blkno(first),
5175                                         num_clusters, new_bucket);
5176
5177         /* The extent must have room for an additional bucket */
5178         BUG_ON(new_bucket >=
5179                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5180
5181         /* end_blk points to the last existing bucket */
5182         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5183
5184         /*
5185          * end_blk is the start of the last existing bucket.
5186          * Thus, (end_blk - target_blk) covers the target bucket and
5187          * every bucket after it up to, but not including, the last
5188          * existing bucket.  Then we add the last existing bucket, the
5189          * new bucket, and the first bucket (3 * blk_per_bucket).
5190          */
5191         credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5192         ret = ocfs2_extend_trans(handle, credits);
5193         if (ret) {
5194                 mlog_errno(ret);
5195                 goto out;
5196         }
5197
5198         ret = ocfs2_xattr_bucket_journal_access(handle, first,
5199                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5200         if (ret) {
5201                 mlog_errno(ret);
5202                 goto out;
5203         }
5204
5205         while (end_blk != target_blk) {
5206                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5207                                             end_blk + blk_per_bucket, 0);
5208                 if (ret)
5209                         goto out;
5210                 end_blk -= blk_per_bucket;
5211         }
5212
5213         /* Move half of the xattr in target_blkno to the next bucket. */
5214         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5215                                         target_blk + blk_per_bucket, NULL, 0);
5216
5217         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5218         ocfs2_xattr_bucket_journal_dirty(handle, first);
5219
5220 out:
5221         return ret;
5222 }
5223
5224 /*
5225  * Add new xattr bucket in an extent record and adjust the buckets
5226  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5227  * bucket we want to insert into.
5228  *
5229  * In the easy case, we will move all the buckets after target down by
5230  * one. Half of target's xattrs will be moved to the next bucket.
5231  *
5232  * If current cluster is full, we'll allocate a new one.  This may not
5233  * be contiguous.  The underlying calls will make sure that there is
5234  * space for the insert, shifting buckets around if necessary.
5235  * 'target' may be moved by those calls.
5236  */
5237 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5238                                       struct buffer_head *xb_bh,
5239                                       struct ocfs2_xattr_bucket *target,
5240                                       struct ocfs2_xattr_set_ctxt *ctxt)
5241 {
5242         struct ocfs2_xattr_block *xb =
5243                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5244         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5245         struct ocfs2_extent_list *el = &xb_root->xt_list;
5246         u32 name_hash =
5247                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5248         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5249         int ret, num_buckets, extend = 1;
5250         u64 p_blkno;
5251         u32 e_cpos, num_clusters;
5252         /* The bucket at the front of the extent */
5253         struct ocfs2_xattr_bucket *first;
5254
5255         trace_ocfs2_add_new_xattr_bucket(
5256                                 (unsigned long long)bucket_blkno(target));
5257
5258         /* The first bucket of the original extent */
5259         first = ocfs2_xattr_bucket_new(inode);
5260         if (!first) {
5261                 ret = -ENOMEM;
5262                 mlog_errno(ret);
5263                 goto out;
5264         }
5265
5266         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5267                                   &num_clusters, el);
5268         if (ret) {
5269                 mlog_errno(ret);
5270                 goto out;
5271         }
5272
5273         ret = ocfs2_read_xattr_bucket(first, p_blkno);
5274         if (ret) {
5275                 mlog_errno(ret);
5276                 goto out;
5277         }
5278
5279         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5280         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5281                 /*
5282                  * This can move first+target if the target bucket moves
5283                  * to the new extent.
5284                  */
5285                 ret = ocfs2_add_new_xattr_cluster(inode,
5286                                                   xb_bh,
5287                                                   first,
5288                                                   target,
5289                                                   &num_clusters,
5290                                                   e_cpos,
5291                                                   &extend,
5292                                                   ctxt);
5293                 if (ret) {
5294                         mlog_errno(ret);
5295                         goto out;
5296                 }
5297         }
5298
5299         if (extend) {
5300                 ret = ocfs2_extend_xattr_bucket(inode,
5301                                                 ctxt->handle,
5302                                                 first,
5303                                                 bucket_blkno(target),
5304                                                 num_clusters);
5305                 if (ret)
5306                         mlog_errno(ret);
5307         }
5308
5309 out:
5310         ocfs2_xattr_bucket_free(first);
5311
5312         return ret;
5313 }
5314
5315 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
5316                                         struct ocfs2_xattr_bucket *bucket,
5317                                         int offs)
5318 {
5319         int block_off = offs >> inode->i_sb->s_blocksize_bits;
5320
5321         offs = offs % inode->i_sb->s_blocksize;
5322         return bucket_block(bucket, block_off) + offs;
5323 }
5324
5325 /*
5326  * Truncate the specified xe_off entry in xattr bucket.
5327  * bucket is indicated by header_bh and len is the new length.
5328  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5329  *
5330  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5331  */
5332 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5333                                              struct ocfs2_xattr_bucket *bucket,
5334                                              int xe_off,
5335                                              int len,
5336                                              struct ocfs2_xattr_set_ctxt *ctxt)
5337 {
5338         int ret, offset;
5339         u64 value_blk;
5340         struct ocfs2_xattr_entry *xe;
5341         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5342         size_t blocksize = inode->i_sb->s_blocksize;
5343         struct ocfs2_xattr_value_buf vb = {
5344                 .vb_access = ocfs2_journal_access,
5345         };
5346
5347         xe = &xh->xh_entries[xe_off];
5348
5349         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5350
5351         offset = le16_to_cpu(xe->xe_name_offset) +
5352                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5353
5354         value_blk = offset / blocksize;
5355
5356         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5357         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5358
5359         vb.vb_bh = bucket->bu_bhs[value_blk];
5360         BUG_ON(!vb.vb_bh);
5361
5362         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5363                 (vb.vb_bh->b_data + offset % blocksize);
5364
5365         /*
5366          * From here on out we have to dirty the bucket.  The generic
5367          * value calls only modify one of the bucket's bhs, but we need
5368          * to send the bucket at once.  So if they error, they *could* have
5369          * modified something.  We have to assume they did, and dirty
5370          * the whole bucket.  This leaves us in a consistent state.
5371          */
5372         trace_ocfs2_xattr_bucket_value_truncate(
5373                         (unsigned long long)bucket_blkno(bucket), xe_off, len);
5374         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5375         if (ret) {
5376                 mlog_errno(ret);
5377                 goto out;
5378         }
5379
5380         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5381                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5382         if (ret) {
5383                 mlog_errno(ret);
5384                 goto out;
5385         }
5386
5387         xe->xe_value_size = cpu_to_le64(len);
5388
5389         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5390
5391 out:
5392         return ret;
5393 }
5394
5395 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5396                                   struct buffer_head *root_bh,
5397                                   u64 blkno,
5398                                   u32 cpos,
5399                                   u32 len,
5400                                   void *para)
5401 {
5402         int ret;
5403         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5404         struct inode *tl_inode = osb->osb_tl_inode;
5405         handle_t *handle;
5406         struct ocfs2_xattr_block *xb =
5407                         (struct ocfs2_xattr_block *)root_bh->b_data;
5408         struct ocfs2_alloc_context *meta_ac = NULL;
5409         struct ocfs2_cached_dealloc_ctxt dealloc;
5410         struct ocfs2_extent_tree et;
5411
5412         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5413                                           ocfs2_delete_xattr_in_bucket, para);
5414         if (ret) {
5415                 mlog_errno(ret);
5416                 return ret;
5417         }
5418
5419         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5420
5421         ocfs2_init_dealloc_ctxt(&dealloc);
5422
5423         trace_ocfs2_rm_xattr_cluster(
5424                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
5425                         (unsigned long long)blkno, cpos, len);
5426
5427         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5428                                                len);
5429
5430         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5431         if (ret) {
5432                 mlog_errno(ret);
5433                 return ret;
5434         }
5435
5436         mutex_lock(&tl_inode->i_mutex);
5437
5438         if (ocfs2_truncate_log_needs_flush(osb)) {
5439                 ret = __ocfs2_flush_truncate_log(osb);
5440                 if (ret < 0) {
5441                         mlog_errno(ret);
5442                         goto out;
5443                 }
5444         }
5445
5446         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5447         if (IS_ERR(handle)) {
5448                 ret = -ENOMEM;
5449                 mlog_errno(ret);
5450                 goto out;
5451         }
5452
5453         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5454                                       OCFS2_JOURNAL_ACCESS_WRITE);
5455         if (ret) {
5456                 mlog_errno(ret);
5457                 goto out_commit;
5458         }
5459
5460         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5461                                   &dealloc);
5462         if (ret) {
5463                 mlog_errno(ret);
5464                 goto out_commit;
5465         }
5466
5467         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5468         ocfs2_journal_dirty(handle, root_bh);
5469
5470         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5471         if (ret)
5472                 mlog_errno(ret);
5473
5474 out_commit:
5475         ocfs2_commit_trans(osb, handle);
5476 out:
5477         ocfs2_schedule_truncate_log_flush(osb, 1);
5478
5479         mutex_unlock(&tl_inode->i_mutex);
5480
5481         if (meta_ac)
5482                 ocfs2_free_alloc_context(meta_ac);
5483
5484         ocfs2_run_deallocs(osb, &dealloc);
5485
5486         return ret;
5487 }
5488
5489 /*
5490  * check whether the xattr bucket is filled up with the same hash value.
5491  * If we want to insert the xattr with the same hash, return -ENOSPC.
5492  * If we want to insert a xattr with different hash value, go ahead
5493  * and ocfs2_divide_xattr_bucket will handle this.
5494  */
5495 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5496                                               struct ocfs2_xattr_bucket *bucket,
5497                                               const char *name)
5498 {
5499         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5500         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5501
5502         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5503                 return 0;
5504
5505         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5506             xh->xh_entries[0].xe_name_hash) {
5507                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5508                      "hash = %u\n",
5509                      (unsigned long long)bucket_blkno(bucket),
5510                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5511                 return -ENOSPC;
5512         }
5513
5514         return 0;
5515 }
5516
5517 /*
5518  * Try to set the entry in the current bucket.  If we fail, the caller
5519  * will handle getting us another bucket.
5520  */
5521 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5522                                         struct ocfs2_xattr_info *xi,
5523                                         struct ocfs2_xattr_search *xs,
5524                                         struct ocfs2_xattr_set_ctxt *ctxt)
5525 {
5526         int ret;
5527         struct ocfs2_xa_loc loc;
5528
5529         trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5530
5531         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5532                                        xs->not_found ? NULL : xs->here);
5533         ret = ocfs2_xa_set(&loc, xi, ctxt);
5534         if (!ret) {
5535                 xs->here = loc.xl_entry;
5536                 goto out;
5537         }
5538         if (ret != -ENOSPC) {
5539                 mlog_errno(ret);
5540                 goto out;
5541         }
5542
5543         /* Ok, we need space.  Let's try defragmenting the bucket. */
5544         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5545                                         xs->bucket);
5546         if (ret) {
5547                 mlog_errno(ret);
5548                 goto out;
5549         }
5550
5551         ret = ocfs2_xa_set(&loc, xi, ctxt);
5552         if (!ret) {
5553                 xs->here = loc.xl_entry;
5554                 goto out;
5555         }
5556         if (ret != -ENOSPC)
5557                 mlog_errno(ret);
5558
5559
5560 out:
5561         return ret;
5562 }
5563
5564 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5565                                              struct ocfs2_xattr_info *xi,
5566                                              struct ocfs2_xattr_search *xs,
5567                                              struct ocfs2_xattr_set_ctxt *ctxt)
5568 {
5569         int ret;
5570
5571         trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5572
5573         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5574         if (!ret)
5575                 goto out;
5576         if (ret != -ENOSPC) {
5577                 mlog_errno(ret);
5578                 goto out;
5579         }
5580
5581         /* Ack, need more space.  Let's try to get another bucket! */
5582
5583         /*
5584          * We do not allow for overlapping ranges between buckets. And
5585          * the maximum number of collisions we will allow for then is
5586          * one bucket's worth, so check it here whether we need to
5587          * add a new bucket for the insert.
5588          */
5589         ret = ocfs2_check_xattr_bucket_collision(inode,
5590                                                  xs->bucket,
5591                                                  xi->xi_name);
5592         if (ret) {
5593                 mlog_errno(ret);
5594                 goto out;
5595         }
5596
5597         ret = ocfs2_add_new_xattr_bucket(inode,
5598                                          xs->xattr_bh,
5599                                          xs->bucket,
5600                                          ctxt);
5601         if (ret) {
5602                 mlog_errno(ret);
5603                 goto out;
5604         }
5605
5606         /*
5607          * ocfs2_add_new_xattr_bucket() will have updated
5608          * xs->bucket if it moved, but it will not have updated
5609          * any of the other search fields.  Thus, we drop it and
5610          * re-search.  Everything should be cached, so it'll be
5611          * quick.
5612          */
5613         ocfs2_xattr_bucket_relse(xs->bucket);
5614         ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5615                                            xi->xi_name_index,
5616                                            xi->xi_name, xs);
5617         if (ret && ret != -ENODATA)
5618                 goto out;
5619         xs->not_found = ret;
5620
5621         /* Ok, we have a new bucket, let's try again */
5622         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5623         if (ret && (ret != -ENOSPC))
5624                 mlog_errno(ret);
5625
5626 out:
5627         return ret;
5628 }
5629
5630 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5631                                         struct ocfs2_xattr_bucket *bucket,
5632                                         void *para)
5633 {
5634         int ret = 0, ref_credits;
5635         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5636         u16 i;
5637         struct ocfs2_xattr_entry *xe;
5638         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5639         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5640         int credits = ocfs2_remove_extent_credits(osb->sb) +
5641                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5642         struct ocfs2_xattr_value_root *xv;
5643         struct ocfs2_rm_xattr_bucket_para *args =
5644                         (struct ocfs2_rm_xattr_bucket_para *)para;
5645
5646         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5647
5648         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5649                 xe = &xh->xh_entries[i];
5650                 if (ocfs2_xattr_is_local(xe))
5651                         continue;
5652
5653                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5654                                                       i, &xv, NULL);
5655
5656                 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5657                                                          args->ref_ci,
5658                                                          args->ref_root_bh,
5659                                                          &ctxt.meta_ac,
5660                                                          &ref_credits);
5661
5662                 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5663                 if (IS_ERR(ctxt.handle)) {
5664                         ret = PTR_ERR(ctxt.handle);
5665                         mlog_errno(ret);
5666                         break;
5667                 }
5668
5669                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5670                                                         i, 0, &ctxt);
5671
5672                 ocfs2_commit_trans(osb, ctxt.handle);
5673                 if (ctxt.meta_ac) {
5674                         ocfs2_free_alloc_context(ctxt.meta_ac);
5675                         ctxt.meta_ac = NULL;
5676                 }
5677                 if (ret) {
5678                         mlog_errno(ret);
5679                         break;
5680                 }
5681         }
5682
5683         if (ctxt.meta_ac)
5684                 ocfs2_free_alloc_context(ctxt.meta_ac);
5685         ocfs2_schedule_truncate_log_flush(osb, 1);
5686         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5687         return ret;
5688 }
5689
5690 /*
5691  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5692  * or change the extent record flag), we need to recalculate
5693  * the metaecc for the whole bucket. So it is done here.
5694  *
5695  * Note:
5696  * We have to give the extra credits for the caller.
5697  */
5698 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5699                                             handle_t *handle,
5700                                             void *para)
5701 {
5702         int ret;
5703         struct ocfs2_xattr_bucket *bucket =
5704                         (struct ocfs2_xattr_bucket *)para;
5705
5706         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5707                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5708         if (ret) {
5709                 mlog_errno(ret);
5710                 return ret;
5711         }
5712
5713         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5714
5715         return 0;
5716 }
5717
5718 /*
5719  * Special action we need if the xattr value is refcounted.
5720  *
5721  * 1. If the xattr is refcounted, lock the tree.
5722  * 2. CoW the xattr if we are setting the new value and the value
5723  *    will be stored outside.
5724  * 3. In other case, decrease_refcount will work for us, so just
5725  *    lock the refcount tree, calculate the meta and credits is OK.
5726  *
5727  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5728  * currently CoW is a completed transaction, while this function
5729  * will also lock the allocators and let us deadlock. So we will
5730  * CoW the whole xattr value.
5731  */
5732 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5733                                         struct ocfs2_dinode *di,
5734                                         struct ocfs2_xattr_info *xi,
5735                                         struct ocfs2_xattr_search *xis,
5736                                         struct ocfs2_xattr_search *xbs,
5737                                         struct ocfs2_refcount_tree **ref_tree,
5738                                         int *meta_add,
5739                                         int *credits)
5740 {
5741         int ret = 0;
5742         struct ocfs2_xattr_block *xb;
5743         struct ocfs2_xattr_entry *xe;
5744         char *base;
5745         u32 p_cluster, num_clusters;
5746         unsigned int ext_flags;
5747         int name_offset, name_len;
5748         struct ocfs2_xattr_value_buf vb;
5749         struct ocfs2_xattr_bucket *bucket = NULL;
5750         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5751         struct ocfs2_post_refcount refcount;
5752         struct ocfs2_post_refcount *p = NULL;
5753         struct buffer_head *ref_root_bh = NULL;
5754
5755         if (!xis->not_found) {
5756                 xe = xis->here;
5757                 name_offset = le16_to_cpu(xe->xe_name_offset);
5758                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5759                 base = xis->base;
5760                 vb.vb_bh = xis->inode_bh;
5761                 vb.vb_access = ocfs2_journal_access_di;
5762         } else {
5763                 int i, block_off = 0;
5764                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5765                 xe = xbs->here;
5766                 name_offset = le16_to_cpu(xe->xe_name_offset);
5767                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5768                 i = xbs->here - xbs->header->xh_entries;
5769
5770                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5771                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5772                                                         bucket_xh(xbs->bucket),
5773                                                         i, &block_off,
5774                                                         &name_offset);
5775                         if (ret) {
5776                                 mlog_errno(ret);
5777                                 goto out;
5778                         }
5779                         base = bucket_block(xbs->bucket, block_off);
5780                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5781                         vb.vb_access = ocfs2_journal_access;
5782
5783                         if (ocfs2_meta_ecc(osb)) {
5784                                 /*create parameters for ocfs2_post_refcount. */
5785                                 bucket = xbs->bucket;
5786                                 refcount.credits = bucket->bu_blocks;
5787                                 refcount.para = bucket;
5788                                 refcount.func =
5789                                         ocfs2_xattr_bucket_post_refcount;
5790                                 p = &refcount;
5791                         }
5792                 } else {
5793                         base = xbs->base;
5794                         vb.vb_bh = xbs->xattr_bh;
5795                         vb.vb_access = ocfs2_journal_access_xb;
5796                 }
5797         }
5798
5799         if (ocfs2_xattr_is_local(xe))
5800                 goto out;
5801
5802         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5803                                 (base + name_offset + name_len);
5804
5805         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5806                                        &num_clusters, &vb.vb_xv->xr_list,
5807                                        &ext_flags);
5808         if (ret) {
5809                 mlog_errno(ret);
5810                 goto out;
5811         }
5812
5813         /*
5814          * We just need to check the 1st extent record, since we always
5815          * CoW the whole xattr. So there shouldn't be a xattr with
5816          * some REFCOUNT extent recs after the 1st one.
5817          */
5818         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5819                 goto out;
5820
5821         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5822                                        1, ref_tree, &ref_root_bh);
5823         if (ret) {
5824                 mlog_errno(ret);
5825                 goto out;
5826         }
5827
5828         /*
5829          * If we are deleting the xattr or the new size will be stored inside,
5830          * cool, leave it there, the xattr truncate process will remove them
5831          * for us(it still needs the refcount tree lock and the meta, credits).
5832          * And the worse case is that every cluster truncate will split the
5833          * refcount tree, and make the original extent become 3. So we will need
5834          * 2 * cluster more extent recs at most.
5835          */
5836         if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5837
5838                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5839                                                          &(*ref_tree)->rf_ci,
5840                                                          ref_root_bh, vb.vb_xv,
5841                                                          meta_add, credits);
5842                 if (ret)
5843                         mlog_errno(ret);
5844                 goto out;
5845         }
5846
5847         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5848                                        *ref_tree, ref_root_bh, 0,
5849                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5850         if (ret)
5851                 mlog_errno(ret);
5852
5853 out:
5854         brelse(ref_root_bh);
5855         return ret;
5856 }
5857
5858 /*
5859  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5860  * The physical clusters will be added to refcount tree.
5861  */
5862 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5863                                 struct ocfs2_xattr_value_root *xv,
5864                                 struct ocfs2_extent_tree *value_et,
5865                                 struct ocfs2_caching_info *ref_ci,
5866                                 struct buffer_head *ref_root_bh,
5867                                 struct ocfs2_cached_dealloc_ctxt *dealloc,
5868                                 struct ocfs2_post_refcount *refcount)
5869 {
5870         int ret = 0;
5871         u32 clusters = le32_to_cpu(xv->xr_clusters);
5872         u32 cpos, p_cluster, num_clusters;
5873         struct ocfs2_extent_list *el = &xv->xr_list;
5874         unsigned int ext_flags;
5875
5876         cpos = 0;
5877         while (cpos < clusters) {
5878                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5879                                                &num_clusters, el, &ext_flags);
5880                 if (ret) {
5881                         mlog_errno(ret);
5882                         break;
5883                 }
5884
5885                 cpos += num_clusters;
5886                 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5887                         continue;
5888
5889                 BUG_ON(!p_cluster);
5890
5891                 ret = ocfs2_add_refcount_flag(inode, value_et,
5892                                               ref_ci, ref_root_bh,
5893                                               cpos - num_clusters,
5894                                               p_cluster, num_clusters,
5895                                               dealloc, refcount);
5896                 if (ret) {
5897                         mlog_errno(ret);
5898                         break;
5899                 }
5900         }
5901
5902         return ret;
5903 }
5904
5905 /*
5906  * Given a normal ocfs2_xattr_header, refcount all the entries which
5907  * have value stored outside.
5908  * Used for xattrs stored in inode and ocfs2_xattr_block.
5909  */
5910 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5911                                 struct ocfs2_xattr_value_buf *vb,
5912                                 struct ocfs2_xattr_header *header,
5913                                 struct ocfs2_caching_info *ref_ci,
5914                                 struct buffer_head *ref_root_bh,
5915                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5916 {
5917
5918         struct ocfs2_xattr_entry *xe;
5919         struct ocfs2_xattr_value_root *xv;
5920         struct ocfs2_extent_tree et;
5921         int i, ret = 0;
5922
5923         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5924                 xe = &header->xh_entries[i];
5925
5926                 if (ocfs2_xattr_is_local(xe))
5927                         continue;
5928
5929                 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5930                         le16_to_cpu(xe->xe_name_offset) +
5931                         OCFS2_XATTR_SIZE(xe->xe_name_len));
5932
5933                 vb->vb_xv = xv;
5934                 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5935
5936                 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5937                                                         ref_ci, ref_root_bh,
5938                                                         dealloc, NULL);
5939                 if (ret) {
5940                         mlog_errno(ret);
5941                         break;
5942                 }
5943         }
5944
5945         return ret;
5946 }
5947
5948 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5949                                 struct buffer_head *fe_bh,
5950                                 struct ocfs2_caching_info *ref_ci,
5951                                 struct buffer_head *ref_root_bh,
5952                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5953 {
5954         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5955         struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5956                                 (fe_bh->b_data + inode->i_sb->s_blocksize -
5957                                 le16_to_cpu(di->i_xattr_inline_size));
5958         struct ocfs2_xattr_value_buf vb = {
5959                 .vb_bh = fe_bh,
5960                 .vb_access = ocfs2_journal_access_di,
5961         };
5962
5963         return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5964                                                   ref_ci, ref_root_bh, dealloc);
5965 }
5966
5967 struct ocfs2_xattr_tree_value_refcount_para {
5968         struct ocfs2_caching_info *ref_ci;
5969         struct buffer_head *ref_root_bh;
5970         struct ocfs2_cached_dealloc_ctxt *dealloc;
5971 };
5972
5973 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5974                                            struct ocfs2_xattr_bucket *bucket,
5975                                            int offset,
5976                                            struct ocfs2_xattr_value_root **xv,
5977                                            struct buffer_head **bh)
5978 {
5979         int ret, block_off, name_offset;
5980         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5981         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
5982         void *base;
5983
5984         ret = ocfs2_xattr_bucket_get_name_value(sb,
5985                                                 bucket_xh(bucket),
5986                                                 offset,
5987                                                 &block_off,
5988                                                 &name_offset);
5989         if (ret) {
5990                 mlog_errno(ret);
5991                 goto out;
5992         }
5993
5994         base = bucket_block(bucket, block_off);
5995
5996         *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
5997                          OCFS2_XATTR_SIZE(xe->xe_name_len));
5998
5999         if (bh)
6000                 *bh = bucket->bu_bhs[block_off];
6001 out:
6002         return ret;
6003 }
6004
6005 /*
6006  * For a given xattr bucket, refcount all the entries which
6007  * have value stored outside.
6008  */
6009 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6010                                              struct ocfs2_xattr_bucket *bucket,
6011                                              void *para)
6012 {
6013         int i, ret = 0;
6014         struct ocfs2_extent_tree et;
6015         struct ocfs2_xattr_tree_value_refcount_para *ref =
6016                         (struct ocfs2_xattr_tree_value_refcount_para *)para;
6017         struct ocfs2_xattr_header *xh =
6018                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6019         struct ocfs2_xattr_entry *xe;
6020         struct ocfs2_xattr_value_buf vb = {
6021                 .vb_access = ocfs2_journal_access,
6022         };
6023         struct ocfs2_post_refcount refcount = {
6024                 .credits = bucket->bu_blocks,
6025                 .para = bucket,
6026                 .func = ocfs2_xattr_bucket_post_refcount,
6027         };
6028         struct ocfs2_post_refcount *p = NULL;
6029
6030         /* We only need post_refcount if we support metaecc. */
6031         if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6032                 p = &refcount;
6033
6034         trace_ocfs2_xattr_bucket_value_refcount(
6035                                 (unsigned long long)bucket_blkno(bucket),
6036                                 le16_to_cpu(xh->xh_count));
6037         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6038                 xe = &xh->xh_entries[i];
6039
6040                 if (ocfs2_xattr_is_local(xe))
6041                         continue;
6042
6043                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6044                                                       &vb.vb_xv, &vb.vb_bh);
6045                 if (ret) {
6046                         mlog_errno(ret);
6047                         break;
6048                 }
6049
6050                 ocfs2_init_xattr_value_extent_tree(&et,
6051                                                    INODE_CACHE(inode), &vb);
6052
6053                 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6054                                                         &et, ref->ref_ci,
6055                                                         ref->ref_root_bh,
6056                                                         ref->dealloc, p);
6057                 if (ret) {
6058                         mlog_errno(ret);
6059                         break;
6060                 }
6061         }
6062
6063         return ret;
6064
6065 }
6066
6067 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6068                                      struct buffer_head *root_bh,
6069                                      u64 blkno, u32 cpos, u32 len, void *para)
6070 {
6071         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6072                                            ocfs2_xattr_bucket_value_refcount,
6073                                            para);
6074 }
6075
6076 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6077                                 struct buffer_head *blk_bh,
6078                                 struct ocfs2_caching_info *ref_ci,
6079                                 struct buffer_head *ref_root_bh,
6080                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
6081 {
6082         int ret = 0;
6083         struct ocfs2_xattr_block *xb =
6084                                 (struct ocfs2_xattr_block *)blk_bh->b_data;
6085
6086         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6087                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6088                 struct ocfs2_xattr_value_buf vb = {
6089                         .vb_bh = blk_bh,
6090                         .vb_access = ocfs2_journal_access_xb,
6091                 };
6092
6093                 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6094                                                          ref_ci, ref_root_bh,
6095                                                          dealloc);
6096         } else {
6097                 struct ocfs2_xattr_tree_value_refcount_para para = {
6098                         .ref_ci = ref_ci,
6099                         .ref_root_bh = ref_root_bh,
6100                         .dealloc = dealloc,
6101                 };
6102
6103                 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6104                                                 ocfs2_refcount_xattr_tree_rec,
6105                                                 &para);
6106         }
6107
6108         return ret;
6109 }
6110
6111 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6112                                      struct buffer_head *fe_bh,
6113                                      struct ocfs2_caching_info *ref_ci,
6114                                      struct buffer_head *ref_root_bh,
6115                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
6116 {
6117         int ret = 0;
6118         struct ocfs2_inode_info *oi = OCFS2_I(inode);
6119         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6120         struct buffer_head *blk_bh = NULL;
6121
6122         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6123                 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6124                                                          ref_ci, ref_root_bh,
6125                                                          dealloc);
6126                 if (ret) {
6127                         mlog_errno(ret);
6128                         goto out;
6129                 }
6130         }
6131
6132         if (!di->i_xattr_loc)
6133                 goto out;
6134
6135         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6136                                      &blk_bh);
6137         if (ret < 0) {
6138                 mlog_errno(ret);
6139                 goto out;
6140         }
6141
6142         ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6143                                                 ref_root_bh, dealloc);
6144         if (ret)
6145                 mlog_errno(ret);
6146
6147         brelse(blk_bh);
6148 out:
6149
6150         return ret;
6151 }
6152
6153 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6154 /*
6155  * Store the information we need in xattr reflink.
6156  * old_bh and new_bh are inode bh for the old and new inode.
6157  */
6158 struct ocfs2_xattr_reflink {
6159         struct inode *old_inode;
6160         struct inode *new_inode;
6161         struct buffer_head *old_bh;
6162         struct buffer_head *new_bh;
6163         struct ocfs2_caching_info *ref_ci;
6164         struct buffer_head *ref_root_bh;
6165         struct ocfs2_cached_dealloc_ctxt *dealloc;
6166         should_xattr_reflinked *xattr_reflinked;
6167 };
6168
6169 /*
6170  * Given a xattr header and xe offset,
6171  * return the proper xv and the corresponding bh.
6172  * xattr in inode, block and xattr tree have different implementaions.
6173  */
6174 typedef int (get_xattr_value_root)(struct super_block *sb,
6175                                    struct buffer_head *bh,
6176                                    struct ocfs2_xattr_header *xh,
6177                                    int offset,
6178                                    struct ocfs2_xattr_value_root **xv,
6179                                    struct buffer_head **ret_bh,
6180                                    void *para);
6181
6182 /*
6183  * Calculate all the xattr value root metadata stored in this xattr header and
6184  * credits we need if we create them from the scratch.
6185  * We use get_xattr_value_root so that all types of xattr container can use it.
6186  */
6187 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6188                                              struct buffer_head *bh,
6189                                              struct ocfs2_xattr_header *xh,
6190                                              int *metas, int *credits,
6191                                              int *num_recs,
6192                                              get_xattr_value_root *func,
6193                                              void *para)
6194 {
6195         int i, ret = 0;
6196         struct ocfs2_xattr_value_root *xv;
6197         struct ocfs2_xattr_entry *xe;
6198
6199         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6200                 xe = &xh->xh_entries[i];
6201                 if (ocfs2_xattr_is_local(xe))
6202                         continue;
6203
6204                 ret = func(sb, bh, xh, i, &xv, NULL, para);
6205                 if (ret) {
6206                         mlog_errno(ret);
6207                         break;
6208                 }
6209
6210                 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6211                           le16_to_cpu(xv->xr_list.l_next_free_rec);
6212
6213                 *credits += ocfs2_calc_extend_credits(sb,
6214                                                 &def_xv.xv.xr_list);
6215
6216                 /*
6217                  * If the value is a tree with depth > 1, We don't go deep
6218                  * to the extent block, so just calculate a maximum record num.
6219                  */
6220                 if (!xv->xr_list.l_tree_depth)
6221                         *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6222                 else
6223                         *num_recs += ocfs2_clusters_for_bytes(sb,
6224                                                               XATTR_SIZE_MAX);
6225         }
6226
6227         return ret;
6228 }
6229
6230 /* Used by xattr inode and block to return the right xv and buffer_head. */
6231 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6232                                       struct buffer_head *bh,
6233                                       struct ocfs2_xattr_header *xh,
6234                                       int offset,
6235                                       struct ocfs2_xattr_value_root **xv,
6236                                       struct buffer_head **ret_bh,
6237                                       void *para)
6238 {
6239         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6240
6241         *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6242                 le16_to_cpu(xe->xe_name_offset) +
6243                 OCFS2_XATTR_SIZE(xe->xe_name_len));
6244
6245         if (ret_bh)
6246                 *ret_bh = bh;
6247
6248         return 0;
6249 }
6250
6251 /*
6252  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6253  * It is only used for inline xattr and xattr block.
6254  */
6255 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6256                                         struct ocfs2_xattr_header *xh,
6257                                         struct buffer_head *ref_root_bh,
6258                                         int *credits,
6259                                         struct ocfs2_alloc_context **meta_ac)
6260 {
6261         int ret, meta_add = 0, num_recs = 0;
6262         struct ocfs2_refcount_block *rb =
6263                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6264
6265         *credits = 0;
6266
6267         ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6268                                                 &meta_add, credits, &num_recs,
6269                                                 ocfs2_get_xattr_value_root,
6270                                                 NULL);
6271         if (ret) {
6272                 mlog_errno(ret);
6273                 goto out;
6274         }
6275
6276         /*
6277          * We need to add/modify num_recs in refcount tree, so just calculate
6278          * an approximate number we need for refcount tree change.
6279          * Sometimes we need to split the tree, and after split,  half recs
6280          * will be moved to the new block, and a new block can only provide
6281          * half number of recs. So we multiple new blocks by 2.
6282          */
6283         num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6284         meta_add += num_recs;
6285         *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6286         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6287                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6288                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6289         else
6290                 *credits += 1;
6291
6292         ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6293         if (ret)
6294                 mlog_errno(ret);
6295
6296 out:
6297         return ret;
6298 }
6299
6300 /*
6301  * Given a xattr header, reflink all the xattrs in this container.
6302  * It can be used for inode, block and bucket.
6303  *
6304  * NOTE:
6305  * Before we call this function, the caller has memcpy the xattr in
6306  * old_xh to the new_xh.
6307  *
6308  * If args.xattr_reflinked is set, call it to decide whether the xe should
6309  * be reflinked or not. If not, remove it from the new xattr header.
6310  */
6311 static int ocfs2_reflink_xattr_header(handle_t *handle,
6312                                       struct ocfs2_xattr_reflink *args,
6313                                       struct buffer_head *old_bh,
6314                                       struct ocfs2_xattr_header *xh,
6315                                       struct buffer_head *new_bh,
6316                                       struct ocfs2_xattr_header *new_xh,
6317                                       struct ocfs2_xattr_value_buf *vb,
6318                                       struct ocfs2_alloc_context *meta_ac,
6319                                       get_xattr_value_root *func,
6320                                       void *para)
6321 {
6322         int ret = 0, i, j;
6323         struct super_block *sb = args->old_inode->i_sb;
6324         struct buffer_head *value_bh;
6325         struct ocfs2_xattr_entry *xe, *last;
6326         struct ocfs2_xattr_value_root *xv, *new_xv;
6327         struct ocfs2_extent_tree data_et;
6328         u32 clusters, cpos, p_cluster, num_clusters;
6329         unsigned int ext_flags = 0;
6330
6331         trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6332                                          le16_to_cpu(xh->xh_count));
6333
6334         last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6335         for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6336                 xe = &xh->xh_entries[i];
6337
6338                 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6339                         xe = &new_xh->xh_entries[j];
6340
6341                         le16_add_cpu(&new_xh->xh_count, -1);
6342                         if (new_xh->xh_count) {
6343                                 memmove(xe, xe + 1,
6344                                         (void *)last - (void *)xe);
6345                                 memset(last, 0,
6346                                        sizeof(struct ocfs2_xattr_entry));
6347                         }
6348
6349                         /*
6350                          * We don't want j to increase in the next round since
6351                          * it is already moved ahead.
6352                          */
6353                         j--;
6354                         continue;
6355                 }
6356
6357                 if (ocfs2_xattr_is_local(xe))
6358                         continue;
6359
6360                 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6361                 if (ret) {
6362                         mlog_errno(ret);
6363                         break;
6364                 }
6365
6366                 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6367                 if (ret) {
6368                         mlog_errno(ret);
6369                         break;
6370                 }
6371
6372                 /*
6373                  * For the xattr which has l_tree_depth = 0, all the extent
6374                  * recs have already be copied to the new xh with the
6375                  * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6376                  * increase the refount count int the refcount tree.
6377                  *
6378                  * For the xattr which has l_tree_depth > 0, we need
6379                  * to initialize it to the empty default value root,
6380                  * and then insert the extents one by one.
6381                  */
6382                 if (xv->xr_list.l_tree_depth) {
6383                         memcpy(new_xv, &def_xv, sizeof(def_xv));
6384                         vb->vb_xv = new_xv;
6385                         vb->vb_bh = value_bh;
6386                         ocfs2_init_xattr_value_extent_tree(&data_et,
6387                                         INODE_CACHE(args->new_inode), vb);
6388                 }
6389
6390                 clusters = le32_to_cpu(xv->xr_clusters);
6391                 cpos = 0;
6392                 while (cpos < clusters) {
6393                         ret = ocfs2_xattr_get_clusters(args->old_inode,
6394                                                        cpos,
6395                                                        &p_cluster,
6396                                                        &num_clusters,
6397                                                        &xv->xr_list,
6398                                                        &ext_flags);
6399                         if (ret) {
6400                                 mlog_errno(ret);
6401                                 goto out;
6402                         }
6403
6404                         BUG_ON(!p_cluster);
6405
6406                         if (xv->xr_list.l_tree_depth) {
6407                                 ret = ocfs2_insert_extent(handle,
6408                                                 &data_et, cpos,
6409                                                 ocfs2_clusters_to_blocks(
6410                                                         args->old_inode->i_sb,
6411                                                         p_cluster),
6412                                                 num_clusters, ext_flags,
6413                                                 meta_ac);
6414                                 if (ret) {
6415                                         mlog_errno(ret);
6416                                         goto out;
6417                                 }
6418                         }
6419
6420                         ret = ocfs2_increase_refcount(handle, args->ref_ci,
6421                                                       args->ref_root_bh,
6422                                                       p_cluster, num_clusters,
6423                                                       meta_ac, args->dealloc);
6424                         if (ret) {
6425                                 mlog_errno(ret);
6426                                 goto out;
6427                         }
6428
6429                         cpos += num_clusters;
6430                 }
6431         }
6432
6433 out:
6434         return ret;
6435 }
6436
6437 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6438 {
6439         int ret = 0, credits = 0;
6440         handle_t *handle;
6441         struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6442         struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6443         int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6444         int header_off = osb->sb->s_blocksize - inline_size;
6445         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6446                                         (args->old_bh->b_data + header_off);
6447         struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6448                                         (args->new_bh->b_data + header_off);
6449         struct ocfs2_alloc_context *meta_ac = NULL;
6450         struct ocfs2_inode_info *new_oi;
6451         struct ocfs2_dinode *new_di;
6452         struct ocfs2_xattr_value_buf vb = {
6453                 .vb_bh = args->new_bh,
6454                 .vb_access = ocfs2_journal_access_di,
6455         };
6456
6457         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6458                                                   &credits, &meta_ac);
6459         if (ret) {
6460                 mlog_errno(ret);
6461                 goto out;
6462         }
6463
6464         handle = ocfs2_start_trans(osb, credits);
6465         if (IS_ERR(handle)) {
6466                 ret = PTR_ERR(handle);
6467                 mlog_errno(ret);
6468                 goto out;
6469         }
6470
6471         ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6472                                       args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6473         if (ret) {
6474                 mlog_errno(ret);
6475                 goto out_commit;
6476         }
6477
6478         memcpy(args->new_bh->b_data + header_off,
6479                args->old_bh->b_data + header_off, inline_size);
6480
6481         new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6482         new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6483
6484         ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6485                                          args->new_bh, new_xh, &vb, meta_ac,
6486                                          ocfs2_get_xattr_value_root, NULL);
6487         if (ret) {
6488                 mlog_errno(ret);
6489                 goto out_commit;
6490         }
6491
6492         new_oi = OCFS2_I(args->new_inode);
6493         /*
6494          * Adjust extent record count to reserve space for extended attribute.
6495          * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6496          */
6497         if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6498             !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6499                 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6500                 le16_add_cpu(&el->l_count, -(inline_size /
6501                                         sizeof(struct ocfs2_extent_rec)));
6502         }
6503         spin_lock(&new_oi->ip_lock);
6504         new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6505         new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6506         spin_unlock(&new_oi->ip_lock);
6507
6508         ocfs2_journal_dirty(handle, args->new_bh);
6509
6510 out_commit:
6511         ocfs2_commit_trans(osb, handle);
6512
6513 out:
6514         if (meta_ac)
6515                 ocfs2_free_alloc_context(meta_ac);
6516         return ret;
6517 }
6518
6519 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6520                                           struct buffer_head *fe_bh,
6521                                           struct buffer_head **ret_bh,
6522                                           int indexed)
6523 {
6524         int ret;
6525         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6526         struct ocfs2_xattr_set_ctxt ctxt;
6527
6528         memset(&ctxt, 0, sizeof(ctxt));
6529         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6530         if (ret < 0) {
6531                 mlog_errno(ret);
6532                 return ret;
6533         }
6534
6535         ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6536         if (IS_ERR(ctxt.handle)) {
6537                 ret = PTR_ERR(ctxt.handle);
6538                 mlog_errno(ret);
6539                 goto out;
6540         }
6541
6542         trace_ocfs2_create_empty_xattr_block(
6543                                 (unsigned long long)fe_bh->b_blocknr, indexed);
6544         ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6545                                        ret_bh);
6546         if (ret)
6547                 mlog_errno(ret);
6548
6549         ocfs2_commit_trans(osb, ctxt.handle);
6550 out:
6551         ocfs2_free_alloc_context(ctxt.meta_ac);
6552         return ret;
6553 }
6554
6555 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6556                                      struct buffer_head *blk_bh,
6557                                      struct buffer_head *new_blk_bh)
6558 {
6559         int ret = 0, credits = 0;
6560         handle_t *handle;
6561         struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6562         struct ocfs2_dinode *new_di;
6563         struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6564         int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6565         struct ocfs2_xattr_block *xb =
6566                         (struct ocfs2_xattr_block *)blk_bh->b_data;
6567         struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6568         struct ocfs2_xattr_block *new_xb =
6569                         (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6570         struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6571         struct ocfs2_alloc_context *meta_ac;
6572         struct ocfs2_xattr_value_buf vb = {
6573                 .vb_bh = new_blk_bh,
6574                 .vb_access = ocfs2_journal_access_xb,
6575         };
6576
6577         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6578                                                   &credits, &meta_ac);
6579         if (ret) {
6580                 mlog_errno(ret);
6581                 return ret;
6582         }
6583
6584         /* One more credits in case we need to add xattr flags in new inode. */
6585         handle = ocfs2_start_trans(osb, credits + 1);
6586         if (IS_ERR(handle)) {
6587                 ret = PTR_ERR(handle);
6588                 mlog_errno(ret);
6589                 goto out;
6590         }
6591
6592         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6593                 ret = ocfs2_journal_access_di(handle,
6594                                               INODE_CACHE(args->new_inode),
6595                                               args->new_bh,
6596                                               OCFS2_JOURNAL_ACCESS_WRITE);
6597                 if (ret) {
6598                         mlog_errno(ret);
6599                         goto out_commit;
6600                 }
6601         }
6602
6603         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6604                                       new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6605         if (ret) {
6606                 mlog_errno(ret);
6607                 goto out_commit;
6608         }
6609
6610         memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6611                osb->sb->s_blocksize - header_off);
6612
6613         ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6614                                          new_blk_bh, new_xh, &vb, meta_ac,
6615                                          ocfs2_get_xattr_value_root, NULL);
6616         if (ret) {
6617                 mlog_errno(ret);
6618                 goto out_commit;
6619         }
6620
6621         ocfs2_journal_dirty(handle, new_blk_bh);
6622
6623         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6624                 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6625                 spin_lock(&new_oi->ip_lock);
6626                 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6627                 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6628                 spin_unlock(&new_oi->ip_lock);
6629
6630                 ocfs2_journal_dirty(handle, args->new_bh);
6631         }
6632
6633 out_commit:
6634         ocfs2_commit_trans(osb, handle);
6635
6636 out:
6637         ocfs2_free_alloc_context(meta_ac);
6638         return ret;
6639 }
6640
6641 struct ocfs2_reflink_xattr_tree_args {
6642         struct ocfs2_xattr_reflink *reflink;
6643         struct buffer_head *old_blk_bh;
6644         struct buffer_head *new_blk_bh;
6645         struct ocfs2_xattr_bucket *old_bucket;
6646         struct ocfs2_xattr_bucket *new_bucket;
6647 };
6648
6649 /*
6650  * NOTE:
6651  * We have to handle the case that both old bucket and new bucket
6652  * will call this function to get the right ret_bh.
6653  * So The caller must give us the right bh.
6654  */
6655 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6656                                         struct buffer_head *bh,
6657                                         struct ocfs2_xattr_header *xh,
6658                                         int offset,
6659                                         struct ocfs2_xattr_value_root **xv,
6660                                         struct buffer_head **ret_bh,
6661                                         void *para)
6662 {
6663         struct ocfs2_reflink_xattr_tree_args *args =
6664                         (struct ocfs2_reflink_xattr_tree_args *)para;
6665         struct ocfs2_xattr_bucket *bucket;
6666
6667         if (bh == args->old_bucket->bu_bhs[0])
6668                 bucket = args->old_bucket;
6669         else
6670                 bucket = args->new_bucket;
6671
6672         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6673                                                xv, ret_bh);
6674 }
6675
6676 struct ocfs2_value_tree_metas {
6677         int num_metas;
6678         int credits;
6679         int num_recs;
6680 };
6681
6682 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6683                                         struct buffer_head *bh,
6684                                         struct ocfs2_xattr_header *xh,
6685                                         int offset,
6686                                         struct ocfs2_xattr_value_root **xv,
6687                                         struct buffer_head **ret_bh,
6688                                         void *para)
6689 {
6690         struct ocfs2_xattr_bucket *bucket =
6691                                 (struct ocfs2_xattr_bucket *)para;
6692
6693         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6694                                                xv, ret_bh);
6695 }
6696
6697 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6698                                       struct ocfs2_xattr_bucket *bucket,
6699                                       void *para)
6700 {
6701         struct ocfs2_value_tree_metas *metas =
6702                         (struct ocfs2_value_tree_metas *)para;
6703         struct ocfs2_xattr_header *xh =
6704                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6705
6706         /* Add the credits for this bucket first. */
6707         metas->credits += bucket->bu_blocks;
6708         return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6709                                         xh, &metas->num_metas,
6710                                         &metas->credits, &metas->num_recs,
6711                                         ocfs2_value_tree_metas_in_bucket,
6712                                         bucket);
6713 }
6714
6715 /*
6716  * Given a xattr extent rec starting from blkno and having len clusters,
6717  * iterate all the buckets calculate how much metadata we need for reflinking
6718  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6719  */
6720 static int ocfs2_lock_reflink_xattr_rec_allocators(
6721                                 struct ocfs2_reflink_xattr_tree_args *args,
6722                                 struct ocfs2_extent_tree *xt_et,
6723                                 u64 blkno, u32 len, int *credits,
6724                                 struct ocfs2_alloc_context **meta_ac,
6725                                 struct ocfs2_alloc_context **data_ac)
6726 {
6727         int ret, num_free_extents;
6728         struct ocfs2_value_tree_metas metas;
6729         struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6730         struct ocfs2_refcount_block *rb;
6731
6732         memset(&metas, 0, sizeof(metas));
6733
6734         ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6735                                           ocfs2_calc_value_tree_metas, &metas);
6736         if (ret) {
6737                 mlog_errno(ret);
6738                 goto out;
6739         }
6740
6741         *credits = metas.credits;
6742
6743         /*
6744          * Calculate we need for refcount tree change.
6745          *
6746          * We need to add/modify num_recs in refcount tree, so just calculate
6747          * an approximate number we need for refcount tree change.
6748          * Sometimes we need to split the tree, and after split,  half recs
6749          * will be moved to the new block, and a new block can only provide
6750          * half number of recs. So we multiple new blocks by 2.
6751          * In the end, we have to add credits for modifying the already
6752          * existed refcount block.
6753          */
6754         rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6755         metas.num_recs =
6756                 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6757                  ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6758         metas.num_metas += metas.num_recs;
6759         *credits += metas.num_recs +
6760                     metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6761         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6762                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6763                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6764         else
6765                 *credits += 1;
6766
6767         /* count in the xattr tree change. */
6768         num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6769         if (num_free_extents < 0) {
6770                 ret = num_free_extents;
6771                 mlog_errno(ret);
6772                 goto out;
6773         }
6774
6775         if (num_free_extents < len)
6776                 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6777
6778         *credits += ocfs2_calc_extend_credits(osb->sb,
6779                                               xt_et->et_root_el);
6780
6781         if (metas.num_metas) {
6782                 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6783                                                         meta_ac);
6784                 if (ret) {
6785                         mlog_errno(ret);
6786                         goto out;
6787                 }
6788         }
6789
6790         if (len) {
6791                 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6792                 if (ret)
6793                         mlog_errno(ret);
6794         }
6795 out:
6796         if (ret) {
6797                 if (*meta_ac) {
6798                         ocfs2_free_alloc_context(*meta_ac);
6799                         *meta_ac = NULL;
6800                 }
6801         }
6802
6803         return ret;
6804 }
6805
6806 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6807                                 u64 blkno, u64 new_blkno, u32 clusters,
6808                                 u32 *cpos, int num_buckets,
6809                                 struct ocfs2_alloc_context *meta_ac,
6810                                 struct ocfs2_alloc_context *data_ac,
6811                                 struct ocfs2_reflink_xattr_tree_args *args)
6812 {
6813         int i, j, ret = 0;
6814         struct super_block *sb = args->reflink->old_inode->i_sb;
6815         int bpb = args->old_bucket->bu_blocks;
6816         struct ocfs2_xattr_value_buf vb = {
6817                 .vb_access = ocfs2_journal_access,
6818         };
6819
6820         for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6821                 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6822                 if (ret) {
6823                         mlog_errno(ret);
6824                         break;
6825                 }
6826
6827                 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno);
6828                 if (ret) {
6829                         mlog_errno(ret);
6830                         break;
6831                 }
6832
6833                 ret = ocfs2_xattr_bucket_journal_access(handle,
6834                                                 args->new_bucket,
6835                                                 OCFS2_JOURNAL_ACCESS_CREATE);
6836                 if (ret) {
6837                         mlog_errno(ret);
6838                         break;
6839                 }
6840
6841                 for (j = 0; j < bpb; j++)
6842                         memcpy(bucket_block(args->new_bucket, j),
6843                                bucket_block(args->old_bucket, j),
6844                                sb->s_blocksize);
6845
6846                 /*
6847                  * Record the start cpos so that we can use it to initialize
6848                  * our xattr tree we also set the xh_num_bucket for the new
6849                  * bucket.
6850                  */
6851                 if (i == 0) {
6852                         *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6853                                             xh_entries[0].xe_name_hash);
6854                         bucket_xh(args->new_bucket)->xh_num_buckets =
6855                                 cpu_to_le16(num_buckets);
6856                 }
6857
6858                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6859
6860                 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6861                                         args->old_bucket->bu_bhs[0],
6862                                         bucket_xh(args->old_bucket),
6863                                         args->new_bucket->bu_bhs[0],
6864                                         bucket_xh(args->new_bucket),
6865                                         &vb, meta_ac,
6866                                         ocfs2_get_reflink_xattr_value_root,
6867                                         args);
6868                 if (ret) {
6869                         mlog_errno(ret);
6870                         break;
6871                 }
6872
6873                 /*
6874                  * Re-access and dirty the bucket to calculate metaecc.
6875                  * Because we may extend the transaction in reflink_xattr_header
6876                  * which will let the already accessed block gone.
6877                  */
6878                 ret = ocfs2_xattr_bucket_journal_access(handle,
6879                                                 args->new_bucket,
6880                                                 OCFS2_JOURNAL_ACCESS_WRITE);
6881                 if (ret) {
6882                         mlog_errno(ret);
6883                         break;
6884                 }
6885
6886                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6887
6888                 ocfs2_xattr_bucket_relse(args->old_bucket);
6889                 ocfs2_xattr_bucket_relse(args->new_bucket);
6890         }
6891
6892         ocfs2_xattr_bucket_relse(args->old_bucket);
6893         ocfs2_xattr_bucket_relse(args->new_bucket);
6894         return ret;
6895 }
6896
6897 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6898                                 struct inode *inode,
6899                                 struct ocfs2_reflink_xattr_tree_args *args,
6900                                 struct ocfs2_extent_tree *et,
6901                                 struct ocfs2_alloc_context *meta_ac,
6902                                 struct ocfs2_alloc_context *data_ac,
6903                                 u64 blkno, u32 cpos, u32 len)
6904 {
6905         int ret, first_inserted = 0;
6906         u32 p_cluster, num_clusters, reflink_cpos = 0;
6907         u64 new_blkno;
6908         unsigned int num_buckets, reflink_buckets;
6909         unsigned int bpc =
6910                 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6911
6912         ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6913         if (ret) {
6914                 mlog_errno(ret);
6915                 goto out;
6916         }
6917         num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6918         ocfs2_xattr_bucket_relse(args->old_bucket);
6919
6920         while (len && num_buckets) {
6921                 ret = ocfs2_claim_clusters(handle, data_ac,
6922                                            1, &p_cluster, &num_clusters);
6923                 if (ret) {
6924                         mlog_errno(ret);
6925                         goto out;
6926                 }
6927
6928                 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6929                 reflink_buckets = min(num_buckets, bpc * num_clusters);
6930
6931                 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6932                                                  new_blkno, num_clusters,
6933                                                  &reflink_cpos, reflink_buckets,
6934                                                  meta_ac, data_ac, args);
6935                 if (ret) {
6936                         mlog_errno(ret);
6937                         goto out;
6938                 }
6939
6940                 /*
6941                  * For the 1st allocated cluster, we make it use the same cpos
6942                  * so that the xattr tree looks the same as the original one
6943                  * in the most case.
6944                  */
6945                 if (!first_inserted) {
6946                         reflink_cpos = cpos;
6947                         first_inserted = 1;
6948                 }
6949                 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6950                                           num_clusters, 0, meta_ac);
6951                 if (ret)
6952                         mlog_errno(ret);
6953
6954                 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6955                                                   num_clusters, reflink_cpos);
6956
6957                 len -= num_clusters;
6958                 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6959                 num_buckets -= reflink_buckets;
6960         }
6961 out:
6962         return ret;
6963 }
6964
6965 /*
6966  * Create the same xattr extent record in the new inode's xattr tree.
6967  */
6968 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6969                                    struct buffer_head *root_bh,
6970                                    u64 blkno,
6971                                    u32 cpos,
6972                                    u32 len,
6973                                    void *para)
6974 {
6975         int ret, credits = 0;
6976         handle_t *handle;
6977         struct ocfs2_reflink_xattr_tree_args *args =
6978                         (struct ocfs2_reflink_xattr_tree_args *)para;
6979         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6980         struct ocfs2_alloc_context *meta_ac = NULL;
6981         struct ocfs2_alloc_context *data_ac = NULL;
6982         struct ocfs2_extent_tree et;
6983
6984         trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
6985
6986         ocfs2_init_xattr_tree_extent_tree(&et,
6987                                           INODE_CACHE(args->reflink->new_inode),
6988                                           args->new_blk_bh);
6989
6990         ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
6991                                                       len, &credits,
6992                                                       &meta_ac, &data_ac);
6993         if (ret) {
6994                 mlog_errno(ret);
6995                 goto out;
6996         }
6997
6998         handle = ocfs2_start_trans(osb, credits);
6999         if (IS_ERR(handle)) {
7000                 ret = PTR_ERR(handle);
7001                 mlog_errno(ret);
7002                 goto out;
7003         }
7004
7005         ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7006                                           meta_ac, data_ac,
7007                                           blkno, cpos, len);
7008         if (ret)
7009                 mlog_errno(ret);
7010
7011         ocfs2_commit_trans(osb, handle);
7012
7013 out:
7014         if (meta_ac)
7015                 ocfs2_free_alloc_context(meta_ac);
7016         if (data_ac)
7017                 ocfs2_free_alloc_context(data_ac);
7018         return ret;
7019 }
7020
7021 /*
7022  * Create reflinked xattr buckets.
7023  * We will add bucket one by one, and refcount all the xattrs in the bucket
7024  * if they are stored outside.
7025  */
7026 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7027                                     struct buffer_head *blk_bh,
7028                                     struct buffer_head *new_blk_bh)
7029 {
7030         int ret;
7031         struct ocfs2_reflink_xattr_tree_args para;
7032
7033         memset(&para, 0, sizeof(para));
7034         para.reflink = args;
7035         para.old_blk_bh = blk_bh;
7036         para.new_blk_bh = new_blk_bh;
7037
7038         para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7039         if (!para.old_bucket) {
7040                 mlog_errno(-ENOMEM);
7041                 return -ENOMEM;
7042         }
7043
7044         para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7045         if (!para.new_bucket) {
7046                 ret = -ENOMEM;
7047                 mlog_errno(ret);
7048                 goto out;
7049         }
7050
7051         ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7052                                               ocfs2_reflink_xattr_rec,
7053                                               &para);
7054         if (ret)
7055                 mlog_errno(ret);
7056
7057 out:
7058         ocfs2_xattr_bucket_free(para.old_bucket);
7059         ocfs2_xattr_bucket_free(para.new_bucket);
7060         return ret;
7061 }
7062
7063 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7064                                         struct buffer_head *blk_bh)
7065 {
7066         int ret, indexed = 0;
7067         struct buffer_head *new_blk_bh = NULL;
7068         struct ocfs2_xattr_block *xb =
7069                         (struct ocfs2_xattr_block *)blk_bh->b_data;
7070
7071
7072         if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7073                 indexed = 1;
7074
7075         ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7076                                              &new_blk_bh, indexed);
7077         if (ret) {
7078                 mlog_errno(ret);
7079                 goto out;
7080         }
7081
7082         if (!indexed)
7083                 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7084         else
7085                 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7086         if (ret)
7087                 mlog_errno(ret);
7088
7089 out:
7090         brelse(new_blk_bh);
7091         return ret;
7092 }
7093
7094 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7095 {
7096         int type = ocfs2_xattr_get_type(xe);
7097
7098         return type != OCFS2_XATTR_INDEX_SECURITY &&
7099                type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7100                type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7101 }
7102
7103 int ocfs2_reflink_xattrs(struct inode *old_inode,
7104                          struct buffer_head *old_bh,
7105                          struct inode *new_inode,
7106                          struct buffer_head *new_bh,
7107                          bool preserve_security)
7108 {
7109         int ret;
7110         struct ocfs2_xattr_reflink args;
7111         struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7112         struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7113         struct buffer_head *blk_bh = NULL;
7114         struct ocfs2_cached_dealloc_ctxt dealloc;
7115         struct ocfs2_refcount_tree *ref_tree;
7116         struct buffer_head *ref_root_bh = NULL;
7117
7118         ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7119                                        le64_to_cpu(di->i_refcount_loc),
7120                                        1, &ref_tree, &ref_root_bh);
7121         if (ret) {
7122                 mlog_errno(ret);
7123                 goto out;
7124         }
7125
7126         ocfs2_init_dealloc_ctxt(&dealloc);
7127
7128         args.old_inode = old_inode;
7129         args.new_inode = new_inode;
7130         args.old_bh = old_bh;
7131         args.new_bh = new_bh;
7132         args.ref_ci = &ref_tree->rf_ci;
7133         args.ref_root_bh = ref_root_bh;
7134         args.dealloc = &dealloc;
7135         if (preserve_security)
7136                 args.xattr_reflinked = NULL;
7137         else
7138                 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7139
7140         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7141                 ret = ocfs2_reflink_xattr_inline(&args);
7142                 if (ret) {
7143                         mlog_errno(ret);
7144                         goto out_unlock;
7145                 }
7146         }
7147
7148         if (!di->i_xattr_loc)
7149                 goto out_unlock;
7150
7151         ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7152                                      &blk_bh);
7153         if (ret < 0) {
7154                 mlog_errno(ret);
7155                 goto out_unlock;
7156         }
7157
7158         ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7159         if (ret)
7160                 mlog_errno(ret);
7161
7162         brelse(blk_bh);
7163
7164 out_unlock:
7165         ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7166                                    ref_tree, 1);
7167         brelse(ref_root_bh);
7168
7169         if (ocfs2_dealloc_has_cluster(&dealloc)) {
7170                 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7171                 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7172         }
7173
7174 out:
7175         return ret;
7176 }
7177
7178 /*
7179  * Initialize security and acl for a already created inode.
7180  * Used for reflink a non-preserve-security file.
7181  *
7182  * It uses common api like ocfs2_xattr_set, so the caller
7183  * must not hold any lock expect i_mutex.
7184  */
7185 int ocfs2_init_security_and_acl(struct inode *dir,
7186                                 struct inode *inode,
7187                                 const struct qstr *qstr)
7188 {
7189         int ret = 0;
7190         struct buffer_head *dir_bh = NULL;
7191
7192         ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7193         if (ret) {
7194                 mlog_errno(ret);
7195                 goto leave;
7196         }
7197
7198         ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7199         if (ret) {
7200                 mlog_errno(ret);
7201                 goto leave;
7202         }
7203
7204         ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7205         if (ret)
7206                 mlog_errno(ret);
7207
7208         ocfs2_inode_unlock(dir, 0);
7209         brelse(dir_bh);
7210 leave:
7211         return ret;
7212 }
7213 /*
7214  * 'security' attributes support
7215  */
7216 static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
7217                                         size_t list_size, const char *name,
7218                                         size_t name_len, int type)
7219 {
7220         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7221         const size_t total_len = prefix_len + name_len + 1;
7222
7223         if (list && total_len <= list_size) {
7224                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7225                 memcpy(list + prefix_len, name, name_len);
7226                 list[prefix_len + name_len] = '\0';
7227         }
7228         return total_len;
7229 }
7230
7231 static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7232                                     void *buffer, size_t size, int type)
7233 {
7234         if (strcmp(name, "") == 0)
7235                 return -EINVAL;
7236         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7237                                name, buffer, size);
7238 }
7239
7240 static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7241                 const void *value, size_t size, int flags, int type)
7242 {
7243         if (strcmp(name, "") == 0)
7244                 return -EINVAL;
7245
7246         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7247                                name, value, size, flags);
7248 }
7249
7250 int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7251                      void *fs_info)
7252 {
7253         const struct xattr *xattr;
7254         int err = 0;
7255
7256         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7257                 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7258                                       xattr->name, xattr->value,
7259                                       xattr->value_len, XATTR_CREATE);
7260                 if (err)
7261                         break;
7262         }
7263         return err;
7264 }
7265
7266 int ocfs2_init_security_get(struct inode *inode,
7267                             struct inode *dir,
7268                             const struct qstr *qstr,
7269                             struct ocfs2_security_xattr_info *si)
7270 {
7271         /* check whether ocfs2 support feature xattr */
7272         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7273                 return -EOPNOTSUPP;
7274         if (si)
7275                 return security_old_inode_init_security(inode, dir, qstr,
7276                                                         &si->name, &si->value,
7277                                                         &si->value_len);
7278
7279         return security_inode_init_security(inode, dir, qstr,
7280                                             &ocfs2_initxattrs, NULL);
7281 }
7282
7283 int ocfs2_init_security_set(handle_t *handle,
7284                             struct inode *inode,
7285                             struct buffer_head *di_bh,
7286                             struct ocfs2_security_xattr_info *si,
7287                             struct ocfs2_alloc_context *xattr_ac,
7288                             struct ocfs2_alloc_context *data_ac)
7289 {
7290         return ocfs2_xattr_set_handle(handle, inode, di_bh,
7291                                      OCFS2_XATTR_INDEX_SECURITY,
7292                                      si->name, si->value, si->value_len, 0,
7293                                      xattr_ac, data_ac);
7294 }
7295
7296 const struct xattr_handler ocfs2_xattr_security_handler = {
7297         .prefix = XATTR_SECURITY_PREFIX,
7298         .list   = ocfs2_xattr_security_list,
7299         .get    = ocfs2_xattr_security_get,
7300         .set    = ocfs2_xattr_security_set,
7301 };
7302
7303 /*
7304  * 'trusted' attributes support
7305  */
7306 static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
7307                                        size_t list_size, const char *name,
7308                                        size_t name_len, int type)
7309 {
7310         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7311         const size_t total_len = prefix_len + name_len + 1;
7312
7313         if (list && total_len <= list_size) {
7314                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7315                 memcpy(list + prefix_len, name, name_len);
7316                 list[prefix_len + name_len] = '\0';
7317         }
7318         return total_len;
7319 }
7320
7321 static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7322                 void *buffer, size_t size, int type)
7323 {
7324         if (strcmp(name, "") == 0)
7325                 return -EINVAL;
7326         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7327                                name, buffer, size);
7328 }
7329
7330 static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7331                 const void *value, size_t size, int flags, int type)
7332 {
7333         if (strcmp(name, "") == 0)
7334                 return -EINVAL;
7335
7336         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7337                                name, value, size, flags);
7338 }
7339
7340 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7341         .prefix = XATTR_TRUSTED_PREFIX,
7342         .list   = ocfs2_xattr_trusted_list,
7343         .get    = ocfs2_xattr_trusted_get,
7344         .set    = ocfs2_xattr_trusted_set,
7345 };
7346
7347 /*
7348  * 'user' attributes support
7349  */
7350 static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
7351                                     size_t list_size, const char *name,
7352                                     size_t name_len, int type)
7353 {
7354         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7355         const size_t total_len = prefix_len + name_len + 1;
7356         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7357
7358         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7359                 return 0;
7360
7361         if (list && total_len <= list_size) {
7362                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7363                 memcpy(list + prefix_len, name, name_len);
7364                 list[prefix_len + name_len] = '\0';
7365         }
7366         return total_len;
7367 }
7368
7369 static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7370                 void *buffer, size_t size, int type)
7371 {
7372         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7373
7374         if (strcmp(name, "") == 0)
7375                 return -EINVAL;
7376         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7377                 return -EOPNOTSUPP;
7378         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
7379                                buffer, size);
7380 }
7381
7382 static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7383                 const void *value, size_t size, int flags, int type)
7384 {
7385         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7386
7387         if (strcmp(name, "") == 0)
7388                 return -EINVAL;
7389         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7390                 return -EOPNOTSUPP;
7391
7392         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7393                                name, value, size, flags);
7394 }
7395
7396 const struct xattr_handler ocfs2_xattr_user_handler = {
7397         .prefix = XATTR_USER_PREFIX,
7398         .list   = ocfs2_xattr_user_list,
7399         .get    = ocfs2_xattr_user_get,
7400         .set    = ocfs2_xattr_user_set,
7401 };