linux/fs/ocfs2/xattr.c
<<
>>
Prefs
   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
  61struct ocfs2_xattr_def_value_root {
  62        struct ocfs2_xattr_value_root   xv;
  63        struct ocfs2_extent_rec         er;
  64};
  65
  66struct 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
  77struct 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
  96static struct ocfs2_xattr_def_value_root def_xv = {
  97        .xv.xr_list.l_count = cpu_to_le16(1),
  98};
  99
 100const 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
 109static 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
 119struct 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
 127struct 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 */
 143struct ocfs2_xa_loc;
 144struct 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 */
 196struct 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 */
 228static 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
 236static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
 237{
 238        return namevalue_size(xi->xi_name_len, xi->xi_value_len);
 239}
 240
 241static 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
 251static 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
 257static int ocfs2_xattr_block_find(struct inode *inode,
 258                                  int name_index,
 259                                  const char *name,
 260                                  struct ocfs2_xattr_search *xs);
 261static 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
 267static 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
 272static int ocfs2_xattr_create_index_block(struct inode *inode,
 273                                          struct ocfs2_xattr_search *xs,
 274                                          struct ocfs2_xattr_set_ctxt *ctxt);
 275
 276static 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
 281typedef int (xattr_tree_rec_func)(struct inode *inode,
 282                                  struct buffer_head *root_bh,
 283                                  u64 blkno, u32 cpos, u32 len, void *para);
 284static 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);
 288static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
 289                                        struct ocfs2_xattr_bucket *bucket,
 290                                        void *para);
 291static 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
 298static 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);
 302static 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);
 310static 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
 316static 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
 321static 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
 330static 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
 346static 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
 356static 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 */
 371static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 372                                   u64 xb_blkno)
 373{
 374        int i, rc = 0;
 375
 376        for (i = 0; i < bucket->bu_blocks; i++) {
 377                bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 378                                              xb_blkno + i);
 379                if (!bucket->bu_bhs[i]) {
 380                        rc = -EIO;
 381                        mlog_errno(rc);
 382                        break;
 383                }
 384
 385                if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 386                                           bucket->bu_bhs[i]))
 387                        ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 388                                                      bucket->bu_bhs[i]);
 389        }
 390
 391        if (rc)
 392                ocfs2_xattr_bucket_relse(bucket);
 393        return rc;
 394}
 395
 396/* Read the xattr bucket at xb_blkno */
 397static 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
 421static 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
 440static 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
 455static 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
 470static 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
 524static 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
 540static 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
 550static 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
 568static 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
 574static 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
 580static 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
 586int 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
 624int 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
 706static int ocfs2_xattr_extend_allocation(struct inode *inode,
 707                                         u32 clusters_to_add,
 708                                         struct ocfs2_xattr_value_buf *vb,
 709                                         struct ocfs2_xattr_set_ctxt *ctxt)
 710{
 711        int status = 0, credits;
 712        handle_t *handle = ctxt->handle;
 713        enum ocfs2_alloc_restarted why;
 714        u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 715        struct ocfs2_extent_tree et;
 716
 717        ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 718
 719        while (clusters_to_add) {
 720                trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 721
 722                status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 723                                       OCFS2_JOURNAL_ACCESS_WRITE);
 724                if (status < 0) {
 725                        mlog_errno(status);
 726                        break;
 727                }
 728
 729                prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 730                status = ocfs2_add_clusters_in_btree(handle,
 731                                                     &et,
 732                                                     &logical_start,
 733                                                     clusters_to_add,
 734                                                     0,
 735                                                     ctxt->data_ac,
 736                                                     ctxt->meta_ac,
 737                                                     &why);
 738                if ((status < 0) && (status != -EAGAIN)) {
 739                        if (status != -ENOSPC)
 740                                mlog_errno(status);
 741                        break;
 742                }
 743
 744                ocfs2_journal_dirty(handle, vb->vb_bh);
 745
 746                clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 747                                         prev_clusters;
 748
 749                if (why != RESTART_NONE && clusters_to_add) {
 750                        /*
 751                         * We can only fail in case the alloc file doesn't give
 752                         * up enough clusters.
 753                         */
 754                        BUG_ON(why == RESTART_META);
 755
 756                        credits = ocfs2_calc_extend_credits(inode->i_sb,
 757                                                            &vb->vb_xv->xr_list,
 758                                                            clusters_to_add);
 759                        status = ocfs2_extend_trans(handle, credits);
 760                        if (status < 0) {
 761                                status = -ENOMEM;
 762                                mlog_errno(status);
 763                                break;
 764                        }
 765                }
 766        }
 767
 768        return status;
 769}
 770
 771static int __ocfs2_remove_xattr_range(struct inode *inode,
 772                                      struct ocfs2_xattr_value_buf *vb,
 773                                      u32 cpos, u32 phys_cpos, u32 len,
 774                                      unsigned int ext_flags,
 775                                      struct ocfs2_xattr_set_ctxt *ctxt)
 776{
 777        int ret;
 778        u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 779        handle_t *handle = ctxt->handle;
 780        struct ocfs2_extent_tree et;
 781
 782        ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 783
 784        ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 785                            OCFS2_JOURNAL_ACCESS_WRITE);
 786        if (ret) {
 787                mlog_errno(ret);
 788                goto out;
 789        }
 790
 791        ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 792                                  &ctxt->dealloc);
 793        if (ret) {
 794                mlog_errno(ret);
 795                goto out;
 796        }
 797
 798        le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 799        ocfs2_journal_dirty(handle, vb->vb_bh);
 800
 801        if (ext_flags & OCFS2_EXT_REFCOUNTED)
 802                ret = ocfs2_decrease_refcount(inode, handle,
 803                                        ocfs2_blocks_to_clusters(inode->i_sb,
 804                                                                 phys_blkno),
 805                                        len, ctxt->meta_ac, &ctxt->dealloc, 1);
 806        else
 807                ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 808                                                  phys_blkno, len);
 809        if (ret)
 810                mlog_errno(ret);
 811
 812out:
 813        return ret;
 814}
 815
 816static int ocfs2_xattr_shrink_size(struct inode *inode,
 817                                   u32 old_clusters,
 818                                   u32 new_clusters,
 819                                   struct ocfs2_xattr_value_buf *vb,
 820                                   struct ocfs2_xattr_set_ctxt *ctxt)
 821{
 822        int ret = 0;
 823        unsigned int ext_flags;
 824        u32 trunc_len, cpos, phys_cpos, alloc_size;
 825        u64 block;
 826
 827        if (old_clusters <= new_clusters)
 828                return 0;
 829
 830        cpos = new_clusters;
 831        trunc_len = old_clusters - new_clusters;
 832        while (trunc_len) {
 833                ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 834                                               &alloc_size,
 835                                               &vb->vb_xv->xr_list, &ext_flags);
 836                if (ret) {
 837                        mlog_errno(ret);
 838                        goto out;
 839                }
 840
 841                if (alloc_size > trunc_len)
 842                        alloc_size = trunc_len;
 843
 844                ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 845                                                 phys_cpos, alloc_size,
 846                                                 ext_flags, ctxt);
 847                if (ret) {
 848                        mlog_errno(ret);
 849                        goto out;
 850                }
 851
 852                block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 853                ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 854                                                       block, alloc_size);
 855                cpos += alloc_size;
 856                trunc_len -= alloc_size;
 857        }
 858
 859out:
 860        return ret;
 861}
 862
 863static int ocfs2_xattr_value_truncate(struct inode *inode,
 864                                      struct ocfs2_xattr_value_buf *vb,
 865                                      int len,
 866                                      struct ocfs2_xattr_set_ctxt *ctxt)
 867{
 868        int ret;
 869        u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 870        u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 871
 872        if (new_clusters == old_clusters)
 873                return 0;
 874
 875        if (new_clusters > old_clusters)
 876                ret = ocfs2_xattr_extend_allocation(inode,
 877                                                    new_clusters - old_clusters,
 878                                                    vb, ctxt);
 879        else
 880                ret = ocfs2_xattr_shrink_size(inode,
 881                                              old_clusters, new_clusters,
 882                                              vb, ctxt);
 883
 884        return ret;
 885}
 886
 887static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 888                                  size_t *result, const char *prefix,
 889                                  const char *name, int name_len)
 890{
 891        char *p = buffer + *result;
 892        int prefix_len = strlen(prefix);
 893        int total_len = prefix_len + name_len + 1;
 894
 895        *result += total_len;
 896
 897        /* we are just looking for how big our buffer needs to be */
 898        if (!size)
 899                return 0;
 900
 901        if (*result > size)
 902                return -ERANGE;
 903
 904        memcpy(p, prefix, prefix_len);
 905        memcpy(p + prefix_len, name, name_len);
 906        p[prefix_len + name_len] = '\0';
 907
 908        return 0;
 909}
 910
 911static int ocfs2_xattr_list_entries(struct inode *inode,
 912                                    struct ocfs2_xattr_header *header,
 913                                    char *buffer, size_t buffer_size)
 914{
 915        size_t result = 0;
 916        int i, type, ret;
 917        const char *prefix, *name;
 918
 919        for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 920                struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 921                type = ocfs2_xattr_get_type(entry);
 922                prefix = ocfs2_xattr_prefix(type);
 923
 924                if (prefix) {
 925                        name = (const char *)header +
 926                                le16_to_cpu(entry->xe_name_offset);
 927
 928                        ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 929                                                     &result, prefix, name,
 930                                                     entry->xe_name_len);
 931                        if (ret)
 932                                return ret;
 933                }
 934        }
 935
 936        return result;
 937}
 938
 939int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 940                                         struct ocfs2_dinode *di)
 941{
 942        struct ocfs2_xattr_header *xh;
 943        int i;
 944
 945        xh = (struct ocfs2_xattr_header *)
 946                 ((void *)di + inode->i_sb->s_blocksize -
 947                 le16_to_cpu(di->i_xattr_inline_size));
 948
 949        for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 950                if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 951                        return 1;
 952
 953        return 0;
 954}
 955
 956static int ocfs2_xattr_ibody_list(struct inode *inode,
 957                                  struct ocfs2_dinode *di,
 958                                  char *buffer,
 959                                  size_t buffer_size)
 960{
 961        struct ocfs2_xattr_header *header = NULL;
 962        struct ocfs2_inode_info *oi = OCFS2_I(inode);
 963        int ret = 0;
 964
 965        if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 966                return ret;
 967
 968        header = (struct ocfs2_xattr_header *)
 969                 ((void *)di + inode->i_sb->s_blocksize -
 970                 le16_to_cpu(di->i_xattr_inline_size));
 971
 972        ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 973
 974        return ret;
 975}
 976
 977static int ocfs2_xattr_block_list(struct inode *inode,
 978                                  struct ocfs2_dinode *di,
 979                                  char *buffer,
 980                                  size_t buffer_size)
 981{
 982        struct buffer_head *blk_bh = NULL;
 983        struct ocfs2_xattr_block *xb;
 984        int ret = 0;
 985
 986        if (!di->i_xattr_loc)
 987                return ret;
 988
 989        ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 990                                     &blk_bh);
 991        if (ret < 0) {
 992                mlog_errno(ret);
 993                return ret;
 994        }
 995
 996        xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 997        if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 998                struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 999                ret = ocfs2_xattr_list_entries(inode, header,
1000                                               buffer, buffer_size);
1001        } else
1002                ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1003                                                   buffer, buffer_size);
1004
1005        brelse(blk_bh);
1006
1007        return ret;
1008}
1009
1010ssize_t ocfs2_listxattr(struct dentry *dentry,
1011                        char *buffer,
1012                        size_t size)
1013{
1014        int ret = 0, i_ret = 0, b_ret = 0;
1015        struct buffer_head *di_bh = NULL;
1016        struct ocfs2_dinode *di = NULL;
1017        struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1018
1019        if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1020                return -EOPNOTSUPP;
1021
1022        if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1023                return ret;
1024
1025        ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1026        if (ret < 0) {
1027                mlog_errno(ret);
1028                return ret;
1029        }
1030
1031        di = (struct ocfs2_dinode *)di_bh->b_data;
1032
1033        down_read(&oi->ip_xattr_sem);
1034        i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1035        if (i_ret < 0)
1036                b_ret = 0;
1037        else {
1038                if (buffer) {
1039                        buffer += i_ret;
1040                        size -= i_ret;
1041                }
1042                b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1043                                               buffer, size);
1044                if (b_ret < 0)
1045                        i_ret = 0;
1046        }
1047        up_read(&oi->ip_xattr_sem);
1048        ocfs2_inode_unlock(dentry->d_inode, 0);
1049
1050        brelse(di_bh);
1051
1052        return i_ret + b_ret;
1053}
1054
1055static int ocfs2_xattr_find_entry(int name_index,
1056                                  const char *name,
1057                                  struct ocfs2_xattr_search *xs)
1058{
1059        struct ocfs2_xattr_entry *entry;
1060        size_t name_len;
1061        int i, cmp = 1;
1062
1063        if (name == NULL)
1064                return -EINVAL;
1065
1066        name_len = strlen(name);
1067        entry = xs->here;
1068        for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1069                cmp = name_index - ocfs2_xattr_get_type(entry);
1070                if (!cmp)
1071                        cmp = name_len - entry->xe_name_len;
1072                if (!cmp)
1073                        cmp = memcmp(name, (xs->base +
1074                                     le16_to_cpu(entry->xe_name_offset)),
1075                                     name_len);
1076                if (cmp == 0)
1077                        break;
1078                entry += 1;
1079        }
1080        xs->here = entry;
1081
1082        return cmp ? -ENODATA : 0;
1083}
1084
1085static int ocfs2_xattr_get_value_outside(struct inode *inode,
1086                                         struct ocfs2_xattr_value_root *xv,
1087                                         void *buffer,
1088                                         size_t len)
1089{
1090        u32 cpos, p_cluster, num_clusters, bpc, clusters;
1091        u64 blkno;
1092        int i, ret = 0;
1093        size_t cplen, blocksize;
1094        struct buffer_head *bh = NULL;
1095        struct ocfs2_extent_list *el;
1096
1097        el = &xv->xr_list;
1098        clusters = le32_to_cpu(xv->xr_clusters);
1099        bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1100        blocksize = inode->i_sb->s_blocksize;
1101
1102        cpos = 0;
1103        while (cpos < clusters) {
1104                ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1105                                               &num_clusters, el, NULL);
1106                if (ret) {
1107                        mlog_errno(ret);
1108                        goto out;
1109                }
1110
1111                blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1112                /* Copy ocfs2_xattr_value */
1113                for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1114                        ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1115                                               &bh, NULL);
1116                        if (ret) {
1117                                mlog_errno(ret);
1118                                goto out;
1119                        }
1120
1121                        cplen = len >= blocksize ? blocksize : len;
1122                        memcpy(buffer, bh->b_data, cplen);
1123                        len -= cplen;
1124                        buffer += cplen;
1125
1126                        brelse(bh);
1127                        bh = NULL;
1128                        if (len == 0)
1129                                break;
1130                }
1131                cpos += num_clusters;
1132        }
1133out:
1134        return ret;
1135}
1136
1137static int ocfs2_xattr_ibody_get(struct inode *inode,
1138                                 int name_index,
1139                                 const char *name,
1140                                 void *buffer,
1141                                 size_t buffer_size,
1142                                 struct ocfs2_xattr_search *xs)
1143{
1144        struct ocfs2_inode_info *oi = OCFS2_I(inode);
1145        struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1146        struct ocfs2_xattr_value_root *xv;
1147        size_t size;
1148        int ret = 0;
1149
1150        if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1151                return -ENODATA;
1152
1153        xs->end = (void *)di + inode->i_sb->s_blocksize;
1154        xs->header = (struct ocfs2_xattr_header *)
1155                        (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1156        xs->base = (void *)xs->header;
1157        xs->here = xs->header->xh_entries;
1158
1159        ret = ocfs2_xattr_find_entry(name_index, name, xs);
1160        if (ret)
1161                return ret;
1162        size = le64_to_cpu(xs->here->xe_value_size);
1163        if (buffer) {
1164                if (size > buffer_size)
1165                        return -ERANGE;
1166                if (ocfs2_xattr_is_local(xs->here)) {
1167                        memcpy(buffer, (void *)xs->base +
1168                               le16_to_cpu(xs->here->xe_name_offset) +
1169                               OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1170                } else {
1171                        xv = (struct ocfs2_xattr_value_root *)
1172                                (xs->base + le16_to_cpu(
1173                                 xs->here->xe_name_offset) +
1174                                OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1175                        ret = ocfs2_xattr_get_value_outside(inode, xv,
1176                                                            buffer, size);
1177                        if (ret < 0) {
1178                                mlog_errno(ret);
1179                                return ret;
1180                        }
1181                }
1182        }
1183
1184        return size;
1185}
1186
1187static int ocfs2_xattr_block_get(struct inode *inode,
1188                                 int name_index,
1189                                 const char *name,
1190                                 void *buffer,
1191                                 size_t buffer_size,
1192                                 struct ocfs2_xattr_search *xs)
1193{
1194        struct ocfs2_xattr_block *xb;
1195        struct ocfs2_xattr_value_root *xv;
1196        size_t size;
1197        int ret = -ENODATA, name_offset, name_len, i;
1198        int uninitialized_var(block_off);
1199
1200        xs->bucket = ocfs2_xattr_bucket_new(inode);
1201        if (!xs->bucket) {
1202                ret = -ENOMEM;
1203                mlog_errno(ret);
1204                goto cleanup;
1205        }
1206
1207        ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1208        if (ret) {
1209                mlog_errno(ret);
1210                goto cleanup;
1211        }
1212
1213        if (xs->not_found) {
1214                ret = -ENODATA;
1215                goto cleanup;
1216        }
1217
1218        xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1219        size = le64_to_cpu(xs->here->xe_value_size);
1220        if (buffer) {
1221                ret = -ERANGE;
1222                if (size > buffer_size)
1223                        goto cleanup;
1224
1225                name_offset = le16_to_cpu(xs->here->xe_name_offset);
1226                name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1227                i = xs->here - xs->header->xh_entries;
1228
1229                if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1230                        ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1231                                                                bucket_xh(xs->bucket),
1232                                                                i,
1233                                                                &block_off,
1234                                                                &name_offset);
1235                        xs->base = bucket_block(xs->bucket, block_off);
1236                }
1237                if (ocfs2_xattr_is_local(xs->here)) {
1238                        memcpy(buffer, (void *)xs->base +
1239                               name_offset + name_len, size);
1240                } else {
1241                        xv = (struct ocfs2_xattr_value_root *)
1242                                (xs->base + name_offset + name_len);
1243                        ret = ocfs2_xattr_get_value_outside(inode, xv,
1244                                                            buffer, size);
1245                        if (ret < 0) {
1246                                mlog_errno(ret);
1247                                goto cleanup;
1248                        }
1249                }
1250        }
1251        ret = size;
1252cleanup:
1253        ocfs2_xattr_bucket_free(xs->bucket);
1254
1255        brelse(xs->xattr_bh);
1256        xs->xattr_bh = NULL;
1257        return ret;
1258}
1259
1260int ocfs2_xattr_get_nolock(struct inode *inode,
1261                           struct buffer_head *di_bh,
1262                           int name_index,
1263                           const char *name,
1264                           void *buffer,
1265                           size_t buffer_size)
1266{
1267        int ret;
1268        struct ocfs2_dinode *di = NULL;
1269        struct ocfs2_inode_info *oi = OCFS2_I(inode);
1270        struct ocfs2_xattr_search xis = {
1271                .not_found = -ENODATA,
1272        };
1273        struct ocfs2_xattr_search xbs = {
1274                .not_found = -ENODATA,
1275        };
1276
1277        if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1278                return -EOPNOTSUPP;
1279
1280        if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1281                ret = -ENODATA;
1282
1283        xis.inode_bh = xbs.inode_bh = di_bh;
1284        di = (struct ocfs2_dinode *)di_bh->b_data;
1285
1286        ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1287                                    buffer_size, &xis);
1288        if (ret == -ENODATA && di->i_xattr_loc)
1289                ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1290                                            buffer_size, &xbs);
1291
1292        return ret;
1293}
1294
1295/* ocfs2_xattr_get()
1296 *
1297 * Copy an extended attribute into the buffer provided.
1298 * Buffer is NULL to compute the size of buffer required.
1299 */
1300static int ocfs2_xattr_get(struct inode *inode,
1301                           int name_index,
1302                           const char *name,
1303                           void *buffer,
1304                           size_t buffer_size)
1305{
1306        int ret;
1307        struct buffer_head *di_bh = NULL;
1308
1309        ret = ocfs2_inode_lock(inode, &di_bh, 0);
1310        if (ret < 0) {
1311                mlog_errno(ret);
1312                return ret;
1313        }
1314        down_read(&OCFS2_I(inode)->ip_xattr_sem);
1315        ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1316                                     name, buffer, buffer_size);
1317        up_read(&OCFS2_I(inode)->ip_xattr_sem);
1318
1319        ocfs2_inode_unlock(inode, 0);
1320
1321        brelse(di_bh);
1322
1323        return ret;
1324}
1325
1326static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1327                                           handle_t *handle,
1328                                           struct ocfs2_xattr_value_buf *vb,
1329                                           const void *value,
1330                                           int value_len)
1331{
1332        int ret = 0, i, cp_len;
1333        u16 blocksize = inode->i_sb->s_blocksize;
1334        u32 p_cluster, num_clusters;
1335        u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1336        u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1337        u64 blkno;
1338        struct buffer_head *bh = NULL;
1339        unsigned int ext_flags;
1340        struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1341
1342        BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1343
1344        while (cpos < clusters) {
1345                ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1346                                               &num_clusters, &xv->xr_list,
1347                                               &ext_flags);
1348                if (ret) {
1349                        mlog_errno(ret);
1350                        goto out;
1351                }
1352
1353                BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1354
1355                blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1356
1357                for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1358                        ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1359                                               &bh, NULL);
1360                        if (ret) {
1361                                mlog_errno(ret);
1362                                goto out;
1363                        }
1364
1365                        ret = ocfs2_journal_access(handle,
1366                                                   INODE_CACHE(inode),
1367                                                   bh,
1368                                                   OCFS2_JOURNAL_ACCESS_WRITE);
1369                        if (ret < 0) {
1370                                mlog_errno(ret);
1371                                goto out;
1372                        }
1373
1374                        cp_len = value_len > blocksize ? blocksize : value_len;
1375                        memcpy(bh->b_data, value, cp_len);
1376                        value_len -= cp_len;
1377                        value += cp_len;
1378                        if (cp_len < blocksize)
1379                                memset(bh->b_data + cp_len, 0,
1380                                       blocksize - cp_len);
1381
1382                        ocfs2_journal_dirty(handle, bh);
1383                        brelse(bh);
1384                        bh = NULL;
1385
1386                        /*
1387                         * XXX: do we need to empty all the following
1388                         * blocks in this cluster?
1389                         */
1390                        if (!value_len)
1391                                break;
1392                }
1393                cpos += num_clusters;
1394        }
1395out:
1396        brelse(bh);
1397
1398        return ret;
1399}
1400
1401static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1402                                       int num_entries)
1403{
1404        int free_space;
1405
1406        if (!needed_space)
1407                return 0;
1408
1409        free_space = free_start -
1410                sizeof(struct ocfs2_xattr_header) -
1411                (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1412                OCFS2_XATTR_HEADER_GAP;
1413        if (free_space < 0)
1414                return -EIO;
1415        if (free_space < needed_space)
1416                return -ENOSPC;
1417
1418        return 0;
1419}
1420
1421static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1422                                   int type)
1423{
1424        return loc->xl_ops->xlo_journal_access(handle, loc, type);
1425}
1426
1427static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1428{
1429        loc->xl_ops->xlo_journal_dirty(handle, loc);
1430}
1431
1432/* Give a pointer into the storage for the given offset */
1433static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1434{
1435        BUG_ON(offset >= loc->xl_size);
1436        return loc->xl_ops->xlo_offset_pointer(loc, offset);
1437}
1438
1439/*
1440 * Wipe the name+value pair and allow the storage to reclaim it.  This
1441 * must be followed by either removal of the entry or a call to
1442 * ocfs2_xa_add_namevalue().
1443 */
1444static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1445{
1446        loc->xl_ops->xlo_wipe_namevalue(loc);
1447}
1448
1449/*
1450 * Find lowest offset to a name+value pair.  This is the start of our
1451 * downward-growing free space.
1452 */
1453static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1454{
1455        return loc->xl_ops->xlo_get_free_start(loc);
1456}
1457
1458/* Can we reuse loc->xl_entry for xi? */
1459static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1460                                    struct ocfs2_xattr_info *xi)
1461{
1462        return loc->xl_ops->xlo_can_reuse(loc, xi);
1463}
1464
1465/* How much free space is needed to set the new value */
1466static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1467                                struct ocfs2_xattr_info *xi)
1468{
1469        return loc->xl_ops->xlo_check_space(loc, xi);
1470}
1471
1472static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1473{
1474        loc->xl_ops->xlo_add_entry(loc, name_hash);
1475        loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1476        /*
1477         * We can't leave the new entry's xe_name_offset at zero or
1478         * add_namevalue() will go nuts.  We set it to the size of our
1479         * storage so that it can never be less than any other entry.
1480         */
1481        loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1482}
1483
1484static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1485                                   struct ocfs2_xattr_info *xi)
1486{
1487        int size = namevalue_size_xi(xi);
1488        int nameval_offset;
1489        char *nameval_buf;
1490
1491        loc->xl_ops->xlo_add_namevalue(loc, size);
1492        loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1493        loc->xl_entry->xe_name_len = xi->xi_name_len;
1494        ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1495        ocfs2_xattr_set_local(loc->xl_entry,
1496                              xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1497
1498        nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1499        nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1500        memset(nameval_buf, 0, size);
1501        memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1502}
1503
1504static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1505                                    struct ocfs2_xattr_value_buf *vb)
1506{
1507        int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1508        int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1509
1510        /* Value bufs are for value trees */
1511        BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1512        BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1513               (name_size + OCFS2_XATTR_ROOT_SIZE));
1514
1515        loc->xl_ops->xlo_fill_value_buf(loc, vb);
1516        vb->vb_xv =
1517                (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1518                                                        nameval_offset +
1519                                                        name_size);
1520}
1521
1522static int ocfs2_xa_block_journal_access(handle_t *handle,
1523                                         struct ocfs2_xa_loc *loc, int type)
1524{
1525        struct buffer_head *bh = loc->xl_storage;
1526        ocfs2_journal_access_func access;
1527
1528        if (loc->xl_size == (bh->b_size -
1529                             offsetof(struct ocfs2_xattr_block,
1530                                      xb_attrs.xb_header)))
1531                access = ocfs2_journal_access_xb;
1532        else
1533                access = ocfs2_journal_access_di;
1534        return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1535}
1536
1537static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1538                                         struct ocfs2_xa_loc *loc)
1539{
1540        struct buffer_head *bh = loc->xl_storage;
1541
1542        ocfs2_journal_dirty(handle, bh);
1543}
1544
1545static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1546                                           int offset)
1547{
1548        return (char *)loc->xl_header + offset;
1549}
1550
1551static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1552                                    struct ocfs2_xattr_info *xi)
1553{
1554        /*
1555         * Block storage is strict.  If the sizes aren't exact, we will
1556         * remove the old one and reinsert the new.
1557         */
1558        return namevalue_size_xe(loc->xl_entry) ==
1559                namevalue_size_xi(xi);
1560}
1561
1562static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1563{
1564        struct ocfs2_xattr_header *xh = loc->xl_header;
1565        int i, count = le16_to_cpu(xh->xh_count);
1566        int offset, free_start = loc->xl_size;
1567
1568        for (i = 0; i < count; i++) {
1569                offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1570                if (offset < free_start)
1571                        free_start = offset;
1572        }
1573
1574        return free_start;
1575}
1576
1577static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1578                                      struct ocfs2_xattr_info *xi)
1579{
1580        int count = le16_to_cpu(loc->xl_header->xh_count);
1581        int free_start = ocfs2_xa_get_free_start(loc);
1582        int needed_space = ocfs2_xi_entry_usage(xi);
1583
1584        /*
1585         * Block storage will reclaim the original entry before inserting
1586         * the new value, so we only need the difference.  If the new
1587         * entry is smaller than the old one, we don't need anything.
1588         */
1589        if (loc->xl_entry) {
1590                /* Don't need space if we're reusing! */
1591                if (ocfs2_xa_can_reuse_entry(loc, xi))
1592                        needed_space = 0;
1593                else
1594                        needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1595        }
1596        if (needed_space < 0)
1597                needed_space = 0;
1598        return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1599}
1600
1601/*
1602 * Block storage for xattrs keeps the name+value pairs compacted.  When
1603 * we remove one, we have to shift any that preceded it towards the end.
1604 */
1605static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1606{
1607        int i, offset;
1608        int namevalue_offset, first_namevalue_offset, namevalue_size;
1609        struct ocfs2_xattr_entry *entry = loc->xl_entry;
1610        struct ocfs2_xattr_header *xh = loc->xl_header;
1611        int count = le16_to_cpu(xh->xh_count);
1612
1613        namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1614        namevalue_size = namevalue_size_xe(entry);
1615        first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1616
1617        /* Shift the name+value pairs */
1618        memmove((char *)xh + first_namevalue_offset + namevalue_size,
1619                (char *)xh + first_namevalue_offset,
1620                namevalue_offset - first_namevalue_offset);
1621        memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1622
1623        /* Now tell xh->xh_entries about it */
1624        for (i = 0; i < count; i++) {
1625                offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1626                if (offset <= namevalue_offset)
1627                        le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1628                                     namevalue_size);
1629        }
1630
1631        /*
1632         * Note that we don't update xh_free_start or xh_name_value_len
1633         * because they're not used in block-stored xattrs.
1634         */
1635}
1636
1637static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1638{
1639        int count = le16_to_cpu(loc->xl_header->xh_count);
1640        loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1641        le16_add_cpu(&loc->xl_header->xh_count, 1);
1642        memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1643}
1644
1645static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1646{
1647        int free_start = ocfs2_xa_get_free_start(loc);
1648
1649        loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1650}
1651
1652static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1653                                          struct ocfs2_xattr_value_buf *vb)
1654{
1655        struct buffer_head *bh = loc->xl_storage;
1656
1657        if (loc->xl_size == (bh->b_size -
1658                             offsetof(struct ocfs2_xattr_block,
1659                                      xb_attrs.xb_header)))
1660                vb->vb_access = ocfs2_journal_access_xb;
1661        else
1662                vb->vb_access = ocfs2_journal_access_di;
1663        vb->vb_bh = bh;
1664}
1665
1666/*
1667 * Operations for xattrs stored in blocks.  This includes inline inode
1668 * storage and unindexed ocfs2_xattr_blocks.
1669 */
1670static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1671        .xlo_journal_access     = ocfs2_xa_block_journal_access,
1672        .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1673        .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1674        .xlo_check_space        = ocfs2_xa_block_check_space,
1675        .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1676        .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1677        .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1678        .xlo_add_entry          = ocfs2_xa_block_add_entry,
1679        .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1680        .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1681};
1682
1683static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1684                                          struct ocfs2_xa_loc *loc, int type)
1685{
1686        struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1687
1688        return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1689}
1690
1691static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1692                                          struct ocfs2_xa_loc *loc)
1693{
1694        struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1695
1696        ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1697}
1698
1699static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1700                                            int offset)
1701{
1702        struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1703        int block, block_offset;
1704
1705        /* The header is at the front of the bucket */
1706        block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1707        block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1708
1709        return bucket_block(bucket, block) + block_offset;
1710}
1711
1712static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1713                                     struct ocfs2_xattr_info *xi)
1714{
1715        return namevalue_size_xe(loc->xl_entry) >=
1716                namevalue_size_xi(xi);
1717}
1718
1719static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1720{
1721        struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1722        return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1723}
1724
1725static int ocfs2_bucket_align_free_start(struct super_block *sb,
1726                                         int free_start, int size)
1727{
1728        /*
1729         * We need to make sure that the name+value pair fits within
1730         * one block.
1731         */
1732        if (((free_start - size) >> sb->s_blocksize_bits) !=
1733            ((free_start - 1) >> sb->s_blocksize_bits))
1734                free_start -= free_start % sb->s_blocksize;
1735
1736        return free_start;
1737}
1738
1739static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1740                                       struct ocfs2_xattr_info *xi)
1741{
1742        int rc;
1743        int count = le16_to_cpu(loc->xl_header->xh_count);
1744        int free_start = ocfs2_xa_get_free_start(loc);
1745        int needed_space = ocfs2_xi_entry_usage(xi);
1746        int size = namevalue_size_xi(xi);
1747        struct super_block *sb = loc->xl_inode->i_sb;
1748
1749        /*
1750         * Bucket storage does not reclaim name+value pairs it cannot
1751         * reuse.  They live as holes until the bucket fills, and then
1752         * the bucket is defragmented.  However, the bucket can reclaim
1753         * the ocfs2_xattr_entry.
1754         */
1755        if (loc->xl_entry) {
1756                /* Don't need space if we're reusing! */
1757                if (ocfs2_xa_can_reuse_entry(loc, xi))
1758                        needed_space = 0;
1759                else
1760                        needed_space -= sizeof(struct ocfs2_xattr_entry);
1761        }
1762        BUG_ON(needed_space < 0);
1763
1764        if (free_start < size) {
1765                if (needed_space)
1766                        return -ENOSPC;
1767        } else {
1768                /*
1769                 * First we check if it would fit in the first place.
1770                 * Below, we align the free start to a block.  This may
1771                 * slide us below the minimum gap.  By checking unaligned
1772                 * first, we avoid that error.
1773                 */
1774                rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1775                                                 count);
1776                if (rc)
1777                        return rc;
1778                free_start = ocfs2_bucket_align_free_start(sb, free_start,
1779                                                           size);
1780        }
1781        return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1782}
1783
1784static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1785{
1786        le16_add_cpu(&loc->xl_header->xh_name_value_len,
1787                     -namevalue_size_xe(loc->xl_entry));
1788}
1789
1790static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1791{
1792        struct ocfs2_xattr_header *xh = loc->xl_header;
1793        int count = le16_to_cpu(xh->xh_count);
1794        int low = 0, high = count - 1, tmp;
1795        struct ocfs2_xattr_entry *tmp_xe;
1796
1797        /*
1798         * We keep buckets sorted by name_hash, so we need to find
1799         * our insert place.
1800         */
1801        while (low <= high && count) {
1802                tmp = (low + high) / 2;
1803                tmp_xe = &xh->xh_entries[tmp];
1804
1805                if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1806                        low = tmp + 1;
1807                else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1808                        high = tmp - 1;
1809                else {
1810                        low = tmp;
1811                        break;
1812                }
1813        }
1814
1815        if (low != count)
1816                memmove(&xh->xh_entries[low + 1],
1817                        &xh->xh_entries[low],
1818                        ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1819
1820        le16_add_cpu(&xh->xh_count, 1);
1821        loc->xl_entry = &xh->xh_entries[low];
1822        memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1823}
1824
1825static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1826{
1827        int free_start = ocfs2_xa_get_free_start(loc);
1828        struct ocfs2_xattr_header *xh = loc->xl_header;
1829        struct super_block *sb = loc->xl_inode->i_sb;
1830        int nameval_offset;
1831
1832        free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1833        nameval_offset = free_start - size;
1834        loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1835        xh->xh_free_start = cpu_to_le16(nameval_offset);
1836        le16_add_cpu(&xh->xh_name_value_len, size);
1837
1838}
1839
1840static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1841                                           struct ocfs2_xattr_value_buf *vb)
1842{
1843        struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1844        struct super_block *sb = loc->xl_inode->i_sb;
1845        int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1846        int size = namevalue_size_xe(loc->xl_entry);
1847        int block_offset = nameval_offset >> sb->s_blocksize_bits;
1848
1849        /* Values are not allowed to straddle block boundaries */
1850        BUG_ON(block_offset !=
1851               ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1852        /* We expect the bucket to be filled in */
1853        BUG_ON(!bucket->bu_bhs[block_offset]);
1854
1855        vb->vb_access = ocfs2_journal_access;
1856        vb->vb_bh = bucket->bu_bhs[block_offset];
1857}
1858
1859/* Operations for xattrs stored in buckets. */
1860static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1861        .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1862        .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1863        .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1864        .xlo_check_space        = ocfs2_xa_bucket_check_space,
1865        .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1866        .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1867        .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1868        .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1869        .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1870        .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1871};
1872
1873static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1874{
1875        struct ocfs2_xattr_value_buf vb;
1876
1877        if (ocfs2_xattr_is_local(loc->xl_entry))
1878                return 0;
1879
1880        ocfs2_xa_fill_value_buf(loc, &vb);
1881        return le32_to_cpu(vb.vb_xv->xr_clusters);
1882}
1883
1884static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1885                                   struct ocfs2_xattr_set_ctxt *ctxt)
1886{
1887        int trunc_rc, access_rc;
1888        struct ocfs2_xattr_value_buf vb;
1889
1890        ocfs2_xa_fill_value_buf(loc, &vb);
1891        trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1892                                              ctxt);
1893
1894        /*
1895         * The caller of ocfs2_xa_value_truncate() has already called
1896         * ocfs2_xa_journal_access on the loc.  However, The truncate code
1897         * calls ocfs2_extend_trans().  This may commit the previous
1898         * transaction and open a new one.  If this is a bucket, truncate
1899         * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1900         * the caller is expecting to dirty the entire bucket.  So we must
1901         * reset the journal work.  We do this even if truncate has failed,
1902         * as it could have failed after committing the extend.
1903         */
1904        access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1905                                            OCFS2_JOURNAL_ACCESS_WRITE);
1906
1907        /* Errors in truncate take precedence */
1908        return trunc_rc ? trunc_rc : access_rc;
1909}
1910
1911static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1912{
1913        int index, count;
1914        struct ocfs2_xattr_header *xh = loc->xl_header;
1915        struct ocfs2_xattr_entry *entry = loc->xl_entry;
1916
1917        ocfs2_xa_wipe_namevalue(loc);
1918        loc->xl_entry = NULL;
1919
1920        le16_add_cpu(&xh->xh_count, -1);
1921        count = le16_to_cpu(xh->xh_count);
1922
1923        /*
1924         * Only zero out the entry if there are more remaining.  This is
1925         * important for an empty bucket, as it keeps track of the
1926         * bucket's hash value.  It doesn't hurt empty block storage.
1927         */
1928        if (count) {
1929                index = ((char *)entry - (char *)&xh->xh_entries) /
1930                        sizeof(struct ocfs2_xattr_entry);
1931                memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1932                        (count - index) * sizeof(struct ocfs2_xattr_entry));
1933                memset(&xh->xh_entries[count], 0,
1934                       sizeof(struct ocfs2_xattr_entry));
1935        }
1936}
1937
1938/*
1939 * If we have a problem adjusting the size of an external value during
1940 * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1941 * in an intermediate state.  For example, the value may be partially
1942 * truncated.
1943 *
1944 * If the value tree hasn't changed, the extend/truncate went nowhere.
1945 * We have nothing to do.  The caller can treat it as a straight error.
1946 *
1947 * If the value tree got partially truncated, we now have a corrupted
1948 * extended attribute.  We're going to wipe its entry and leak the
1949 * clusters.  Better to leak some storage than leave a corrupt entry.
1950 *
1951 * If the value tree grew, it obviously didn't grow enough for the
1952 * new entry.  We're not going to try and reclaim those clusters either.
1953 * If there was already an external value there (orig_clusters != 0),
1954 * the new clusters are attached safely and we can just leave the old
1955 * value in place.  If there was no external value there, we remove
1956 * the entry.
1957 *
1958 * This way, the xattr block we store in the journal will be consistent.
1959 * If the size change broke because of the journal, no changes will hit
1960 * disk anyway.
1961 */
1962static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1963                                            const char *what,
1964                                            unsigned int orig_clusters)
1965{
1966        unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1967        char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1968                                le16_to_cpu(loc->xl_entry->xe_name_offset));
1969
1970        if (new_clusters < orig_clusters) {
1971                mlog(ML_ERROR,
1972                     "Partial truncate while %s xattr %.*s.  Leaking "
1973                     "%u clusters and removing the entry\n",
1974                     what, loc->xl_entry->xe_name_len, nameval_buf,
1975                     orig_clusters - new_clusters);
1976                ocfs2_xa_remove_entry(loc);
1977        } else if (!orig_clusters) {
1978                mlog(ML_ERROR,
1979                     "Unable to allocate an external value for xattr "
1980                     "%.*s safely.  Leaking %u clusters and removing the "
1981                     "entry\n",
1982                     loc->xl_entry->xe_name_len, nameval_buf,
1983                     new_clusters - orig_clusters);
1984                ocfs2_xa_remove_entry(loc);
1985        } else if (new_clusters > orig_clusters)
1986                mlog(ML_ERROR,
1987                     "Unable to grow xattr %.*s safely.  %u new clusters "
1988                     "have been added, but the value will not be "
1989                     "modified\n",
1990                     loc->xl_entry->xe_name_len, nameval_buf,
1991                     new_clusters - orig_clusters);
1992}
1993
1994static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
1995                           struct ocfs2_xattr_set_ctxt *ctxt)
1996{
1997        int rc = 0;
1998        unsigned int orig_clusters;
1999
2000        if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2001                orig_clusters = ocfs2_xa_value_clusters(loc);
2002                rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2003                if (rc) {
2004                        mlog_errno(rc);
2005                        /*
2006                         * Since this is remove, we can return 0 if
2007                         * ocfs2_xa_cleanup_value_truncate() is going to
2008                         * wipe the entry anyway.  So we check the
2009                         * cluster count as well.
2010                         */
2011                        if (orig_clusters != ocfs2_xa_value_clusters(loc))
2012                                rc = 0;
2013                        ocfs2_xa_cleanup_value_truncate(loc, "removing",
2014                                                        orig_clusters);
2015                        if (rc)
2016                                goto out;
2017                }
2018        }
2019
2020        ocfs2_xa_remove_entry(loc);
2021
2022out:
2023        return rc;
2024}
2025
2026static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2027{
2028        int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2029        char *nameval_buf;
2030
2031        nameval_buf = ocfs2_xa_offset_pointer(loc,
2032                                le16_to_cpu(loc->xl_entry->xe_name_offset));
2033        memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2034}
2035
2036/*
2037 * Take an existing entry and make it ready for the new value.  This
2038 * won't allocate space, but it may free space.  It should be ready for
2039 * ocfs2_xa_prepare_entry() to finish the work.
2040 */
2041static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2042                                struct ocfs2_xattr_info *xi,
2043                                struct ocfs2_xattr_set_ctxt *ctxt)
2044{
2045        int rc = 0;
2046        int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2047        unsigned int orig_clusters;
2048        char *nameval_buf;
2049        int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2050        int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2051
2052        BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2053               name_size);
2054
2055        nameval_buf = ocfs2_xa_offset_pointer(loc,
2056                                le16_to_cpu(loc->xl_entry->xe_name_offset));
2057        if (xe_local) {
2058                memset(nameval_buf + name_size, 0,
2059                       namevalue_size_xe(loc->xl_entry) - name_size);
2060                if (!xi_local)
2061                        ocfs2_xa_install_value_root(loc);
2062        } else {
2063                orig_clusters = ocfs2_xa_value_clusters(loc);
2064                if (xi_local) {
2065                        rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2066                        if (rc < 0)
2067                                mlog_errno(rc);
2068                        else
2069                                memset(nameval_buf + name_size, 0,
2070                                       namevalue_size_xe(loc->xl_entry) -
2071                                       name_size);
2072                } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2073                           xi->xi_value_len) {
2074                        rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2075                                                     ctxt);
2076                        if (rc < 0)
2077                                mlog_errno(rc);
2078                }
2079
2080                if (rc) {
2081                        ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2082                                                        orig_clusters);
2083                        goto out;
2084                }
2085        }
2086
2087        loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2088        ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2089
2090out:
2091        return rc;
2092}
2093
2094/*
2095 * Prepares loc->xl_entry to receive the new xattr.  This includes
2096 * properly setting up the name+value pair region.  If loc->xl_entry
2097 * already exists, it will take care of modifying it appropriately.
2098 *
2099 * Note that this modifies the data.  You did journal_access already,
2100 * right?
2101 */
2102static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2103                                  struct ocfs2_xattr_info *xi,
2104                                  u32 name_hash,
2105                                  struct ocfs2_xattr_set_ctxt *ctxt)
2106{
2107        int rc = 0;
2108        unsigned int orig_clusters;
2109        __le64 orig_value_size = 0;
2110
2111        rc = ocfs2_xa_check_space(loc, xi);
2112        if (rc)
2113                goto out;
2114
2115        if (loc->xl_entry) {
2116                if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2117                        orig_value_size = loc->xl_entry->xe_value_size;
2118                        rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2119                        if (rc)
2120                                goto out;
2121                        goto alloc_value;
2122                }
2123
2124                if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2125                        orig_clusters = ocfs2_xa_value_clusters(loc);
2126                        rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2127                        if (rc) {
2128                                mlog_errno(rc);
2129                                ocfs2_xa_cleanup_value_truncate(loc,
2130                                                                "overwriting",
2131                                                                orig_clusters);
2132                                goto out;
2133                        }
2134                }
2135                ocfs2_xa_wipe_namevalue(loc);
2136        } else
2137                ocfs2_xa_add_entry(loc, name_hash);
2138
2139        /*
2140         * If we get here, we have a blank entry.  Fill it.  We grow our
2141         * name+value pair back from the end.
2142         */
2143        ocfs2_xa_add_namevalue(loc, xi);
2144        if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2145                ocfs2_xa_install_value_root(loc);
2146
2147alloc_value:
2148        if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2149                orig_clusters = ocfs2_xa_value_clusters(loc);
2150                rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2151                if (rc < 0) {
2152                        ctxt->set_abort = 1;
2153                        ocfs2_xa_cleanup_value_truncate(loc, "growing",
2154                                                        orig_clusters);
2155                        /*
2156                         * If we were growing an existing value,
2157                         * ocfs2_xa_cleanup_value_truncate() won't remove
2158                         * the entry. We need to restore the original value
2159                         * size.
2160                         */
2161                        if (loc->xl_entry) {
2162                                BUG_ON(!orig_value_size);
2163                                loc->xl_entry->xe_value_size = orig_value_size;
2164                        }
2165                        mlog_errno(rc);
2166                }
2167        }
2168
2169out:
2170        return rc;
2171}
2172
2173/*
2174 * Store the value portion of the name+value pair.  This will skip
2175 * values that are stored externally.  Their tree roots were set up
2176 * by ocfs2_xa_prepare_entry().
2177 */
2178static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2179                                struct ocfs2_xattr_info *xi,
2180                                struct ocfs2_xattr_set_ctxt *ctxt)
2181{
2182        int rc = 0;
2183        int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2184        int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2185        char *nameval_buf;
2186        struct ocfs2_xattr_value_buf vb;
2187
2188        nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2189        if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2190                ocfs2_xa_fill_value_buf(loc, &vb);
2191                rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2192                                                     ctxt->handle, &vb,
2193                                                     xi->xi_value,
2194                                                     xi->xi_value_len);
2195        } else
2196                memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2197
2198        return rc;
2199}
2200
2201static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2202                        struct ocfs2_xattr_info *xi,
2203                        struct ocfs2_xattr_set_ctxt *ctxt)
2204{
2205        int ret;
2206        u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2207                                              xi->xi_name_len);
2208
2209        ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2210                                      OCFS2_JOURNAL_ACCESS_WRITE);
2211        if (ret) {
2212                mlog_errno(ret);
2213                goto out;
2214        }
2215
2216        /*
2217         * From here on out, everything is going to modify the buffer a
2218         * little.  Errors are going to leave the xattr header in a
2219         * sane state.  Thus, even with errors we dirty the sucker.
2220         */
2221
2222        /* Don't worry, we are never called with !xi_value and !xl_entry */
2223        if (!xi->xi_value) {
2224                ret = ocfs2_xa_remove(loc, ctxt);
2225                goto out_dirty;
2226        }
2227
2228        ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2229        if (ret) {
2230                if (ret != -ENOSPC)
2231                        mlog_errno(ret);
2232                goto out_dirty;
2233        }
2234
2235        ret = ocfs2_xa_store_value(loc, xi, ctxt);
2236        if (ret)
2237                mlog_errno(ret);
2238
2239out_dirty:
2240        ocfs2_xa_journal_dirty(ctxt->handle, loc);
2241
2242out:
2243        return ret;
2244}
2245
2246static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2247                                     struct inode *inode,
2248                                     struct buffer_head *bh,
2249                                     struct ocfs2_xattr_entry *entry)
2250{
2251        struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2252
2253        BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2254
2255        loc->xl_inode = inode;
2256        loc->xl_ops = &ocfs2_xa_block_loc_ops;
2257        loc->xl_storage = bh;
2258        loc->xl_entry = entry;
2259        loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2260        loc->xl_header =
2261                (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2262                                              loc->xl_size);
2263}
2264
2265static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2266                                          struct inode *inode,
2267                                          struct buffer_head *bh,
2268                                          struct ocfs2_xattr_entry *entry)
2269{
2270        struct ocfs2_xattr_block *xb =
2271                (struct ocfs2_xattr_block *)bh->b_data;
2272
2273        BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2274
2275        loc->xl_inode = inode;
2276        loc->xl_ops = &ocfs2_xa_block_loc_ops;
2277        loc->xl_storage = bh;
2278        loc->xl_header = &(xb->xb_attrs.xb_header);
2279        loc->xl_entry = entry;
2280        loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2281                                             xb_attrs.xb_header);
2282}
2283
2284static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2285                                           struct ocfs2_xattr_bucket *bucket,
2286                                           struct ocfs2_xattr_entry *entry)
2287{
2288        loc->xl_inode = bucket->bu_inode;
2289        loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2290        loc->xl_storage = bucket;
2291        loc->xl_header = bucket_xh(bucket);
2292        loc->xl_entry = entry;
2293        loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2294}
2295
2296/*
2297 * In xattr remove, if it is stored outside and refcounted, we may have
2298 * the chance to split the refcount tree. So need the allocators.
2299 */
2300static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2301                                        struct ocfs2_xattr_value_root *xv,
2302                                        struct ocfs2_caching_info *ref_ci,
2303                                        struct buffer_head *ref_root_bh,
2304                                        struct ocfs2_alloc_context **meta_ac,
2305                                        int *ref_credits)
2306{
2307        int ret, meta_add = 0;
2308        u32 p_cluster, num_clusters;
2309        unsigned int ext_flags;
2310
2311        *ref_credits = 0;
2312        ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2313                                       &num_clusters,
2314                                       &xv->xr_list,
2315                                       &ext_flags);
2316        if (ret) {
2317                mlog_errno(ret);
2318                goto out;
2319        }
2320
2321        if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2322                goto out;
2323
2324        ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2325                                                 ref_root_bh, xv,
2326                                                 &meta_add, ref_credits);
2327        if (ret) {
2328                mlog_errno(ret);
2329                goto out;
2330        }
2331
2332        ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2333                                                meta_add, meta_ac);
2334        if (ret)
2335                mlog_errno(ret);
2336
2337out:
2338        return ret;
2339}
2340
2341static int ocfs2_remove_value_outside(struct inode*inode,
2342                                      struct ocfs2_xattr_value_buf *vb,
2343                                      struct ocfs2_xattr_header *header,
2344                                      struct ocfs2_caching_info *ref_ci,
2345                                      struct buffer_head *ref_root_bh)
2346{
2347        int ret = 0, i, ref_credits;
2348        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2349        struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2350        void *val;
2351
2352        ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2353
2354        for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2355                struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2356
2357                if (ocfs2_xattr_is_local(entry))
2358                        continue;
2359
2360                val = (void *)header +
2361                        le16_to_cpu(entry->xe_name_offset);
2362                vb->vb_xv = (struct ocfs2_xattr_value_root *)
2363                        (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2364
2365                ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2366                                                         ref_ci, ref_root_bh,
2367                                                         &ctxt.meta_ac,
2368                                                         &ref_credits);
2369
2370                ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2371                                        ocfs2_remove_extent_credits(osb->sb));
2372                if (IS_ERR(ctxt.handle)) {
2373                        ret = PTR_ERR(ctxt.handle);
2374                        mlog_errno(ret);
2375                        break;
2376                }
2377
2378                ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2379
2380                ocfs2_commit_trans(osb, ctxt.handle);
2381                if (ctxt.meta_ac) {
2382                        ocfs2_free_alloc_context(ctxt.meta_ac);
2383                        ctxt.meta_ac = NULL;
2384                }
2385
2386                if (ret < 0) {
2387                        mlog_errno(ret);
2388                        break;
2389                }
2390
2391        }
2392
2393        if (ctxt.meta_ac)
2394                ocfs2_free_alloc_context(ctxt.meta_ac);
2395        ocfs2_schedule_truncate_log_flush(osb, 1);
2396        ocfs2_run_deallocs(osb, &ctxt.dealloc);
2397        return ret;
2398}
2399
2400static int ocfs2_xattr_ibody_remove(struct inode *inode,
2401                                    struct buffer_head *di_bh,
2402                                    struct ocfs2_caching_info *ref_ci,
2403                                    struct buffer_head *ref_root_bh)
2404{
2405
2406        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2407        struct ocfs2_xattr_header *header;
2408        int ret;
2409        struct ocfs2_xattr_value_buf vb = {
2410                .vb_bh = di_bh,
2411                .vb_access = ocfs2_journal_access_di,
2412        };
2413
2414        header = (struct ocfs2_xattr_header *)
2415                 ((void *)di + inode->i_sb->s_blocksize -
2416                 le16_to_cpu(di->i_xattr_inline_size));
2417
2418        ret = ocfs2_remove_value_outside(inode, &vb, header,
2419                                         ref_ci, ref_root_bh);
2420
2421        return ret;
2422}
2423
2424struct ocfs2_rm_xattr_bucket_para {
2425        struct ocfs2_caching_info *ref_ci;
2426        struct buffer_head *ref_root_bh;
2427};
2428
2429static int ocfs2_xattr_block_remove(struct inode *inode,
2430                                    struct buffer_head *blk_bh,
2431                                    struct ocfs2_caching_info *ref_ci,
2432                                    struct buffer_head *ref_root_bh)
2433{
2434        struct ocfs2_xattr_block *xb;
2435        int ret = 0;
2436        struct ocfs2_xattr_value_buf vb = {
2437                .vb_bh = blk_bh,
2438                .vb_access = ocfs2_journal_access_xb,
2439        };
2440        struct ocfs2_rm_xattr_bucket_para args = {
2441                .ref_ci = ref_ci,
2442                .ref_root_bh = ref_root_bh,
2443        };
2444
2445        xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2446        if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2447                struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2448                ret = ocfs2_remove_value_outside(inode, &vb, header,
2449                                                 ref_ci, ref_root_bh);
2450        } else
2451                ret = ocfs2_iterate_xattr_index_block(inode,
2452                                                blk_bh,
2453                                                ocfs2_rm_xattr_cluster,
2454                                                &args);
2455
2456        return ret;
2457}
2458
2459static int ocfs2_xattr_free_block(struct inode *inode,
2460                                  u64 block,
2461                                  struct ocfs2_caching_info *ref_ci,
2462                                  struct buffer_head *ref_root_bh)
2463{
2464        struct inode *xb_alloc_inode;
2465        struct buffer_head *xb_alloc_bh = NULL;
2466        struct buffer_head *blk_bh = NULL;
2467        struct ocfs2_xattr_block *xb;
2468        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2469        handle_t *handle;
2470        int ret = 0;
2471        u64 blk, bg_blkno;
2472        u16 bit;
2473
2474        ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2475        if (ret < 0) {
2476                mlog_errno(ret);
2477                goto out;
2478        }
2479
2480        ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2481        if (ret < 0) {
2482                mlog_errno(ret);
2483                goto out;
2484        }
2485
2486        xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2487        blk = le64_to_cpu(xb->xb_blkno);
2488        bit = le16_to_cpu(xb->xb_suballoc_bit);
2489        if (xb->xb_suballoc_loc)
2490                bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2491        else
2492                bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2493
2494        xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2495                                EXTENT_ALLOC_SYSTEM_INODE,
2496                                le16_to_cpu(xb->xb_suballoc_slot));
2497        if (!xb_alloc_inode) {
2498                ret = -ENOMEM;
2499                mlog_errno(ret);
2500                goto out;
2501        }
2502        mutex_lock(&xb_alloc_inode->i_mutex);
2503
2504        ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2505        if (ret < 0) {
2506                mlog_errno(ret);
2507                goto out_mutex;
2508        }
2509
2510        handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2511        if (IS_ERR(handle)) {
2512                ret = PTR_ERR(handle);
2513                mlog_errno(ret);
2514                goto out_unlock;
2515        }
2516
2517        ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2518                                       bit, bg_blkno, 1);
2519        if (ret < 0)
2520                mlog_errno(ret);
2521
2522        ocfs2_commit_trans(osb, handle);
2523out_unlock:
2524        ocfs2_inode_unlock(xb_alloc_inode, 1);
2525        brelse(xb_alloc_bh);
2526out_mutex:
2527        mutex_unlock(&xb_alloc_inode->i_mutex);
2528        iput(xb_alloc_inode);
2529out:
2530        brelse(blk_bh);
2531        return ret;
2532}
2533
2534/*
2535 * ocfs2_xattr_remove()
2536 *
2537 * Free extended attribute resources associated with this inode.
2538 */
2539int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2540{
2541        struct ocfs2_inode_info *oi = OCFS2_I(inode);
2542        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2543        struct ocfs2_refcount_tree *ref_tree = NULL;
2544        struct buffer_head *ref_root_bh = NULL;
2545        struct ocfs2_caching_info *ref_ci = NULL;
2546        handle_t *handle;
2547        int ret;
2548
2549        if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2550                return 0;
2551
2552        if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2553                return 0;
2554
2555        if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2556                ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2557                                               le64_to_cpu(di->i_refcount_loc),
2558                                               1, &ref_tree, &ref_root_bh);
2559                if (ret) {
2560                        mlog_errno(ret);
2561                        goto out;
2562                }
2563                ref_ci = &ref_tree->rf_ci;
2564
2565        }
2566
2567        if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2568                ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2569                                               ref_ci, ref_root_bh);
2570                if (ret < 0) {
2571                        mlog_errno(ret);
2572                        goto out;
2573                }
2574        }
2575
2576        if (di->i_xattr_loc) {
2577                ret = ocfs2_xattr_free_block(inode,
2578                                             le64_to_cpu(di->i_xattr_loc),
2579                                             ref_ci, ref_root_bh);
2580                if (ret < 0) {
2581                        mlog_errno(ret);
2582                        goto out;
2583                }
2584        }
2585
2586        handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2587                                   OCFS2_INODE_UPDATE_CREDITS);
2588        if (IS_ERR(handle)) {
2589                ret = PTR_ERR(handle);
2590                mlog_errno(ret);
2591                goto out;
2592        }
2593        ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2594                                      OCFS2_JOURNAL_ACCESS_WRITE);
2595        if (ret) {
2596                mlog_errno(ret);
2597                goto out_commit;
2598        }
2599
2600        di->i_xattr_loc = 0;
2601
2602        spin_lock(&oi->ip_lock);
2603        oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2604        di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2605        spin_unlock(&oi->ip_lock);
2606
2607        ocfs2_journal_dirty(handle, di_bh);
2608out_commit:
2609        ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2610out:
2611        if (ref_tree)
2612                ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2613        brelse(ref_root_bh);
2614        return ret;
2615}
2616
2617static int ocfs2_xattr_has_space_inline(struct inode *inode,
2618                                        struct ocfs2_dinode *di)
2619{
2620        struct ocfs2_inode_info *oi = OCFS2_I(inode);
2621        unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2622        int free;
2623
2624        if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2625                return 0;
2626
2627        if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2628                struct ocfs2_inline_data *idata = &di->id2.i_data;
2629                free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2630        } else if (ocfs2_inode_is_fast_symlink(inode)) {
2631                free = ocfs2_fast_symlink_chars(inode->i_sb) -
2632                        le64_to_cpu(di->i_size);
2633        } else {
2634                struct ocfs2_extent_list *el = &di->id2.i_list;
2635                free = (le16_to_cpu(el->l_count) -
2636                        le16_to_cpu(el->l_next_free_rec)) *
2637                        sizeof(struct ocfs2_extent_rec);
2638        }
2639        if (free >= xattrsize)
2640                return 1;
2641
2642        return 0;
2643}
2644
2645/*
2646 * ocfs2_xattr_ibody_find()
2647 *
2648 * Find extended attribute in inode block and
2649 * fill search info into struct ocfs2_xattr_search.
2650 */
2651static int ocfs2_xattr_ibody_find(struct inode *inode,
2652                                  int name_index,
2653                                  const char *name,
2654                                  struct ocfs2_xattr_search *xs)
2655{
2656        struct ocfs2_inode_info *oi = OCFS2_I(inode);
2657        struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2658        int ret;
2659        int has_space = 0;
2660
2661        if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2662                return 0;
2663
2664        if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2665                down_read(&oi->ip_alloc_sem);
2666                has_space = ocfs2_xattr_has_space_inline(inode, di);
2667                up_read(&oi->ip_alloc_sem);
2668                if (!has_space)
2669                        return 0;
2670        }
2671
2672        xs->xattr_bh = xs->inode_bh;
2673        xs->end = (void *)di + inode->i_sb->s_blocksize;
2674        if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2675                xs->header = (struct ocfs2_xattr_header *)
2676                        (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2677        else
2678                xs->header = (struct ocfs2_xattr_header *)
2679                        (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2680        xs->base = (void *)xs->header;
2681        xs->here = xs->header->xh_entries;
2682
2683        /* Find the named attribute. */
2684        if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2685                ret = ocfs2_xattr_find_entry(name_index, name, xs);
2686                if (ret && ret != -ENODATA)
2687                        return ret;
2688                xs->not_found = ret;
2689        }
2690
2691        return 0;
2692}
2693
2694static int ocfs2_xattr_ibody_init(struct inode *inode,
2695                                  struct buffer_head *di_bh,
2696                                  struct ocfs2_xattr_set_ctxt *ctxt)
2697{
2698        int ret;
2699        struct ocfs2_inode_info *oi = OCFS2_I(inode);
2700        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2701        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2702        unsigned int xattrsize = osb->s_xattr_inline_size;
2703
2704        if (!ocfs2_xattr_has_space_inline(inode, di)) {
2705                ret = -ENOSPC;
2706                goto out;
2707        }
2708
2709        ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2710                                      OCFS2_JOURNAL_ACCESS_WRITE);
2711        if (ret) {
2712                mlog_errno(ret);
2713                goto out;
2714        }
2715
2716        /*
2717         * Adjust extent record count or inline data size
2718         * to reserve space for extended attribute.
2719         */
2720        if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2721                struct ocfs2_inline_data *idata = &di->id2.i_data;
2722                le16_add_cpu(&idata->id_count, -xattrsize);
2723        } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2724                struct ocfs2_extent_list *el = &di->id2.i_list;
2725                le16_add_cpu(&el->l_count, -(xattrsize /
2726                                             sizeof(struct ocfs2_extent_rec)));
2727        }
2728        di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2729
2730        spin_lock(&oi->ip_lock);
2731        oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2732        di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2733        spin_unlock(&oi->ip_lock);
2734
2735        ocfs2_journal_dirty(ctxt->handle, di_bh);
2736
2737out:
2738        return ret;
2739}
2740
2741/*
2742 * ocfs2_xattr_ibody_set()
2743 *
2744 * Set, replace or remove an extended attribute into inode block.
2745 *
2746 */
2747static int ocfs2_xattr_ibody_set(struct inode *inode,
2748                                 struct ocfs2_xattr_info *xi,
2749                                 struct ocfs2_xattr_search *xs,
2750                                 struct ocfs2_xattr_set_ctxt *ctxt)
2751{
2752        int ret;
2753        struct ocfs2_inode_info *oi = OCFS2_I(inode);
2754        struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2755        struct ocfs2_xa_loc loc;
2756
2757        if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2758                return -ENOSPC;
2759
2760        down_write(&oi->ip_alloc_sem);
2761        if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2762                if (!ocfs2_xattr_has_space_inline(inode, di)) {
2763                        ret = -ENOSPC;
2764                        goto out;
2765                }
2766        }
2767
2768        if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2769                ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2770                if (ret) {
2771                        if (ret != -ENOSPC)
2772                                mlog_errno(ret);
2773                        goto out;
2774                }
2775        }
2776
2777        ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2778                                 xs->not_found ? NULL : xs->here);
2779        ret = ocfs2_xa_set(&loc, xi, ctxt);
2780        if (ret) {
2781                if (ret != -ENOSPC)
2782                        mlog_errno(ret);
2783                goto out;
2784        }
2785        xs->here = loc.xl_entry;
2786
2787out:
2788        up_write(&oi->ip_alloc_sem);
2789
2790        return ret;
2791}
2792
2793/*
2794 * ocfs2_xattr_block_find()
2795 *
2796 * Find extended attribute in external block and
2797 * fill search info into struct ocfs2_xattr_search.
2798 */
2799static int ocfs2_xattr_block_find(struct inode *inode,
2800                                  int name_index,
2801                                  const char *name,
2802                                  struct ocfs2_xattr_search *xs)
2803{
2804        struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2805        struct buffer_head *blk_bh = NULL;
2806        struct ocfs2_xattr_block *xb;
2807        int ret = 0;
2808
2809        if (!di->i_xattr_loc)
2810                return ret;
2811
2812        ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2813                                     &blk_bh);
2814        if (ret < 0) {
2815                mlog_errno(ret);
2816                return ret;
2817        }
2818
2819        xs->xattr_bh = blk_bh;
2820        xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2821
2822        if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2823                xs->header = &xb->xb_attrs.xb_header;
2824                xs->base = (void *)xs->header;
2825                xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2826                xs->here = xs->header->xh_entries;
2827
2828                ret = ocfs2_xattr_find_entry(name_index, name, xs);
2829        } else
2830                ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2831                                                   name_index,
2832                                                   name, xs);
2833
2834        if (ret && ret != -ENODATA) {
2835                xs->xattr_bh = NULL;
2836                goto cleanup;
2837        }
2838        xs->not_found = ret;
2839        return 0;
2840cleanup:
2841        brelse(blk_bh);
2842
2843        return ret;
2844}
2845
2846static int ocfs2_create_xattr_block(struct inode *inode,
2847                                    struct buffer_head *inode_bh,
2848                                    struct ocfs2_xattr_set_ctxt *ctxt,
2849                                    int indexed,
2850                                    struct buffer_head **ret_bh)
2851{
2852        int ret;
2853        u16 suballoc_bit_start;
2854        u32 num_got;
2855        u64 suballoc_loc, first_blkno;
2856        struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2857        struct buffer_head *new_bh = NULL;
2858        struct ocfs2_xattr_block *xblk;
2859
2860        ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2861                                      inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2862        if (ret < 0) {
2863                mlog_errno(ret);
2864                goto end;
2865        }
2866
2867        ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2868                                   &suballoc_loc, &suballoc_bit_start,
2869                                   &num_got, &first_blkno);
2870        if (ret < 0) {
2871                mlog_errno(ret);
2872                goto end;
2873        }
2874
2875        new_bh = sb_getblk(inode->i_sb, first_blkno);
2876        ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2877
2878        ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2879                                      new_bh,
2880                                      OCFS2_JOURNAL_ACCESS_CREATE);
2881        if (ret < 0) {
2882                mlog_errno(ret);
2883                goto end;
2884        }
2885
2886        /* Initialize ocfs2_xattr_block */
2887        xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2888        memset(xblk, 0, inode->i_sb->s_blocksize);
2889        strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2890        xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2891        xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2892        xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2893        xblk->xb_fs_generation =
2894                cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2895        xblk->xb_blkno = cpu_to_le64(first_blkno);
2896        if (indexed) {
2897                struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2898                xr->xt_clusters = cpu_to_le32(1);
2899                xr->xt_last_eb_blk = 0;
2900                xr->xt_list.l_tree_depth = 0;
2901                xr->xt_list.l_count = cpu_to_le16(
2902                                        ocfs2_xattr_recs_per_xb(inode->i_sb));
2903                xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2904                xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2905        }
2906        ocfs2_journal_dirty(ctxt->handle, new_bh);
2907
2908        /* Add it to the inode */
2909        di->i_xattr_loc = cpu_to_le64(first_blkno);
2910
2911        spin_lock(&OCFS2_I(inode)->ip_lock);
2912        OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2913        di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2914        spin_unlock(&OCFS2_I(inode)->ip_lock);
2915
2916        ocfs2_journal_dirty(ctxt->handle, inode_bh);
2917
2918        *ret_bh = new_bh;
2919        new_bh = NULL;
2920
2921end:
2922        brelse(new_bh);
2923        return ret;
2924}
2925
2926/*
2927 * ocfs2_xattr_block_set()
2928 *
2929 * Set, replace or remove an extended attribute into external block.
2930 *
2931 */
2932static int ocfs2_xattr_block_set(struct inode *inode,
2933                                 struct ocfs2_xattr_info *xi,
2934                                 struct ocfs2_xattr_search *xs,
2935                                 struct ocfs2_xattr_set_ctxt *ctxt)
2936{
2937        struct buffer_head *new_bh = NULL;
2938        struct ocfs2_xattr_block *xblk = NULL;
2939        int ret;
2940        struct ocfs2_xa_loc loc;
2941
2942        if (!xs->xattr_bh) {
2943                ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2944                                               0, &new_bh);
2945                if (ret) {
2946                        mlog_errno(ret);
2947                        goto end;
2948                }
2949
2950                xs->xattr_bh = new_bh;
2951                xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2952                xs->header = &xblk->xb_attrs.xb_header;
2953                xs->base = (void *)xs->header;
2954                xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2955                xs->here = xs->header->xh_entries;
2956        } else
2957                xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2958
2959        if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2960                ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2961                                              xs->not_found ? NULL : xs->here);
2962
2963                ret = ocfs2_xa_set(&loc, xi, ctxt);
2964                if (!ret)
2965                        xs->here = loc.xl_entry;
2966                else if ((ret != -ENOSPC) || ctxt->set_abort)
2967                        goto end;
2968                else {
2969                        ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2970                        if (ret)
2971                                goto end;
2972                }
2973        }
2974
2975        if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2976                ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2977
2978end:
2979        return ret;
2980}
2981
2982/* Check whether the new xattr can be inserted into the inode. */
2983static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2984                                       struct ocfs2_xattr_info *xi,
2985                                       struct ocfs2_xattr_search *xs)
2986{
2987        struct ocfs2_xattr_entry *last;
2988        int free, i;
2989        size_t min_offs = xs->end - xs->base;
2990
2991        if (!xs->header)
2992                return 0;
2993
2994        last = xs->header->xh_entries;
2995
2996        for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2997                size_t offs = le16_to_cpu(last->xe_name_offset);
2998                if (offs < min_offs)
2999                        min_offs = offs;
3000                last += 1;
3001        }
3002
3003        free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3004        if (free < 0)
3005                return 0;
3006
3007        BUG_ON(!xs->not_found);
3008
3009        if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3010                return 1;
3011
3012        return 0;
3013}
3014
3015static int ocfs2_calc_xattr_set_need(struct inode *inode,
3016                                     struct ocfs2_dinode *di,
3017                                     struct ocfs2_xattr_info *xi,
3018                                     struct ocfs2_xattr_search *xis,
3019                                     struct ocfs2_xattr_search *xbs,
3020                                     int *clusters_need,
3021                                     int *meta_need,
3022                                     int *credits_need)
3023{
3024        int ret = 0, old_in_xb = 0;
3025        int clusters_add = 0, meta_add = 0, credits = 0;
3026        struct buffer_head *bh = NULL;
3027        struct ocfs2_xattr_block *xb = NULL;
3028        struct ocfs2_xattr_entry *xe = NULL;
3029        struct ocfs2_xattr_value_root *xv = NULL;
3030        char *base = NULL;
3031        int name_offset, name_len = 0;
3032        u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3033                                                    xi->xi_value_len);
3034        u64 value_size;
3035
3036        /*
3037         * Calculate the clusters we need to write.
3038         * No matter whether we replace an old one or add a new one,
3039         * we need this for writing.
3040         */
3041        if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3042                credits += new_clusters *
3043                           ocfs2_clusters_to_blocks(inode->i_sb, 1);
3044
3045        if (xis->not_found && xbs->not_found) {
3046                credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3047
3048                if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3049                        clusters_add += new_clusters;
3050                        credits += ocfs2_calc_extend_credits(inode->i_sb,
3051                                                        &def_xv.xv.xr_list,
3052                                                        new_clusters);
3053                }
3054
3055                goto meta_guess;
3056        }
3057
3058        if (!xis->not_found) {
3059                xe = xis->here;
3060                name_offset = le16_to_cpu(xe->xe_name_offset);
3061                name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3062                base = xis->base;
3063                credits += OCFS2_INODE_UPDATE_CREDITS;
3064        } else {
3065                int i, block_off = 0;
3066                xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3067                xe = xbs->here;
3068                name_offset = le16_to_cpu(xe->xe_name_offset);
3069                name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3070                i = xbs->here - xbs->header->xh_entries;
3071                old_in_xb = 1;
3072
3073                if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3074                        ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3075                                                        bucket_xh(xbs->bucket),
3076                                                        i, &block_off,
3077                                                        &name_offset);
3078                        base = bucket_block(xbs->bucket, block_off);
3079                        credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3080                } else {
3081                        base = xbs->base;
3082                        credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3083                }
3084        }
3085
3086        /*
3087         * delete a xattr doesn't need metadata and cluster allocation.
3088         * so just calculate the credits and return.
3089         *
3090         * The credits for removing the value tree will be extended
3091         * by ocfs2_remove_extent itself.
3092         */
3093        if (!xi->xi_value) {
3094                if (!ocfs2_xattr_is_local(xe))
3095                        credits += ocfs2_remove_extent_credits(inode->i_sb);
3096
3097                goto out;
3098        }
3099
3100        /* do cluster allocation guess first. */
3101        value_size = le64_to_cpu(xe->xe_value_size);
3102
3103        if (old_in_xb) {
3104                /*
3105                 * In xattr set, we always try to set the xe in inode first,
3106                 * so if it can be inserted into inode successfully, the old
3107                 * one will be removed from the xattr block, and this xattr
3108                 * will be inserted into inode as a new xattr in inode.
3109                 */
3110                if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3111                        clusters_add += new_clusters;
3112                        credits += ocfs2_remove_extent_credits(inode->i_sb) +
3113                                    OCFS2_INODE_UPDATE_CREDITS;
3114                        if (!ocfs2_xattr_is_local(xe))
3115                                credits += ocfs2_calc_extend_credits(
3116                                                        inode->i_sb,
3117                                                        &def_xv.xv.xr_list,
3118                                                        new_clusters);
3119                        goto out;
3120                }
3121        }
3122
3123        if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3124                /* the new values will be stored outside. */
3125                u32 old_clusters = 0;
3126
3127                if (!ocfs2_xattr_is_local(xe)) {
3128                        old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3129                                                                 value_size);
3130                        xv = (struct ocfs2_xattr_value_root *)
3131                             (base + name_offset + name_len);
3132                        value_size = OCFS2_XATTR_ROOT_SIZE;
3133                } else
3134                        xv = &def_xv.xv;
3135
3136                if (old_clusters >= new_clusters) {
3137                        credits += ocfs2_remove_extent_credits(inode->i_sb);
3138                        goto out;
3139                } else {
3140                        meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3141                        clusters_add += new_clusters - old_clusters;
3142                        credits += ocfs2_calc_extend_credits(inode->i_sb,
3143                                                             &xv->xr_list,
3144                                                             new_clusters -
3145                                                             old_clusters);
3146                        if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3147                                goto out;
3148                }
3149        } else {
3150                /*
3151                 * Now the new value will be stored inside. So if the new
3152                 * value is smaller than the size of value root or the old
3153                 * value, we don't need any allocation, otherwise we have
3154                 * to guess metadata allocation.
3155                 */
3156                if ((ocfs2_xattr_is_local(xe) &&
3157                     (value_size >= xi->xi_value_len)) ||
3158                    (!ocfs2_xattr_is_local(xe) &&
3159                     OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3160                        goto out;
3161        }
3162
3163meta_guess:
3164        /* calculate metadata allocation. */
3165        if (di->i_xattr_loc) {
3166                if (!xbs->xattr_bh) {
3167                        ret = ocfs2_read_xattr_block(inode,
3168                                                     le64_to_cpu(di->i_xattr_loc),
3169                                                     &bh);
3170                        if (ret) {
3171                                mlog_errno(ret);
3172                                goto out;
3173                        }
3174
3175                        xb = (struct ocfs2_xattr_block *)bh->b_data;
3176                } else
3177                        xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3178
3179                /*
3180                 * If there is already an xattr tree, good, we can calculate
3181                 * like other b-trees. Otherwise we may have the chance of
3182                 * create a tree, the credit calculation is borrowed from
3183                 * ocfs2_calc_extend_credits with root_el = NULL. And the
3184                 * new tree will be cluster based, so no meta is needed.
3185                 */
3186                if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3187                        struct ocfs2_extent_list *el =
3188                                 &xb->xb_attrs.xb_root.xt_list;
3189                        meta_add += ocfs2_extend_meta_needed(el);
3190                        credits += ocfs2_calc_extend_credits(inode->i_sb,
3191                                                             el, 1);
3192                } else
3193                        credits += OCFS2_SUBALLOC_ALLOC + 1;
3194
3195                /*
3196                 * This cluster will be used either for new bucket or for
3197                 * new xattr block.
3198                 * If the cluster size is the same as the bucket size, one
3199                 * more is needed since we may need to extend the bucket
3200                 * also.
3201                 */
3202                clusters_add += 1;
3203                credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3204                if (OCFS2_XATTR_BUCKET_SIZE ==
3205                        OCFS2_SB(inode->i_sb)->s_clustersize) {
3206                        credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3207                        clusters_add += 1;
3208                }
3209        } else {
3210                meta_add += 1;
3211                credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3212        }
3213out:
3214        if (clusters_need)
3215                *clusters_need = clusters_add;
3216        if (meta_need)
3217                *meta_need = meta_add;
3218        if (credits_need)
3219                *credits_need = credits;
3220        brelse(bh);
3221        return ret;
3222}
3223
3224static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3225                                     struct ocfs2_dinode *di,
3226                                     struct ocfs2_xattr_info *xi,
3227                                     struct ocfs2_xattr_search *xis,
3228                                     struct ocfs2_xattr_search *xbs,
3229                                     struct ocfs2_xattr_set_ctxt *ctxt,
3230                                     int extra_meta,
3231                                     int *credits)
3232{
3233        int clusters_add, meta_add, ret;
3234        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3235
3236        memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3237
3238        ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3239
3240        ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3241                                        &clusters_add, &meta_add, credits);
3242        if (ret) {
3243                mlog_errno(ret);
3244                return ret;
3245        }
3246
3247        meta_add += extra_meta;
3248        trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3249                                        clusters_add, *credits);
3250
3251        if (meta_add) {
3252                ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3253                                                        &ctxt->meta_ac);
3254                if (ret) {
3255                        mlog_errno(ret);
3256                        goto out;
3257                }
3258        }
3259
3260        if (clusters_add) {
3261                ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3262                if (ret)
3263                        mlog_errno(ret);
3264        }
3265out:
3266        if (ret) {
3267                if (ctxt->meta_ac) {
3268                        ocfs2_free_alloc_context(ctxt->meta_ac);
3269                        ctxt->meta_ac = NULL;
3270                }
3271
3272                /*
3273                 * We cannot have an error and a non null ctxt->data_ac.
3274                 */
3275        }
3276
3277        return ret;
3278}
3279
3280static int __ocfs2_xattr_set_handle(struct inode *inode,
3281                                    struct ocfs2_dinode *di,
3282                                    struct ocfs2_xattr_info *xi,
3283                                    struct ocfs2_xattr_search *xis,
3284                                    struct ocfs2_xattr_search *xbs,
3285                                    struct ocfs2_xattr_set_ctxt *ctxt)
3286{
3287        int ret = 0, credits, old_found;
3288
3289        if (!xi->xi_value) {
3290                /* Remove existing extended attribute */
3291                if (!xis->not_found)
3292                        ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3293                else if (!xbs->not_found)
3294                        ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3295        } else {
3296                /* We always try to set extended attribute into inode first*/
3297                ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3298                if (!ret && !xbs->not_found) {
3299                        /*
3300                         * If succeed and that extended attribute existing in
3301                         * external block, then we will remove it.
3302                         */
3303                        xi->xi_value = NULL;
3304                        xi->xi_value_len = 0;
3305
3306                        old_found = xis->not_found;
3307                        xis->not_found = -ENODATA;
3308                        ret = ocfs2_calc_xattr_set_need(inode,
3309                                                        di,
3310                                                        xi,
3311                                                        xis,
3312                                                        xbs,
3313                                                        NULL,
3314                                                        NULL,
3315                                                        &credits);
3316                        xis->not_found = old_found;
3317                        if (ret) {
3318                                mlog_errno(ret);
3319                                goto out;
3320                        }
3321
3322                        ret = ocfs2_extend_trans(ctxt->handle, credits);
3323                        if (ret) {
3324                                mlog_errno(ret);
3325                                goto out;
3326                        }
3327                        ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3328                } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3329                        if (di->i_xattr_loc && !xbs->xattr_bh) {
3330                                ret = ocfs2_xattr_block_find(inode,
3331                                                             xi->xi_name_index,
3332                                                             xi->xi_name, xbs);
3333                                if (ret)
3334                                        goto out;
3335
3336                                old_found = xis->not_found;
3337                                xis->not_found = -ENODATA;
3338                                ret = ocfs2_calc_xattr_set_need(inode,
3339                                                                di,
3340                                                                xi,
3341                                                                xis,
3342                                                                xbs,
3343                                                                NULL,
3344                                                                NULL,
3345                                                                &credits);
3346                                xis->not_found = old_found;
3347                                if (ret) {
3348                                        mlog_errno(ret);
3349                                        goto out;
3350                                }
3351
3352                                ret = ocfs2_extend_trans(ctxt->handle, credits);
3353                                if (ret) {
3354                                        mlog_errno(ret);
3355                                        goto out;
3356                                }
3357                        }
3358                        /*
3359                         * If no space in inode, we will set extended attribute
3360                         * into external block.
3361                         */
3362                        ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3363                        if (ret)
3364                                goto out;
3365                        if (!xis->not_found) {
3366                                /*
3367                                 * If succeed and that extended attribute
3368                                 * existing in inode, we will remove it.
3369                                 */
3370                                xi->xi_value = NULL;
3371                                xi->xi_value_len = 0;
3372                                xbs->not_found = -ENODATA;
3373                                ret = ocfs2_calc_xattr_set_need(inode,
3374                                                                di,
3375                                                                xi,
3376                                                                xis,
3377                                                                xbs,
3378                                                                NULL,
3379                                                                NULL,
3380                                                                &credits);
3381                                if (ret) {
3382                                        mlog_errno(ret);
3383                                        goto out;
3384                                }
3385
3386                                ret = ocfs2_extend_trans(ctxt->handle, credits);
3387                                if (ret) {
3388                                        mlog_errno(ret);
3389                                        goto out;
3390                                }
3391                                ret = ocfs2_xattr_ibody_set(inode, xi,
3392                                                            xis, ctxt);
3393                        }
3394                }
3395        }
3396
3397        if (!ret) {
3398                /* Update inode ctime. */
3399                ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3400                                              xis->inode_bh,
3401                                              OCFS2_JOURNAL_ACCESS_WRITE);
3402                if (ret) {
3403                        mlog_errno(ret);
3404                        goto out;
3405                }
3406
3407                inode->i_ctime = CURRENT_TIME;
3408                di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3409                di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3410                ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3411        }
3412out:
3413        return ret;
3414}
3415
3416/*
3417 * This function only called duing creating inode
3418 * for init security/acl xattrs of the new inode.
3419 * All transanction credits have been reserved in mknod.
3420 */
3421int ocfs2_xattr_set_handle(handle_t *handle,
3422                           struct inode *inode,
3423                           struct buffer_head *di_bh,
3424                           int name_index,
3425                           const char *name,
3426                           const void *value,
3427                           size_t value_len,
3428                           int flags,
3429                           struct ocfs2_alloc_context *meta_ac,
3430                           struct ocfs2_alloc_context *data_ac)
3431{
3432        struct ocfs2_dinode *di;
3433        int ret;
3434
3435        struct ocfs2_xattr_info xi = {
3436                .xi_name_index = name_index,
3437                .xi_name = name,
3438                .xi_name_len = strlen(name),
3439                .xi_value = value,
3440                .xi_value_len = value_len,
3441        };
3442
3443        struct ocfs2_xattr_search xis = {
3444                .not_found = -ENODATA,
3445        };
3446
3447        struct ocfs2_xattr_search xbs = {
3448                .not_found = -ENODATA,
3449        };
3450
3451        struct ocfs2_xattr_set_ctxt ctxt = {
3452                .handle = handle,
3453                .meta_ac = meta_ac,
3454                .data_ac = data_ac,
3455        };
3456
3457        if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3458                return -EOPNOTSUPP;
3459
3460        /*
3461         * In extreme situation, may need xattr bucket when
3462         * block size is too small. And we have already reserved
3463         * the credits for bucket in mknod.
3464         */
3465        if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3466                xbs.bucket = ocfs2_xattr_bucket_new(inode);
3467                if (!xbs.bucket) {
3468                        mlog_errno(-ENOMEM);
3469                        return -ENOMEM;
3470                }
3471        }
3472
3473        xis.inode_bh = xbs.inode_bh = di_bh;
3474        di = (struct ocfs2_dinode *)di_bh->b_data;
3475
3476        down_write(&OCFS2_I(inode)->ip_xattr_sem);
3477
3478        ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3479        if (ret)
3480                goto cleanup;
3481        if (xis.not_found) {
3482                ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3483                if (ret)
3484                        goto cleanup;
3485        }
3486
3487        ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3488
3489cleanup:
3490        up_write(&OCFS2_I(inode)->ip_xattr_sem);
3491        brelse(xbs.xattr_bh);
3492        ocfs2_xattr_bucket_free(xbs.bucket);
3493
3494        return ret;
3495}
3496
3497/*
3498 * ocfs2_xattr_set()
3499 *
3500 * Set, replace or remove an extended attribute for this inode.
3501 * value is NULL to remove an existing extended attribute, else either
3502 * create or replace an extended attribute.
3503 */
3504int ocfs2_xattr_set(struct inode *inode,
3505                    int name_index,
3506                    const char *name,
3507                    const void *value,
3508                    size_t value_len,
3509                    int flags)
3510{
3511        struct buffer_head *di_bh = NULL;
3512        struct ocfs2_dinode *di;
3513        int ret, credits, ref_meta = 0, ref_credits = 0;
3514        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3515        struct inode *tl_inode = osb->osb_tl_inode;
3516        struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
3517        struct ocfs2_refcount_tree *ref_tree = NULL;
3518
3519        struct ocfs2_xattr_info xi = {
3520                .xi_name_index = name_index,
3521                .xi_name = name,
3522                .xi_name_len = strlen(name),
3523                .xi_value = value,
3524                .xi_value_len = value_len,
3525        };
3526
3527        struct ocfs2_xattr_search xis = {
3528                .not_found = -ENODATA,
3529        };
3530
3531        struct ocfs2_xattr_search xbs = {
3532                .not_found = -ENODATA,
3533        };
3534
3535        if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3536                return -EOPNOTSUPP;
3537
3538        /*
3539         * Only xbs will be used on indexed trees.  xis doesn't need a
3540         * bucket.
3541         */
3542        xbs.bucket = ocfs2_xattr_bucket_new(inode);
3543        if (!xbs.bucket) {
3544                mlog_errno(-ENOMEM);
3545                return -ENOMEM;
3546        }
3547
3548        ret = ocfs2_inode_lock(inode, &di_bh, 1);
3549        if (ret < 0) {
3550                mlog_errno(ret);
3551                goto cleanup_nolock;
3552        }
3553        xis.inode_bh = xbs.inode_bh = di_bh;
3554        di = (struct ocfs2_dinode *)di_bh->b_data;
3555
3556        down_write(&OCFS2_I(inode)->ip_xattr_sem);
3557        /*
3558         * Scan inode and external block to find the same name
3559         * extended attribute and collect search information.
3560         */
3561        ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3562        if (ret)
3563                goto cleanup;
3564        if (xis.not_found) {
3565                ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3566                if (ret)
3567                        goto cleanup;
3568        }
3569
3570        if (xis.not_found && xbs.not_found) {
3571                ret = -ENODATA;
3572                if (flags & XATTR_REPLACE)
3573                        goto cleanup;
3574                ret = 0;
3575                if (!value)
3576                        goto cleanup;
3577        } else {
3578                ret = -EEXIST;
3579                if (flags & XATTR_CREATE)
3580                        goto cleanup;
3581        }
3582
3583        /* Check whether the value is refcounted and do some preparation. */
3584        if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3585            (!xis.not_found || !xbs.not_found)) {
3586                ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3587                                                   &xis, &xbs, &ref_tree,
3588                                                   &ref_meta, &ref_credits);
3589                if (ret) {
3590                        mlog_errno(ret);
3591                        goto cleanup;
3592                }
3593        }
3594
3595        mutex_lock(&tl_inode->i_mutex);
3596
3597        if (ocfs2_truncate_log_needs_flush(osb)) {
3598                ret = __ocfs2_flush_truncate_log(osb);
3599                if (ret < 0) {
3600                        mutex_unlock(&tl_inode->i_mutex);
3601                        mlog_errno(ret);
3602                        goto cleanup;
3603                }
3604        }
3605        mutex_unlock(&tl_inode->i_mutex);
3606
3607        ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3608                                        &xbs, &ctxt, ref_meta, &credits);
3609        if (ret) {
3610                mlog_errno(ret);
3611                goto cleanup;
3612        }
3613
3614        /* we need to update inode's ctime field, so add credit for it. */
3615        credits += OCFS2_INODE_UPDATE_CREDITS;
3616        ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3617        if (IS_ERR(ctxt.handle)) {
3618                ret = PTR_ERR(ctxt.handle);
3619                mlog_errno(ret);
3620                goto cleanup;
3621        }
3622
3623        ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3624
3625        ocfs2_commit_trans(osb, ctxt.handle);
3626
3627        if (ctxt.data_ac)
3628                ocfs2_free_alloc_context(ctxt.data_ac);
3629        if (ctxt.meta_ac)
3630                ocfs2_free_alloc_context(ctxt.meta_ac);
3631        if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3632                ocfs2_schedule_truncate_log_flush(osb, 1);
3633        ocfs2_run_deallocs(osb, &ctxt.dealloc);
3634
3635cleanup:
3636        if (ref_tree)
3637                ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3638        up_write(&OCFS2_I(inode)->ip_xattr_sem);
3639        if (!value && !ret) {
3640                ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3641                if (ret)
3642                        mlog_errno(ret);
3643        }
3644        ocfs2_inode_unlock(inode, 1);
3645cleanup_nolock:
3646        brelse(di_bh);
3647        brelse(xbs.xattr_bh);
3648        ocfs2_xattr_bucket_free(xbs.bucket);
3649
3650        return ret;
3651}
3652
3653/*
3654 * Find the xattr extent rec which may contains name_hash.
3655 * e_cpos will be the first name hash of the xattr rec.
3656 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3657 */
3658static int ocfs2_xattr_get_rec(struct inode *inode,
3659                               u32 name_hash,
3660                               u64 *p_blkno,
3661                               u32 *e_cpos,
3662                               u32 *num_clusters,
3663                               struct ocfs2_extent_list *el)
3664{
3665        int ret = 0, i;
3666        struct buffer_head *eb_bh = NULL;
3667        struct ocfs2_extent_block *eb;
3668        struct ocfs2_extent_rec *rec = NULL;
3669        u64 e_blkno = 0;
3670
3671        if (el->l_tree_depth) {
3672                ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3673                                      &eb_bh);
3674                if (ret) {
3675                        mlog_errno(ret);
3676                        goto out;
3677                }
3678
3679                eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3680                el = &eb->h_list;
3681
3682                if (el->l_tree_depth) {
3683                        ocfs2_error(inode->i_sb,
3684                                    "Inode %lu has non zero tree depth in "
3685                                    "xattr tree block %llu\n", inode->i_ino,
3686                                    (unsigned long long)eb_bh->b_blocknr);
3687                        ret = -EROFS;
3688                        goto out;
3689                }
3690        }
3691
3692        for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3693                rec = &el->l_recs[i];
3694
3695                if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3696                        e_blkno = le64_to_cpu(rec->e_blkno);
3697                        break;
3698                }
3699        }
3700
3701        if (!e_blkno) {
3702                ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3703                            "record (%u, %u, 0) in xattr", inode->i_ino,
3704                            le32_to_cpu(rec->e_cpos),
3705                            ocfs2_rec_clusters(el, rec));
3706                ret = -EROFS;
3707                goto out;
3708        }
3709
3710        *p_blkno = le64_to_cpu(rec->e_blkno);
3711        *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3712        if (e_cpos)
3713                *e_cpos = le32_to_cpu(rec->e_cpos);
3714out:
3715        brelse(eb_bh);
3716        return ret;
3717}
3718
3719typedef int (xattr_bucket_func)(struct inode *inode,
3720                                struct ocfs2_xattr_bucket *bucket,
3721                                void *para);
3722
3723static int ocfs2_find_xe_in_bucket(struct inode *inode,
3724                                   struct ocfs2_xattr_bucket *bucket,
3725                                   int name_index,
3726                                   const char *name,
3727                                   u32 name_hash,
3728                                   u16 *xe_index,
3729                                   int *found)
3730{
3731        int i, ret = 0, cmp = 1, block_off, new_offset;
3732        struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3733        size_t name_len = strlen(name);
3734        struct ocfs2_xattr_entry *xe = NULL;
3735        char *xe_name;
3736
3737        /*
3738         * We don't use binary search in the bucket because there
3739         * may be multiple entries with the same name hash.
3740         */
3741        for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3742                xe = &xh->xh_entries[i];
3743
3744                if (name_hash > le32_to_cpu(xe->xe_name_hash))
3745                        continue;
3746                else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3747                        break;
3748
3749                cmp = name_index - ocfs2_xattr_get_type(xe);
3750                if (!cmp)
3751                        cmp = name_len - xe->xe_name_len;
3752                if (cmp)
3753                        continue;
3754
3755                ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3756                                                        xh,
3757                                                        i,
3758                                                        &block_off,
3759                                                        &new_offset);
3760                if (ret) {
3761                        mlog_errno(ret);
3762                        break;
3763                }
3764
3765
3766                xe_name = bucket_block(bucket, block_off) + new_offset;
3767                if (!memcmp(name, xe_name, name_len)) {
3768                        *xe_index = i;
3769                        *found = 1;
3770                        ret = 0;
3771                        break;
3772                }
3773        }
3774
3775        return ret;
3776}
3777
3778/*
3779 * Find the specified xattr entry in a series of buckets.
3780 * This series start from p_blkno and last for num_clusters.
3781 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3782 * the num of the valid buckets.
3783 *
3784 * Return the buffer_head this xattr should reside in. And if the xattr's
3785 * hash is in the gap of 2 buckets, return the lower bucket.
3786 */
3787static int ocfs2_xattr_bucket_find(struct inode *inode,
3788                                   int name_index,
3789                                   const char *name,
3790                                   u32 name_hash,
3791                                   u64 p_blkno,
3792                                   u32 first_hash,
3793                                   u32 num_clusters,
3794                                   struct ocfs2_xattr_search *xs)
3795{
3796        int ret, found = 0;
3797        struct ocfs2_xattr_header *xh = NULL;
3798        struct ocfs2_xattr_entry *xe = NULL;
3799        u16 index = 0;
3800        u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3801        int low_bucket = 0, bucket, high_bucket;
3802        struct ocfs2_xattr_bucket *search;
3803        u32 last_hash;
3804        u64 blkno, lower_blkno = 0;
3805
3806        search = ocfs2_xattr_bucket_new(inode);
3807        if (!search) {
3808                ret = -ENOMEM;
3809                mlog_errno(ret);
3810                goto out;
3811        }
3812
3813        ret = ocfs2_read_xattr_bucket(search, p_blkno);
3814        if (ret) {
3815                mlog_errno(ret);
3816                goto out;
3817        }
3818
3819        xh = bucket_xh(search);
3820        high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3821        while (low_bucket <= high_bucket) {
3822                ocfs2_xattr_bucket_relse(search);
3823
3824                bucket = (low_bucket + high_bucket) / 2;
3825                blkno = p_blkno + bucket * blk_per_bucket;
3826                ret = ocfs2_read_xattr_bucket(search, blkno);
3827                if (ret) {
3828                        mlog_errno(ret);
3829                        goto out;
3830                }
3831
3832                xh = bucket_xh(search);
3833                xe = &xh->xh_entries[0];
3834                if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3835                        high_bucket = bucket - 1;
3836                        continue;
3837                }
3838
3839                /*
3840                 * Check whether the hash of the last entry in our
3841                 * bucket is larger than the search one. for an empty
3842                 * bucket, the last one is also the first one.
3843                 */
3844                if (xh->xh_count)
3845                        xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3846
3847                last_hash = le32_to_cpu(xe->xe_name_hash);
3848
3849                /* record lower_blkno which may be the insert place. */
3850                lower_blkno = blkno;
3851
3852                if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3853                        low_bucket = bucket + 1;
3854                        continue;
3855                }
3856
3857                /* the searched xattr should reside in this bucket if exists. */
3858                ret = ocfs2_find_xe_in_bucket(inode, search,
3859                                              name_index, name, name_hash,
3860                                              &index, &found);
3861                if (ret) {
3862                        mlog_errno(ret);
3863                        goto out;
3864                }
3865                break;
3866        }
3867
3868        /*
3869         * Record the bucket we have found.
3870         * When the xattr's hash value is in the gap of 2 buckets, we will
3871         * always set it to the previous bucket.
3872         */
3873        if (!lower_blkno)
3874                lower_blkno = p_blkno;
3875
3876        /* This should be in cache - we just read it during the search */
3877        ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3878        if (ret) {
3879                mlog_errno(ret);
3880                goto out;
3881        }
3882
3883        xs->header = bucket_xh(xs->bucket);
3884        xs->base = bucket_block(xs->bucket, 0);
3885        xs->end = xs->base + inode->i_sb->s_blocksize;
3886
3887        if (found) {
3888                xs->here = &xs->header->xh_entries[index];
3889                trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3890                        name, name_index, name_hash,
3891                        (unsigned long long)bucket_blkno(xs->bucket),
3892                        index);
3893        } else
3894                ret = -ENODATA;
3895
3896out:
3897        ocfs2_xattr_bucket_free(search);
3898        return ret;
3899}
3900
3901static int ocfs2_xattr_index_block_find(struct inode *inode,
3902                                        struct buffer_head *root_bh,
3903                                        int name_index,
3904                                        const char *name,
3905                                        struct ocfs2_xattr_search *xs)
3906{
3907        int ret;
3908        struct ocfs2_xattr_block *xb =
3909                        (struct ocfs2_xattr_block *)root_bh->b_data;
3910        struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3911        struct ocfs2_extent_list *el = &xb_root->xt_list;
3912        u64 p_blkno = 0;
3913        u32 first_hash, num_clusters = 0;
3914        u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3915
3916        if (le16_to_cpu(el->l_next_free_rec) == 0)
3917                return -ENODATA;
3918
3919        trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3920                                        name, name_index, name_hash,
3921                                        (unsigned long long)root_bh->b_blocknr,
3922                                        -1);
3923
3924        ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3925                                  &num_clusters, el);
3926        if (ret) {
3927                mlog_errno(ret);
3928                goto out;
3929        }
3930
3931        BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3932
3933        trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3934                                        name, name_index, first_hash,
3935                                        (unsigned long long)p_blkno,
3936                                        num_clusters);
3937
3938        ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3939                                      p_blkno, first_hash, num_clusters, xs);
3940
3941out:
3942        return ret;
3943}
3944
3945static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3946                                       u64 blkno,
3947                                       u32 clusters,
3948                                       xattr_bucket_func *func,
3949                                       void *para)
3950{
3951        int i, ret = 0;
3952        u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3953        u32 num_buckets = clusters * bpc;
3954        struct ocfs2_xattr_bucket *bucket;
3955
3956        bucket = ocfs2_xattr_bucket_new(inode);
3957        if (!bucket) {
3958                mlog_errno(-ENOMEM);
3959                return -ENOMEM;
3960        }
3961
3962        trace_ocfs2_iterate_xattr_buckets(
3963                (unsigned long long)OCFS2_I(inode)->ip_blkno,
3964                (unsigned long long)blkno, clusters);
3965
3966        for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3967                ret = ocfs2_read_xattr_bucket(bucket, blkno);
3968                if (ret) {
3969                        mlog_errno(ret);
3970                        break;
3971                }
3972
3973                /*
3974                 * The real bucket num in this series of blocks is stored
3975                 * in the 1st bucket.
3976                 */
3977                if (i == 0)
3978                        num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3979
3980                trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
3981                     le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3982                if (func) {
3983                        ret = func(inode, bucket, para);
3984                        if (ret && ret != -ERANGE)
3985                                mlog_errno(ret);
3986                        /* Fall through to bucket_relse() */
3987                }
3988
3989                ocfs2_xattr_bucket_relse(bucket);
3990                if (ret)
3991                        break;
3992        }
3993
3994        ocfs2_xattr_bucket_free(bucket);
3995        return ret;
3996}
3997
3998struct ocfs2_xattr_tree_list {
3999        char *buffer;
4000        size_t buffer_size;
4001        size_t result;
4002};
4003
4004static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4005                                             struct ocfs2_xattr_header *xh,
4006                                             int index,
4007                                             int *block_off,
4008                                             int *new_offset)
4009{
4010        u16 name_offset;
4011
4012        if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4013                return -EINVAL;
4014
4015        name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4016
4017        *block_off = name_offset >> sb->s_blocksize_bits;
4018        *new_offset = name_offset % sb->s_blocksize;
4019
4020        return 0;
4021}
4022
4023static int ocfs2_list_xattr_bucket(struct inode *inode,
4024                                   struct ocfs2_xattr_bucket *bucket,
4025                                   void *para)
4026{
4027        int ret = 0, type;
4028        struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4029        int i, block_off, new_offset;
4030        const char *prefix, *name;
4031
4032        for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4033                struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4034                type = ocfs2_xattr_get_type(entry);
4035                prefix = ocfs2_xattr_prefix(type);
4036
4037                if (prefix) {
4038                        ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4039                                                                bucket_xh(bucket),
4040                                                                i,
4041                                                                &block_off,
4042                                                                &new_offset);
4043                        if (ret)
4044                                break;
4045
4046                        name = (const char *)bucket_block(bucket, block_off) +
4047                                new_offset;
4048                        ret = ocfs2_xattr_list_entry(xl->buffer,
4049                                                     xl->buffer_size,
4050                                                     &xl->result,
4051                                                     prefix, name,
4052                                                     entry->xe_name_len);
4053                        if (ret)
4054                                break;
4055                }
4056        }
4057
4058        return ret;
4059}
4060
4061static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4062                                           struct buffer_head *blk_bh,
4063                                           xattr_tree_rec_func *rec_func,
4064                                           void *para)
4065{
4066        struct ocfs2_xattr_block *xb =
4067                        (struct ocfs2_xattr_block *)blk_bh->b_data;
4068        struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4069        int ret = 0;
4070        u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4071        u64 p_blkno = 0;
4072
4073        if (!el->l_next_free_rec || !rec_func)
4074                return 0;
4075
4076        while (name_hash > 0) {
4077                ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4078                                          &e_cpos, &num_clusters, el);
4079                if (ret) {
4080                        mlog_errno(ret);
4081                        break;
4082                }
4083
4084                ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4085                               num_clusters, para);
4086                if (ret) {
4087                        if (ret != -ERANGE)
4088                                mlog_errno(ret);
4089                        break;
4090                }
4091
4092                if (e_cpos == 0)
4093                        break;
4094
4095                name_hash = e_cpos - 1;
4096        }
4097
4098        return ret;
4099
4100}
4101
4102static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4103                                     struct buffer_head *root_bh,
4104                                     u64 blkno, u32 cpos, u32 len, void *para)
4105{
4106        return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4107                                           ocfs2_list_xattr_bucket, para);
4108}
4109
4110static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4111                                             struct buffer_head *blk_bh,
4112                                             char *buffer,
4113                                             size_t buffer_size)
4114{
4115        int ret;
4116        struct ocfs2_xattr_tree_list xl = {
4117                .buffer = buffer,
4118                .buffer_size = buffer_size,
4119                .result = 0,
4120        };
4121
4122        ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4123                                              ocfs2_list_xattr_tree_rec, &xl);
4124        if (ret) {
4125                mlog_errno(ret);
4126                goto out;
4127        }
4128
4129        ret = xl.result;
4130out:
4131        return ret;
4132}
4133
4134static int cmp_xe(const void *a, const void *b)
4135{
4136        const struct ocfs2_xattr_entry *l = a, *r = b;
4137        u32 l_hash = le32_to_cpu(l->xe_name_hash);
4138        u32 r_hash = le32_to_cpu(r->xe_name_hash);
4139
4140        if (l_hash > r_hash)
4141                return 1;
4142        if (l_hash < r_hash)
4143                return -1;
4144        return 0;
4145}
4146
4147static void swap_xe(void *a, void *b, int size)
4148{
4149        struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4150
4151        tmp = *l;
4152        memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4153        memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4154}
4155
4156/*
4157 * When the ocfs2_xattr_block is filled up, new bucket will be created
4158 * and all the xattr entries will be moved to the new bucket.
4159 * The header goes at the start of the bucket, and the names+values are
4160 * filled from the end.  This is why *target starts as the last buffer.
4161 * Note: we need to sort the entries since they are not saved in order
4162 * in the ocfs2_xattr_block.
4163 */
4164static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4165                                           struct buffer_head *xb_bh,
4166                                           struct ocfs2_xattr_bucket *bucket)
4167{
4168        int i, blocksize = inode->i_sb->s_blocksize;
4169        int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4170        u16 offset, size, off_change;
4171        struct ocfs2_xattr_entry *xe;
4172        struct ocfs2_xattr_block *xb =
4173                                (struct ocfs2_xattr_block *)xb_bh->b_data;
4174        struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4175        struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4176        u16 count = le16_to_cpu(xb_xh->xh_count);
4177        char *src = xb_bh->b_data;
4178        char *target = bucket_block(bucket, blks - 1);
4179
4180        trace_ocfs2_cp_xattr_block_to_bucket_begin(
4181                                (unsigned long long)xb_bh->b_blocknr,
4182                                (unsigned long long)bucket_blkno(bucket));
4183
4184        for (i = 0; i < blks; i++)
4185                memset(bucket_block(bucket, i), 0, blocksize);
4186
4187        /*
4188         * Since the xe_name_offset is based on ocfs2_xattr_header,
4189         * there is a offset change corresponding to the change of
4190         * ocfs2_xattr_header's position.
4191         */
4192        off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4193        xe = &xb_xh->xh_entries[count - 1];
4194        offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4195        size = blocksize - offset;
4196
4197        /* copy all the names and values. */
4198        memcpy(target + offset, src + offset, size);
4199
4200        /* Init new header now. */
4201        xh->xh_count = xb_xh->xh_count;
4202        xh->xh_num_buckets = cpu_to_le16(1);
4203        xh->xh_name_value_len = cpu_to_le16(size);
4204        xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4205
4206        /* copy all the entries. */
4207        target = bucket_block(bucket, 0);
4208        offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4209        size = count * sizeof(struct ocfs2_xattr_entry);
4210        memcpy(target + offset, (char *)xb_xh + offset, size);
4211
4212        /* Change the xe offset for all the xe because of the move. */
4213        off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4214                 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4215        for (i = 0; i < count; i++)
4216                le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4217
4218        trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4219
4220        sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4221             cmp_xe, swap_xe);
4222}
4223
4224/*
4225 * After we move xattr from block to index btree, we have to
4226 * update ocfs2_xattr_search to the new xe and base.
4227 *
4228 * When the entry is in xattr block, xattr_bh indicates the storage place.
4229 * While if the entry is in index b-tree, "bucket" indicates the
4230 * real place of the xattr.
4231 */
4232static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4233                                            struct ocfs2_xattr_search *xs,
4234                                            struct buffer_head *old_bh)
4235{
4236        char *buf = old_bh->b_data;
4237        struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4238        struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4239        int i;
4240
4241        xs->header = bucket_xh(xs->bucket);
4242        xs->base = bucket_block(xs->bucket, 0);
4243        xs->end = xs->base + inode->i_sb->s_blocksize;
4244
4245        if (xs->not_found)
4246                return;
4247
4248        i = xs->here - old_xh->xh_entries;
4249        xs->here = &xs->header->xh_entries[i];
4250}
4251
4252static int ocfs2_xattr_create_index_block(struct inode *inode,
4253                                          struct ocfs2_xattr_search *xs,
4254                                          struct ocfs2_xattr_set_ctxt *ctxt)
4255{
4256        int ret;
4257        u32 bit_off, len;
4258        u64 blkno;
4259        handle_t *handle = ctxt->handle;
4260        struct ocfs2_inode_info *oi = OCFS2_I(inode);
4261        struct buffer_head *xb_bh = xs->xattr_bh;
4262        struct ocfs2_xattr_block *xb =
4263                        (struct ocfs2_xattr_block *)xb_bh->b_data;
4264        struct ocfs2_xattr_tree_root *xr;
4265        u16 xb_flags = le16_to_cpu(xb->xb_flags);
4266
4267        trace_ocfs2_xattr_create_index_block_begin(
4268                                (unsigned long long)xb_bh->b_blocknr);
4269
4270        BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4271        BUG_ON(!xs->bucket);
4272
4273        /*
4274         * XXX:
4275         * We can use this lock for now, and maybe move to a dedicated mutex
4276         * if performance becomes a problem later.
4277         */
4278        down_write(&oi->ip_alloc_sem);
4279
4280        ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4281                                      OCFS2_JOURNAL_ACCESS_WRITE);
4282        if (ret) {
4283                mlog_errno(ret);
4284                goto out;
4285        }
4286
4287        ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4288                                     1, 1, &bit_off, &len);
4289        if (ret) {
4290                mlog_errno(ret);
4291                goto out;
4292        }
4293
4294        /*
4295         * The bucket may spread in many blocks, and
4296         * we will only touch the 1st block and the last block
4297         * in the whole bucket(one for entry and one for data).
4298         */
4299        blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4300
4301        trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4302
4303        ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
4304        if (ret) {
4305                mlog_errno(ret);
4306                goto out;
4307        }
4308
4309        ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4310                                                OCFS2_JOURNAL_ACCESS_CREATE);
4311        if (ret) {
4312                mlog_errno(ret);
4313                goto out;
4314        }
4315
4316        ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4317        ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4318
4319        ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4320
4321        /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4322        memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4323               offsetof(struct ocfs2_xattr_block, xb_attrs));
4324
4325        xr = &xb->xb_attrs.xb_root;
4326        xr->xt_clusters = cpu_to_le32(1);
4327        xr->xt_last_eb_blk = 0;
4328        xr->xt_list.l_tree_depth = 0;
4329        xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4330        xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4331
4332        xr->xt_list.l_recs[0].e_cpos = 0;
4333        xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4334        xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4335
4336        xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4337
4338        ocfs2_journal_dirty(handle, xb_bh);
4339
4340out:
4341        up_write(&oi->ip_alloc_sem);
4342
4343        return ret;
4344}
4345
4346static int cmp_xe_offset(const void *a, const void *b)
4347{
4348        const struct ocfs2_xattr_entry *l = a, *r = b;
4349        u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4350        u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4351
4352        if (l_name_offset < r_name_offset)
4353                return 1;
4354        if (l_name_offset > r_name_offset)
4355                return -1;
4356        return 0;
4357}
4358
4359/*
4360 * defrag a xattr bucket if we find that the bucket has some
4361 * holes beteen name/value pairs.
4362 * We will move all the name/value pairs to the end of the bucket
4363 * so that we can spare some space for insertion.
4364 */
4365static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4366                                     handle_t *handle,
4367                                     struct ocfs2_xattr_bucket *bucket)
4368{
4369        int ret, i;
4370        size_t end, offset, len;
4371        struct ocfs2_xattr_header *xh;
4372        char *entries, *buf, *bucket_buf = NULL;
4373        u64 blkno = bucket_blkno(bucket);
4374        u16 xh_free_start;
4375        size_t blocksize = inode->i_sb->s_blocksize;
4376        struct ocfs2_xattr_entry *xe;
4377
4378        /*
4379         * In order to make the operation more efficient and generic,
4380         * we copy all the blocks into a contiguous memory and do the
4381         * defragment there, so if anything is error, we will not touch
4382         * the real block.
4383         */
4384        bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4385        if (!bucket_buf) {
4386                ret = -EIO;
4387                goto out;
4388        }
4389
4390        buf = bucket_buf;
4391        for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4392                memcpy(buf, bucket_block(bucket, i), blocksize);
4393
4394        ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4395                                                OCFS2_JOURNAL_ACCESS_WRITE);
4396        if (ret < 0) {
4397                mlog_errno(ret);
4398                goto out;
4399        }
4400
4401        xh = (struct ocfs2_xattr_header *)bucket_buf;
4402        entries = (char *)xh->xh_entries;
4403        xh_free_start = le16_to_cpu(xh->xh_free_start);
4404
4405        trace_ocfs2_defrag_xattr_bucket(
4406             (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4407             xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4408
4409        /*
4410         * sort all the entries by their offset.
4411         * the largest will be the first, so that we can
4412         * move them to the end one by one.
4413         */
4414        sort(entries, le16_to_cpu(xh->xh_count),
4415             sizeof(struct ocfs2_xattr_entry),
4416             cmp_xe_offset, swap_xe);
4417
4418        /* Move all name/values to the end of the bucket. */
4419        xe = xh->xh_entries;
4420        end = OCFS2_XATTR_BUCKET_SIZE;
4421        for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4422                offset = le16_to_cpu(xe->xe_name_offset);
4423                len = namevalue_size_xe(xe);
4424
4425                /*
4426                 * We must make sure that the name/value pair
4427                 * exist in the same block. So adjust end to
4428                 * the previous block end if needed.
4429                 */
4430                if (((end - len) / blocksize !=
4431                        (end - 1) / blocksize))
4432                        end = end - end % blocksize;
4433
4434                if (end > offset + len) {
4435                        memmove(bucket_buf + end - len,
4436                                bucket_buf + offset, len);
4437                        xe->xe_name_offset = cpu_to_le16(end - len);
4438                }
4439
4440                mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4441                                "bucket %llu\n", (unsigned long long)blkno);
4442
4443                end -= len;
4444        }
4445
4446        mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4447                        "bucket %llu\n", (unsigned long long)blkno);
4448
4449        if (xh_free_start == end)
4450                goto out;
4451
4452        memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4453        xh->xh_free_start = cpu_to_le16(end);
4454
4455        /* sort the entries by their name_hash. */
4456        sort(entries, le16_to_cpu(xh->xh_count),
4457             sizeof(struct ocfs2_xattr_entry),
4458             cmp_xe, swap_xe);
4459
4460        buf = bucket_buf;
4461        for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4462                memcpy(bucket_block(bucket, i), buf, blocksize);
4463        ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4464
4465out:
4466        kfree(bucket_buf);
4467        return ret;
4468}
4469
4470/*
4471 * prev_blkno points to the start of an existing extent.  new_blkno
4472 * points to a newly allocated extent.  Because we know each of our
4473 * clusters contains more than bucket, we can easily split one cluster
4474 * at a bucket boundary.  So we take the last cluster of the existing
4475 * extent and split it down the middle.  We move the last half of the
4476 * buckets in the last cluster of the existing extent over to the new
4477 * extent.
4478 *
4479 * first_bh is the buffer at prev_blkno so we can update the existing
4480 * extent's bucket count.  header_bh is the bucket were we were hoping
4481 * to insert our xattr.  If the bucket move places the target in the new
4482 * extent, we'll update first_bh and header_bh after modifying the old
4483 * extent.
4484 *
4485 * first_hash will be set as the 1st xe's name_hash in the new extent.
4486 */
4487static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4488                                               handle_t *handle,
4489                                               struct ocfs2_xattr_bucket *first,
4490                                               struct ocfs2_xattr_bucket *target,
4491                                               u64 new_blkno,
4492                                               u32 num_clusters,
4493                                               u32 *first_hash)
4494{
4495        int ret;
4496        struct super_block *sb = inode->i_sb;
4497        int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4498        int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4499        int to_move = num_buckets / 2;
4500        u64 src_blkno;
4501        u64 last_cluster_blkno = bucket_blkno(first) +
4502                ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4503
4504        BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4505        BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4506
4507        trace_ocfs2_mv_xattr_bucket_cross_cluster(
4508                                (unsigned long long)last_cluster_blkno,
4509                                (unsigned long long)new_blkno);
4510
4511        ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4512                                     last_cluster_blkno, new_blkno,
4513                                     to_move, first_hash);
4514        if (ret) {
4515                mlog_errno(ret);
4516                goto out;
4517        }
4518
4519        /* This is the first bucket that got moved */
4520        src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4521
4522        /*
4523         * If the target bucket was part of the moved buckets, we need to
4524         * update first and target.
4525         */
4526        if (bucket_blkno(target) >= src_blkno) {
4527                /* Find the block for the new target bucket */
4528                src_blkno = new_blkno +
4529                        (bucket_blkno(target) - src_blkno);
4530
4531                ocfs2_xattr_bucket_relse(first);
4532                ocfs2_xattr_bucket_relse(target);
4533
4534                /*
4535                 * These shouldn't fail - the buffers are in the
4536                 * journal from ocfs2_cp_xattr_bucket().
4537                 */
4538                ret = ocfs2_read_xattr_bucket(first, new_blkno);
4539                if (ret) {
4540                        mlog_errno(ret);
4541                        goto out;
4542                }
4543                ret = ocfs2_read_xattr_bucket(target, src_blkno);
4544                if (ret)
4545                        mlog_errno(ret);
4546
4547        }
4548
4549out:
4550        return ret;
4551}
4552
4553/*
4554 * Find the suitable pos when we divide a bucket into 2.
4555 * We have to make sure the xattrs with the same hash value exist
4556 * in the same bucket.
4557 *
4558 * If this ocfs2_xattr_header covers more than one hash value, find a
4559 * place where the hash value changes.  Try to find the most even split.
4560 * The most common case is that all entries have different hash values,
4561 * and the first check we make will find a place to split.
4562 */
4563static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4564{
4565        struct ocfs2_xattr_entry *entries = xh->xh_entries;
4566        int count = le16_to_cpu(xh->xh_count);
4567        int delta, middle = count / 2;
4568
4569        /*
4570         * We start at the middle.  Each step gets farther away in both
4571         * directions.  We therefore hit the change in hash value
4572         * nearest to the middle.  Note that this loop does not execute for
4573         * count < 2.
4574         */
4575        for (delta = 0; delta < middle; delta++) {
4576                /* Let's check delta earlier than middle */
4577                if (cmp_xe(&entries[middle - delta - 1],
4578                           &entries[middle - delta]))
4579                        return middle - delta;
4580
4581                /* For even counts, don't walk off the end */
4582                if ((middle + delta + 1) == count)
4583                        continue;
4584
4585                /* Now try delta past middle */
4586                if (cmp_xe(&entries[middle + delta],
4587                           &entries[middle + delta + 1]))
4588                        return middle + delta + 1;
4589        }
4590
4591        /* Every entry had the same hash */
4592        return count;
4593}
4594
4595/*
4596 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4597 * first_hash will record the 1st hash of the new bucket.
4598 *
4599 * Normally half of the xattrs will be moved.  But we have to make
4600 * sure that the xattrs with the same hash value are stored in the
4601 * same bucket. If all the xattrs in this bucket have the same hash
4602 * value, the new bucket will be initialized as an empty one and the
4603 * first_hash will be initialized as (hash_value+1).
4604 */
4605static int ocfs2_divide_xattr_bucket(struct inode *inode,
4606                                    handle_t *handle,
4607                                    u64 blk,
4608                                    u64 new_blk,
4609                                    u32 *first_hash,
4610                                    int new_bucket_head)
4611{
4612        int ret, i;
4613        int count, start, len, name_value_len = 0, name_offset = 0;
4614        struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4615        struct ocfs2_xattr_header *xh;
4616        struct ocfs2_xattr_entry *xe;
4617        int blocksize = inode->i_sb->s_blocksize;
4618
4619        trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4620                                              (unsigned long long)new_blk);
4621
4622        s_bucket = ocfs2_xattr_bucket_new(inode);
4623        t_bucket = ocfs2_xattr_bucket_new(inode);
4624        if (!s_bucket || !t_bucket) {
4625                ret = -ENOMEM;
4626                mlog_errno(ret);
4627                goto out;
4628        }
4629
4630        ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4631        if (ret) {
4632                mlog_errno(ret);
4633                goto out;
4634        }
4635
4636        ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4637                                                OCFS2_JOURNAL_ACCESS_WRITE);
4638        if (ret) {
4639                mlog_errno(ret);
4640                goto out;
4641        }
4642
4643        /*
4644         * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4645         * there's no need to read it.
4646         */
4647        ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4648        if (ret) {
4649                mlog_errno(ret);
4650                goto out;
4651        }
4652
4653        /*
4654         * Hey, if we're overwriting t_bucket, what difference does
4655         * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4656         * same part of ocfs2_cp_xattr_bucket().
4657         */
4658        ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4659                                                new_bucket_head ?
4660                                                OCFS2_JOURNAL_ACCESS_CREATE :
4661                                                OCFS2_JOURNAL_ACCESS_WRITE);
4662        if (ret) {
4663                mlog_errno(ret);
4664                goto out;
4665        }
4666
4667        xh = bucket_xh(s_bucket);
4668        count = le16_to_cpu(xh->xh_count);
4669        start = ocfs2_xattr_find_divide_pos(xh);
4670
4671        if (start == count) {
4672                xe = &xh->xh_entries[start-1];
4673
4674                /*
4675                 * initialized a new empty bucket here.
4676                 * The hash value is set as one larger than
4677                 * that of the last entry in the previous bucket.
4678                 */
4679                for (i = 0; i < t_bucket->bu_blocks; i++)
4680                        memset(bucket_block(t_bucket, i), 0, blocksize);
4681
4682                xh = bucket_xh(t_bucket);
4683                xh->xh_free_start = cpu_to_le16(blocksize);
4684                xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4685                le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4686
4687                goto set_num_buckets;
4688        }
4689
4690        /* copy the whole bucket to the new first. */
4691        ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4692
4693        /* update the new bucket. */
4694        xh = bucket_xh(t_bucket);
4695
4696        /*
4697         * Calculate the total name/value len and xh_free_start for
4698         * the old bucket first.
4699         */
4700        name_offset = OCFS2_XATTR_BUCKET_SIZE;
4701        name_value_len = 0;
4702        for (i = 0; i < start; i++) {
4703                xe = &xh->xh_entries[i];
4704                name_value_len += namevalue_size_xe(xe);
4705                if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4706                        name_offset = le16_to_cpu(xe->xe_name_offset);
4707        }
4708
4709        /*
4710         * Now begin the modification to the new bucket.
4711         *
4712         * In the new bucket, We just move the xattr entry to the beginning
4713         * and don't touch the name/value. So there will be some holes in the
4714         * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4715         * called.
4716         */
4717        xe = &xh->xh_entries[start];
4718        len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4719        trace_ocfs2_divide_xattr_bucket_move(len,
4720                        (int)((char *)xe - (char *)xh),
4721                        (int)((char *)xh->xh_entries - (char *)xh));
4722        memmove((char *)xh->xh_entries, (char *)xe, len);
4723        xe = &xh->xh_entries[count - start];
4724        len = sizeof(struct ocfs2_xattr_entry) * start;
4725        memset((char *)xe, 0, len);
4726
4727        le16_add_cpu(&xh->xh_count, -start);
4728        le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4729
4730        /* Calculate xh_free_start for the new bucket. */
4731        xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4732        for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4733                xe = &xh->xh_entries[i];
4734                if (le16_to_cpu(xe->xe_name_offset) <
4735                    le16_to_cpu(xh->xh_free_start))
4736                        xh->xh_free_start = xe->xe_name_offset;
4737        }
4738
4739set_num_buckets:
4740        /* set xh->xh_num_buckets for the new xh. */
4741        if (new_bucket_head)
4742                xh->xh_num_buckets = cpu_to_le16(1);
4743        else
4744                xh->xh_num_buckets = 0;
4745
4746        ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4747
4748        /* store the first_hash of the new bucket. */
4749        if (first_hash)
4750                *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4751
4752        /*
4753         * Now only update the 1st block of the old bucket.  If we
4754         * just added a new empty bucket, there is no need to modify
4755         * it.
4756         */
4757        if (start == count)
4758                goto out;
4759
4760        xh = bucket_xh(s_bucket);
4761        memset(&xh->xh_entries[start], 0,
4762               sizeof(struct ocfs2_xattr_entry) * (count - start));
4763        xh->xh_count = cpu_to_le16(start);
4764        xh->xh_free_start = cpu_to_le16(name_offset);
4765        xh->xh_name_value_len = cpu_to_le16(name_value_len);
4766
4767        ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4768
4769out:
4770        ocfs2_xattr_bucket_free(s_bucket);
4771        ocfs2_xattr_bucket_free(t_bucket);
4772
4773        return ret;
4774}
4775
4776/*
4777 * Copy xattr from one bucket to another bucket.
4778 *
4779 * The caller must make sure that the journal transaction
4780 * has enough space for journaling.
4781 */
4782static int ocfs2_cp_xattr_bucket(struct inode *inode,
4783                                 handle_t *handle,
4784                                 u64 s_blkno,
4785                                 u64 t_blkno,
4786                                 int t_is_new)
4787{
4788        int ret;
4789        struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4790
4791        BUG_ON(s_blkno == t_blkno);
4792
4793        trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4794                                    (unsigned long long)t_blkno,
4795                                    t_is_new);
4796
4797        s_bucket = ocfs2_xattr_bucket_new(inode);
4798        t_bucket = ocfs2_xattr_bucket_new(inode);
4799        if (!s_bucket || !t_bucket) {
4800                ret = -ENOMEM;
4801                mlog_errno(ret);
4802                goto out;
4803        }
4804
4805        ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4806        if (ret)
4807                goto out;
4808
4809        /*
4810         * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4811         * there's no need to read it.
4812         */
4813        ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4814        if (ret)
4815                goto out;
4816
4817        /*
4818         * Hey, if we're overwriting t_bucket, what difference does
4819         * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4820         * cluster to fill, we came here from
4821         * ocfs2_mv_xattr_buckets(), and it is really new -
4822         * ACCESS_CREATE is required.  But we also might have moved data
4823         * out of t_bucket before extending back into it.
4824         * ocfs2_add_new_xattr_bucket() can do this - its call to
4825         * ocfs2_add_new_xattr_cluster() may have created a new extent
4826         * and copied out the end of the old extent.  Then it re-extends
4827         * the old extent back to create space for new xattrs.  That's
4828         * how we get here, and the bucket isn't really new.
4829         */
4830        ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4831                                                t_is_new ?
4832                                                OCFS2_JOURNAL_ACCESS_CREATE :
4833                                                OCFS2_JOURNAL_ACCESS_WRITE);
4834        if (ret)
4835                goto out;
4836
4837        ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4838        ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4839
4840out:
4841        ocfs2_xattr_bucket_free(t_bucket);
4842        ocfs2_xattr_bucket_free(s_bucket);
4843
4844        return ret;
4845}
4846
4847/*
4848 * src_blk points to the start of an existing extent.  last_blk points to
4849 * last cluster in that extent.  to_blk points to a newly allocated
4850 * extent.  We copy the buckets from the cluster at last_blk to the new
4851 * extent.  If start_bucket is non-zero, we skip that many buckets before
4852 * we start copying.  The new extent's xh_num_buckets gets set to the
4853 * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4854 * by the same amount.
4855 */
4856static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4857                                  u64 src_blk, u64 last_blk, u64 to_blk,
4858                                  unsigned int start_bucket,
4859                                  u32 *first_hash)
4860{
4861        int i, ret, credits;
4862        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4863        int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4864        int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4865        struct ocfs2_xattr_bucket *old_first, *new_first;
4866
4867        trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4868                                     (unsigned long long)to_blk);
4869
4870        BUG_ON(start_bucket >= num_buckets);
4871        if (start_bucket) {
4872                num_buckets -= start_bucket;
4873                last_blk += (start_bucket * blks_per_bucket);
4874        }
4875
4876        /* The first bucket of the original extent */
4877        old_first = ocfs2_xattr_bucket_new(inode);
4878        /* The first bucket of the new extent */
4879        new_first = ocfs2_xattr_bucket_new(inode);
4880        if (!old_first || !new_first) {
4881                ret = -ENOMEM;
4882                mlog_errno(ret);
4883                goto out;
4884        }
4885
4886        ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4887        if (ret) {
4888                mlog_errno(ret);
4889                goto out;
4890        }
4891
4892        /*
4893         * We need to update the first bucket of the old extent and all
4894         * the buckets going to the new extent.
4895         */
4896        credits = ((num_buckets + 1) * blks_per_bucket);
4897        ret = ocfs2_extend_trans(handle, credits);
4898        if (ret) {
4899                mlog_errno(ret);
4900                goto out;
4901        }
4902
4903        ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4904                                                OCFS2_JOURNAL_ACCESS_WRITE);
4905        if (ret) {
4906                mlog_errno(ret);
4907                goto out;
4908        }
4909
4910        for (i = 0; i < num_buckets; i++) {
4911                ret = ocfs2_cp_xattr_bucket(inode, handle,
4912                                            last_blk + (i * blks_per_bucket),
4913                                            to_blk + (i * blks_per_bucket),
4914                                            1);
4915                if (ret) {
4916                        mlog_errno(ret);
4917                        goto out;
4918                }
4919        }
4920
4921        /*
4922         * Get the new bucket ready before we dirty anything
4923         * (This actually shouldn't fail, because we already dirtied
4924         * it once in ocfs2_cp_xattr_bucket()).
4925         */
4926        ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4927        if (ret) {
4928                mlog_errno(ret);
4929                goto out;
4930        }
4931        ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4932                                                OCFS2_JOURNAL_ACCESS_WRITE);
4933        if (ret) {
4934                mlog_errno(ret);
4935                goto out;
4936        }
4937
4938        /* Now update the headers */
4939        le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4940        ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4941
4942        bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4943        ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4944
4945        if (first_hash)
4946                *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4947
4948out:
4949        ocfs2_xattr_bucket_free(new_first);
4950        ocfs2_xattr_bucket_free(old_first);
4951        return ret;
4952}
4953
4954/*
4955 * Move some xattrs in this cluster to the new cluster.
4956 * This function should only be called when bucket size == cluster size.
4957 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4958 */
4959static int ocfs2_divide_xattr_cluster(struct inode *inode,
4960                                      handle_t *handle,
4961                                      u64 prev_blk,
4962                                      u64 new_blk,
4963                                      u32 *first_hash)
4964{
4965        u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4966        int ret, credits = 2 * blk_per_bucket;
4967
4968        BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4969
4970        ret = ocfs2_extend_trans(handle, credits);
4971        if (ret) {
4972                mlog_errno(ret);
4973                return ret;
4974        }
4975
4976        /* Move half of the xattr in start_blk to the next bucket. */
4977        return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4978                                          new_blk, first_hash, 1);
4979}
4980
4981/*
4982 * Move some xattrs from the old cluster to the new one since they are not
4983 * contiguous in ocfs2 xattr tree.
4984 *
4985 * new_blk starts a new separate cluster, and we will move some xattrs from
4986 * prev_blk to it. v_start will be set as the first name hash value in this
4987 * new cluster so that it can be used as e_cpos during tree insertion and
4988 * don't collide with our original b-tree operations. first_bh and header_bh
4989 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4990 * to extend the insert bucket.
4991 *
4992 * The problem is how much xattr should we move to the new one and when should
4993 * we update first_bh and header_bh?
4994 * 1. If cluster size > bucket size, that means the previous cluster has more
4995 *    than 1 bucket, so just move half nums of bucket into the new cluster and
4996 *    update the first_bh and header_bh if the insert bucket has been moved
4997 *    to the new cluster.
4998 * 2. If cluster_size == bucket_size:
4999 *    a) If the previous extent rec has more than one cluster and the insert
5000 *       place isn't in the last cluster, copy the entire last cluster to the
5001 *       new one. This time, we don't need to upate the first_bh and header_bh
5002 *       since they will not be moved into the new cluster.
5003 *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
5004 *       the new one. And we set the extend flag to zero if the insert place is
5005 *       moved into the new allocated cluster since no extend is needed.
5006 */
5007static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5008                                            handle_t *handle,
5009                                            struct ocfs2_xattr_bucket *first,
5010                                            struct ocfs2_xattr_bucket *target,
5011                                            u64 new_blk,
5012                                            u32 prev_clusters,
5013                                            u32 *v_start,
5014                                            int *extend)
5015{
5016        int ret;
5017
5018        trace_ocfs2_adjust_xattr_cross_cluster(
5019                        (unsigned long long)bucket_blkno(first),
5020                        (unsigned long long)new_blk, prev_clusters);
5021
5022        if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5023                ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5024                                                          handle,
5025                                                          first, target,
5026                                                          new_blk,
5027                                                          prev_clusters,
5028                                                          v_start);
5029                if (ret)
5030                        mlog_errno(ret);
5031        } else {
5032                /* The start of the last cluster in the first extent */
5033                u64 last_blk = bucket_blkno(first) +
5034                        ((prev_clusters - 1) *
5035                         ocfs2_clusters_to_blocks(inode->i_sb, 1));
5036
5037                if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5038                        ret = ocfs2_mv_xattr_buckets(inode, handle,
5039                                                     bucket_blkno(first),
5040                                                     last_blk, new_blk, 0,
5041                                                     v_start);
5042                        if (ret)
5043                                mlog_errno(ret);
5044                } else {
5045                        ret = ocfs2_divide_xattr_cluster(inode, handle,
5046                                                         last_blk, new_blk,
5047                                                         v_start);
5048                        if (ret)
5049                                mlog_errno(ret);
5050
5051                        if ((bucket_blkno(target) == last_blk) && extend)
5052                                *extend = 0;
5053                }
5054        }
5055
5056        return ret;
5057}
5058
5059/*
5060 * Add a new cluster for xattr storage.
5061 *
5062 * If the new cluster is contiguous with the previous one, it will be
5063 * appended to the same extent record, and num_clusters will be updated.
5064 * If not, we will insert a new extent for it and move some xattrs in
5065 * the last cluster into the new allocated one.
5066 * We also need to limit the maximum size of a btree leaf, otherwise we'll
5067 * lose the benefits of hashing because we'll have to search large leaves.
5068 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5069 * if it's bigger).
5070 *
5071 * first_bh is the first block of the previous extent rec and header_bh
5072 * indicates the bucket we will insert the new xattrs. They will be updated
5073 * when the header_bh is moved into the new cluster.
5074 */
5075static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5076                                       struct buffer_head *root_bh,
5077                                       struct ocfs2_xattr_bucket *first,
5078                                       struct ocfs2_xattr_bucket *target,
5079                                       u32 *num_clusters,
5080                                       u32 prev_cpos,
5081                                       int *extend,
5082                                       struct ocfs2_xattr_set_ctxt *ctxt)
5083{
5084        int ret;
5085        u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5086        u32 prev_clusters = *num_clusters;
5087        u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5088        u64 block;
5089        handle_t *handle = ctxt->handle;
5090        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5091        struct ocfs2_extent_tree et;
5092
5093        trace_ocfs2_add_new_xattr_cluster_begin(
5094                (unsigned long long)OCFS2_I(inode)->ip_blkno,
5095                (unsigned long long)bucket_blkno(first),
5096                prev_cpos, prev_clusters);
5097
5098        ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5099
5100        ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5101                                      OCFS2_JOURNAL_ACCESS_WRITE);
5102        if (ret < 0) {
5103                mlog_errno(ret);
5104                goto leave;
5105        }
5106
5107        ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5108                                     clusters_to_add, &bit_off, &num_bits);
5109        if (ret < 0) {
5110                if (ret != -ENOSPC)
5111                        mlog_errno(ret);
5112                goto leave;
5113        }
5114
5115        BUG_ON(num_bits > clusters_to_add);
5116
5117        block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5118        trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5119
5120        if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5121            (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5122             OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5123                /*
5124                 * If this cluster is contiguous with the old one and
5125                 * adding this new cluster, we don't surpass the limit of
5126                 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5127                 * initialized and used like other buckets in the previous
5128                 * cluster.
5129                 * So add it as a contiguous one. The caller will handle
5130                 * its init process.
5131                 */
5132                v_start = prev_cpos + prev_clusters;
5133                *num_clusters = prev_clusters + num_bits;
5134        } else {
5135                ret = ocfs2_adjust_xattr_cross_cluster(inode,
5136                                                       handle,
5137                                                       first,
5138                                                       target,
5139                                                       block,
5140                                                       prev_clusters,
5141                                                       &v_start,
5142                                                       extend);
5143                if (ret) {
5144                        mlog_errno(ret);
5145                        goto leave;
5146                }
5147        }
5148
5149        trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5150                                                 v_start, num_bits);
5151        ret = ocfs2_insert_extent(handle, &et, v_start, block,
5152                                  num_bits, 0, ctxt->meta_ac);
5153        if (ret < 0) {
5154                mlog_errno(ret);
5155                goto leave;
5156        }
5157
5158        ocfs2_journal_dirty(handle, root_bh);
5159
5160leave:
5161        return ret;
5162}
5163
5164/*
5165 * We are given an extent.  'first' is the bucket at the very front of
5166 * the extent.  The extent has space for an additional bucket past
5167 * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5168 * of the target bucket.  We wish to shift every bucket past the target
5169 * down one, filling in that additional space.  When we get back to the
5170 * target, we split the target between itself and the now-empty bucket
5171 * at target+1 (aka, target_blkno + blks_per_bucket).
5172 */
5173static int ocfs2_extend_xattr_bucket(struct inode *inode,
5174                                     handle_t *handle,
5175                                     struct ocfs2_xattr_bucket *first,
5176                                     u64 target_blk,
5177                                     u32 num_clusters)
5178{
5179        int ret, credits;
5180        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5181        u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5182        u64 end_blk;
5183        u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5184
5185        trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5186                                        (unsigned long long)bucket_blkno(first),
5187                                        num_clusters, new_bucket);
5188
5189        /* The extent must have room for an additional bucket */
5190        BUG_ON(new_bucket >=
5191               (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5192
5193        /* end_blk points to the last existing bucket */
5194        end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5195
5196        /*
5197         * end_blk is the start of the last existing bucket.
5198         * Thus, (end_blk - target_blk) covers the target bucket and
5199         * every bucket after it up to, but not including, the last
5200         * existing bucket.  Then we add the last existing bucket, the
5201         * new bucket, and the first bucket (3 * blk_per_bucket).
5202         */
5203        credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5204        ret = ocfs2_extend_trans(handle, credits);
5205        if (ret) {
5206                mlog_errno(ret);
5207                goto out;
5208        }
5209
5210        ret = ocfs2_xattr_bucket_journal_access(handle, first,
5211                                                OCFS2_JOURNAL_ACCESS_WRITE);
5212        if (ret) {
5213                mlog_errno(ret);
5214                goto out;
5215        }
5216
5217        while (end_blk != target_blk) {
5218                ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5219                                            end_blk + blk_per_bucket, 0);
5220                if (ret)
5221                        goto out;
5222                end_blk -= blk_per_bucket;
5223        }
5224
5225        /* Move half of the xattr in target_blkno to the next bucket. */
5226        ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5227                                        target_blk + blk_per_bucket, NULL, 0);
5228
5229        le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5230        ocfs2_xattr_bucket_journal_dirty(handle, first);
5231
5232out:
5233        return ret;
5234}
5235
5236/*
5237 * Add new xattr bucket in an extent record and adjust the buckets
5238 * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5239 * bucket we want to insert into.
5240 *
5241 * In the easy case, we will move all the buckets after target down by
5242 * one. Half of target's xattrs will be moved to the next bucket.
5243 *
5244 * If current cluster is full, we'll allocate a new one.  This may not
5245 * be contiguous.  The underlying calls will make sure that there is
5246 * space for the insert, shifting buckets around if necessary.
5247 * 'target' may be moved by those calls.
5248 */
5249static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5250                                      struct buffer_head *xb_bh,
5251                                      struct ocfs2_xattr_bucket *target,
5252                                      struct ocfs2_xattr_set_ctxt *ctxt)
5253{
5254        struct ocfs2_xattr_block *xb =
5255                        (struct ocfs2_xattr_block *)xb_bh->b_data;
5256        struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5257        struct ocfs2_extent_list *el = &xb_root->xt_list;
5258        u32 name_hash =
5259                le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5260        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5261        int ret, num_buckets, extend = 1;
5262        u64 p_blkno;
5263        u32 e_cpos, num_clusters;
5264        /* The bucket at the front of the extent */
5265        struct ocfs2_xattr_bucket *first;
5266
5267        trace_ocfs2_add_new_xattr_bucket(
5268                                (unsigned long long)bucket_blkno(target));
5269
527