ocfs2: Find proper end cpos for a leaf refcount block.
[safe/jmp/linux-2.6] / fs / ocfs2 / refcounttree.c
index a5b5bef..9c3956f 100644 (file)
@@ -33,6 +33,7 @@
 #include "extent_map.h"
 #include "aops.h"
 #include "xattr.h"
+#include "namei.h"
 
 #include <linux/bio.h>
 #include <linux/blkdev.h>
 #include <linux/writeback.h>
 #include <linux/pagevec.h>
 #include <linux/swap.h>
+#include <linux/security.h>
+#include <linux/fsnotify.h>
+#include <linux/quotaops.h>
+#include <linux/namei.h>
+#include <linux/mount.h>
 
 struct ocfs2_cow_context {
        struct inode *inode;
@@ -927,6 +933,139 @@ out:
 }
 
 /*
+ * Try to remove refcount tree. The mechanism is:
+ * 1) Check whether i_clusters == 0, if no, exit.
+ * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
+ * 3) Check whether we have inline xattr stored outside, if yes, exit.
+ * 4) Remove the tree.
+ */
+int ocfs2_try_remove_refcount_tree(struct inode *inode,
+                                  struct buffer_head *di_bh)
+{
+       int ret;
+       struct ocfs2_inode_info *oi = OCFS2_I(inode);
+       struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+
+       down_write(&oi->ip_xattr_sem);
+       down_write(&oi->ip_alloc_sem);
+
+       if (oi->ip_clusters)
+               goto out;
+
+       if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
+               goto out;
+
+       if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
+           ocfs2_has_inline_xattr_value_outside(inode, di))
+               goto out;
+
+       ret = ocfs2_remove_refcount_tree(inode, di_bh);
+       if (ret)
+               mlog_errno(ret);
+out:
+       up_write(&oi->ip_alloc_sem);
+       up_write(&oi->ip_xattr_sem);
+       return 0;
+}
+
+/*
+ * Find the end range for a leaf refcount block indicated by
+ * el->l_recs[index].e_blkno.
+ */
+static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
+                                      struct buffer_head *ref_root_bh,
+                                      struct ocfs2_extent_block *eb,
+                                      struct ocfs2_extent_list *el,
+                                      int index,  u32 *cpos_end)
+{
+       int ret, i, subtree_root;
+       u32 cpos;
+       u64 blkno;
+       struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+       struct ocfs2_path *left_path = NULL, *right_path = NULL;
+       struct ocfs2_extent_tree et;
+       struct ocfs2_extent_list *tmp_el;
+
+       if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
+               /*
+                * We have a extent rec after index, so just use the e_cpos
+                * of the next extent rec.
+                */
+               *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
+               return 0;
+       }
+
+       if (!eb || (eb && !eb->h_next_leaf_blk)) {
+               /*
+                * We are the last extent rec, so any high cpos should
+                * be stored in this leaf refcount block.
+                */
+               *cpos_end = UINT_MAX;
+               return 0;
+       }
+
+       /*
+        * If the extent block isn't the last one, we have to find
+        * the subtree root between this extent block and the next
+        * leaf extent block and get the corresponding e_cpos from
+        * the subroot. Otherwise we may corrupt the b-tree.
+        */
+       ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
+
+       left_path = ocfs2_new_path_from_et(&et);
+       if (!left_path) {
+               ret = -ENOMEM;
+               mlog_errno(ret);
+               goto out;
+       }
+
+       cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
+       ret = ocfs2_find_path(ci, left_path, cpos);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       right_path = ocfs2_new_path_from_path(left_path);
+       if (!right_path) {
+               ret = -ENOMEM;
+               mlog_errno(ret);
+               goto out;
+       }
+
+       ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       ret = ocfs2_find_path(ci, right_path, cpos);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       subtree_root = ocfs2_find_subtree_root(&et, left_path,
+                                              right_path);
+
+       tmp_el = left_path->p_node[subtree_root].el;
+       blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
+       for (i = 0; i < le32_to_cpu(tmp_el->l_next_free_rec); i++) {
+               if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
+                       *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
+                       break;
+               }
+       }
+
+       BUG_ON(i == le32_to_cpu(tmp_el->l_next_free_rec));
+
+out:
+       ocfs2_free_path(left_path);
+       ocfs2_free_path(right_path);
+       return ret;
+}
+
+/*
  * Given a cpos and len, try to find the refcount record which contains cpos.
  * 1. If cpos can be found in one refcount record, return the record.
  * 2. If cpos can't be found, return a fake record which start from cpos
@@ -941,10 +1080,10 @@ static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
                                  struct buffer_head **ret_bh)
 {
        int ret = 0, i, found;
-       u32 low_cpos;
+       u32 low_cpos, uninitialized_var(cpos_end);
        struct ocfs2_extent_list *el;
-       struct ocfs2_extent_rec *tmp, *rec = NULL;
-       struct ocfs2_extent_block *eb;
+       struct ocfs2_extent_rec *rec = NULL;
+       struct ocfs2_extent_block *eb = NULL;
        struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
        struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
        struct ocfs2_refcount_block *rb =
@@ -992,12 +1131,16 @@ static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
                }
        }
 
-       /* adjust len when we have ocfs2_extent_rec after it. */
-       if (found && i < le16_to_cpu(el->l_next_free_rec) - 1) {
-               tmp = &el->l_recs[i+1];
+       if (found) {
+               ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
+                                                 eb, el, i, &cpos_end);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
 
-               if (le32_to_cpu(tmp->e_cpos) < cpos + len)
-                       len = le32_to_cpu(tmp->e_cpos) - cpos;
+               if (cpos_end < low_cpos + len)
+                       len = cpos_end - low_cpos;
        }
 
        ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
@@ -1111,7 +1254,7 @@ static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
 static int ocfs2_change_refcount_rec(handle_t *handle,
                                     struct ocfs2_caching_info *ci,
                                     struct buffer_head *ref_leaf_bh,
-                                    int index, int change)
+                                    int index, int merge, int change)
 {
        int ret;
        struct ocfs2_refcount_block *rb =
@@ -1140,7 +1283,7 @@ static int ocfs2_change_refcount_rec(handle_t *handle,
                }
 
                le16_add_cpu(&rl->rl_used, -1);
-       } else
+       } else if (merge)
                ocfs2_refcount_rec_merge(rb, index);
 
        ret = ocfs2_journal_dirty(handle, ref_leaf_bh);
@@ -1616,7 +1759,7 @@ static int ocfs2_insert_refcount_rec(handle_t *handle,
                                     struct buffer_head *ref_root_bh,
                                     struct buffer_head *ref_leaf_bh,
                                     struct ocfs2_refcount_rec *rec,
-                                    int index,
+                                    int index, int merge,
                                     struct ocfs2_alloc_context *meta_ac)
 {
        int ret;
@@ -1674,7 +1817,8 @@ static int ocfs2_insert_refcount_rec(handle_t *handle,
 
        le16_add_cpu(&rf_list->rl_used, 1);
 
-       ocfs2_refcount_rec_merge(rb, index);
+       if (merge)
+               ocfs2_refcount_rec_merge(rb, index);
 
        ret = ocfs2_journal_dirty(handle, ref_leaf_bh);
        if (ret) {
@@ -1708,7 +1852,7 @@ static int ocfs2_split_refcount_rec(handle_t *handle,
                                    struct buffer_head *ref_root_bh,
                                    struct buffer_head *ref_leaf_bh,
                                    struct ocfs2_refcount_rec *split_rec,
-                                   int index,
+                                   int index, int merge,
                                    struct ocfs2_alloc_context *meta_ac,
                                    struct ocfs2_cached_dealloc_ctxt *dealloc)
 {
@@ -1846,7 +1990,8 @@ static int ocfs2_split_refcount_rec(handle_t *handle,
                     le32_to_cpu(split_rec->r_refcount),
                     (unsigned long long)ref_leaf_bh->b_blocknr, index);
 
-               ocfs2_refcount_rec_merge(rb, index);
+               if (merge)
+                       ocfs2_refcount_rec_merge(rb, index);
        }
 
        ret = ocfs2_journal_dirty(handle, ref_leaf_bh);
@@ -1861,7 +2006,7 @@ out:
 static int __ocfs2_increase_refcount(handle_t *handle,
                                     struct ocfs2_caching_info *ci,
                                     struct buffer_head *ref_root_bh,
-                                    u64 cpos, u32 len,
+                                    u64 cpos, u32 len, int merge,
                                     struct ocfs2_alloc_context *meta_ac,
                                     struct ocfs2_cached_dealloc_ctxt *dealloc)
 {
@@ -1901,7 +2046,8 @@ static int __ocfs2_increase_refcount(handle_t *handle,
                             "count %u\n", (unsigned long long)cpos, set_len,
                             le32_to_cpu(rec.r_refcount));
                        ret = ocfs2_change_refcount_rec(handle, ci,
-                                                       ref_leaf_bh, index, 1);
+                                                       ref_leaf_bh, index,
+                                                       merge, 1);
                        if (ret) {
                                mlog_errno(ret);
                                goto out;
@@ -1914,7 +2060,8 @@ static int __ocfs2_increase_refcount(handle_t *handle,
                             set_len);
                        ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
                                                        ref_leaf_bh,
-                                                       &rec, index, meta_ac);
+                                                       &rec, index,
+                                                       merge, meta_ac);
                        if (ret) {
                                mlog_errno(ret);
                                goto out;
@@ -1932,7 +2079,7 @@ static int __ocfs2_increase_refcount(handle_t *handle,
                             set_len, le32_to_cpu(rec.r_refcount));
                        ret = ocfs2_split_refcount_rec(handle, ci,
                                                       ref_root_bh, ref_leaf_bh,
-                                                      &rec, index,
+                                                      &rec, index, merge,
                                                       meta_ac, dealloc);
                        if (ret) {
                                mlog_errno(ret);
@@ -2025,6 +2172,18 @@ out:
        return ret;
 }
 
+int ocfs2_increase_refcount(handle_t *handle,
+                           struct ocfs2_caching_info *ci,
+                           struct buffer_head *ref_root_bh,
+                           u64 cpos, u32 len,
+                           struct ocfs2_alloc_context *meta_ac,
+                           struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+       return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
+                                        cpos, len, 1,
+                                        meta_ac, dealloc);
+}
+
 static int ocfs2_decrease_refcount_rec(handle_t *handle,
                                struct ocfs2_caching_info *ci,
                                struct buffer_head *ref_root_bh,
@@ -2045,7 +2204,7 @@ static int ocfs2_decrease_refcount_rec(handle_t *handle,
        if (cpos == le64_to_cpu(rec->r_cpos) &&
            len == le32_to_cpu(rec->r_clusters))
                ret = ocfs2_change_refcount_rec(handle, ci,
-                                               ref_leaf_bh, index, -1);
+                                               ref_leaf_bh, index, 1, -1);
        else {
                struct ocfs2_refcount_rec split = *rec;
                split.r_cpos = cpu_to_le64(cpos);
@@ -2061,7 +2220,7 @@ static int ocfs2_decrease_refcount_rec(handle_t *handle,
                     le32_to_cpu(rec->r_clusters));
                ret = ocfs2_split_refcount_rec(handle, ci,
                                               ref_root_bh, ref_leaf_bh,
-                                              &split, index,
+                                              &split, index, 1,
                                               meta_ac, dealloc);
        }
 
@@ -3547,7 +3706,8 @@ int ocfs2_add_refcount_flag(struct inode *inode,
                            struct ocfs2_caching_info *ref_ci,
                            struct buffer_head *ref_root_bh,
                            u32 cpos, u32 p_cluster, u32 num_clusters,
-                           struct ocfs2_cached_dealloc_ctxt *dealloc)
+                           struct ocfs2_cached_dealloc_ctxt *dealloc,
+                           struct ocfs2_post_refcount *post)
 {
        int ret;
        handle_t *handle;
@@ -3576,6 +3736,9 @@ int ocfs2_add_refcount_flag(struct inode *inode,
                }
        }
 
+       if (post)
+               credits += post->credits;
+
        handle = ocfs2_start_trans(osb, credits);
        if (IS_ERR(handle)) {
                ret = PTR_ERR(handle);
@@ -3592,10 +3755,18 @@ int ocfs2_add_refcount_flag(struct inode *inode,
        }
 
        ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
-                                       p_cluster, num_clusters,
+                                       p_cluster, num_clusters, 0,
                                        meta_ac, dealloc);
-       if (ret)
+       if (ret) {
                mlog_errno(ret);
+               goto out_commit;
+       }
+
+       if (post && post->func) {
+               ret = post->func(inode, handle, post->para);
+               if (ret)
+                       mlog_errno(ret);
+       }
 
 out_commit:
        ocfs2_commit_trans(osb, handle);
@@ -3673,6 +3844,9 @@ static int ocfs2_attach_refcount_tree(struct inode *inode,
                goto out;
        }
 
+       if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+               goto attach_xattr;
+
        ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
 
        size = i_size_read(inode);
@@ -3688,7 +3862,7 @@ static int ocfs2_attach_refcount_tree(struct inode *inode,
                                                      &ref_tree->rf_ci,
                                                      ref_root_bh, cpos,
                                                      p_cluster, num_clusters,
-                                                     &dealloc);
+                                                     &dealloc, NULL);
                        if (ret) {
                                mlog_errno(ret);
                                goto unlock;
@@ -3699,6 +3873,18 @@ static int ocfs2_attach_refcount_tree(struct inode *inode,
                cpos += num_clusters;
        }
 
+attach_xattr:
+       if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
+               ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
+                                                      &ref_tree->rf_ci,
+                                                      ref_root_bh,
+                                                      &dealloc);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto unlock;
+               }
+       }
+
        if (data_changed) {
                ret = ocfs2_change_ctime(inode, di_bh);
                if (ret)
@@ -3763,9 +3949,9 @@ static int ocfs2_add_refcounted_extent(struct inode *inode,
                goto out_commit;
        }
 
-       ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
-                                       p_cluster, num_clusters,
-                                       meta_ac, dealloc);
+       ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
+                                     p_cluster, num_clusters,
+                                     meta_ac, dealloc);
        if (ret)
                mlog_errno(ret);
 
@@ -3777,6 +3963,49 @@ out:
        return ret;
 }
 
+static int ocfs2_duplicate_inline_data(struct inode *s_inode,
+                                      struct buffer_head *s_bh,
+                                      struct inode *t_inode,
+                                      struct buffer_head *t_bh)
+{
+       int ret;
+       handle_t *handle;
+       struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
+       struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
+       struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
+
+       BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
+
+       handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+       if (IS_ERR(handle)) {
+               ret = PTR_ERR(handle);
+               mlog_errno(ret);
+               goto out;
+       }
+
+       ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
+                                     OCFS2_JOURNAL_ACCESS_WRITE);
+       if (ret) {
+               mlog_errno(ret);
+               goto out_commit;
+       }
+
+       t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
+       memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
+              le16_to_cpu(s_di->id2.i_data.id_count));
+       spin_lock(&OCFS2_I(t_inode)->ip_lock);
+       OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
+       t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
+       spin_unlock(&OCFS2_I(t_inode)->ip_lock);
+
+       ocfs2_journal_dirty(handle, t_bh);
+
+out_commit:
+       ocfs2_commit_trans(osb, handle);
+out:
+       return ret;
+}
+
 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
                                struct inode *t_inode,
                                struct buffer_head *t_bh,
@@ -3829,7 +4058,8 @@ out:
 static int ocfs2_complete_reflink(struct inode *s_inode,
                                  struct buffer_head *s_bh,
                                  struct inode *t_inode,
-                                 struct buffer_head *t_bh)
+                                 struct buffer_head *t_bh,
+                                 bool preserve)
 {
        int ret;
        handle_t *handle;
@@ -3864,22 +4094,26 @@ static int ocfs2_complete_reflink(struct inode *s_inode,
        di->i_size = s_di->i_size;
        di->i_dyn_features = s_di->i_dyn_features;
        di->i_attr = s_di->i_attr;
-       di->i_uid = s_di->i_uid;
-       di->i_gid = s_di->i_gid;
-       di->i_mode = s_di->i_mode;
 
-       /*
-        * update time.
-        * we want mtime to appear identical to the source and update ctime.
-        */
-       t_inode->i_ctime = CURRENT_TIME;
+       if (preserve) {
+               di->i_uid = s_di->i_uid;
+               di->i_gid = s_di->i_gid;
+               di->i_mode = s_di->i_mode;
 
-       di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
-       di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
+               /*
+                * update time.
+                * we want mtime to appear identical to the source and
+                * update ctime.
+                */
+               t_inode->i_ctime = CURRENT_TIME;
 
-       t_inode->i_mtime = s_inode->i_mtime;
-       di->i_mtime = s_di->i_mtime;
-       di->i_mtime_nsec = s_di->i_mtime_nsec;
+               di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
+               di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
+
+               t_inode->i_mtime = s_inode->i_mtime;
+               di->i_mtime = s_di->i_mtime;
+               di->i_mtime_nsec = s_di->i_mtime_nsec;
+       }
 
        ocfs2_journal_dirty(handle, t_bh);
 
@@ -3891,7 +4125,8 @@ out_commit:
 static int ocfs2_create_reflink_node(struct inode *s_inode,
                                     struct buffer_head *s_bh,
                                     struct inode *t_inode,
-                                    struct buffer_head *t_bh)
+                                    struct buffer_head *t_bh,
+                                    bool preserve)
 {
        int ret;
        struct buffer_head *ref_root_bh = NULL;
@@ -3910,6 +4145,14 @@ static int ocfs2_create_reflink_node(struct inode *s_inode,
                goto out;
        }
 
+       if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+               ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
+                                                 t_inode, t_bh);
+               if (ret)
+                       mlog_errno(ret);
+               goto out;
+       }
+
        ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
                                       1, &ref_tree, &ref_root_bh);
        if (ret) {
@@ -3926,10 +4169,6 @@ static int ocfs2_create_reflink_node(struct inode *s_inode,
                goto out_unlock_refcount;
        }
 
-       ret = ocfs2_complete_reflink(s_inode, s_bh, t_inode, t_bh);
-       if (ret)
-               mlog_errno(ret);
-
 out_unlock_refcount:
        ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
        brelse(ref_root_bh);
@@ -3941,3 +4180,294 @@ out:
 
        return ret;
 }
+
+static int __ocfs2_reflink(struct dentry *old_dentry,
+                          struct buffer_head *old_bh,
+                          struct inode *new_inode,
+                          bool preserve)
+{
+       int ret;
+       struct inode *inode = old_dentry->d_inode;
+       struct buffer_head *new_bh = NULL;
+
+       ret = filemap_fdatawrite(inode->i_mapping);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       ret = ocfs2_attach_refcount_tree(inode, old_bh);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       mutex_lock(&new_inode->i_mutex);
+       ret = ocfs2_inode_lock(new_inode, &new_bh, 1);
+       if (ret) {
+               mlog_errno(ret);
+               goto out_unlock;
+       }
+
+       ret = ocfs2_create_reflink_node(inode, old_bh,
+                                       new_inode, new_bh, preserve);
+       if (ret) {
+               mlog_errno(ret);
+               goto inode_unlock;
+       }
+
+       if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
+               ret = ocfs2_reflink_xattrs(inode, old_bh,
+                                          new_inode, new_bh,
+                                          preserve);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto inode_unlock;
+               }
+       }
+
+       ret = ocfs2_complete_reflink(inode, old_bh,
+                                    new_inode, new_bh, preserve);
+       if (ret)
+               mlog_errno(ret);
+
+inode_unlock:
+       ocfs2_inode_unlock(new_inode, 1);
+       brelse(new_bh);
+out_unlock:
+       mutex_unlock(&new_inode->i_mutex);
+out:
+       if (!ret) {
+               ret = filemap_fdatawait(inode->i_mapping);
+               if (ret)
+                       mlog_errno(ret);
+       }
+       return ret;
+}
+
+static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
+                        struct dentry *new_dentry, bool preserve)
+{
+       int error;
+       struct inode *inode = old_dentry->d_inode;
+       struct buffer_head *old_bh = NULL;
+       struct inode *new_orphan_inode = NULL;
+
+       if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
+               return -EOPNOTSUPP;
+
+       error = ocfs2_create_inode_in_orphan(dir, inode->i_mode,
+                                            &new_orphan_inode);
+       if (error) {
+               mlog_errno(error);
+               goto out;
+       }
+
+       error = ocfs2_inode_lock(inode, &old_bh, 1);
+       if (error) {
+               mlog_errno(error);
+               goto out;
+       }
+
+       down_write(&OCFS2_I(inode)->ip_xattr_sem);
+       down_write(&OCFS2_I(inode)->ip_alloc_sem);
+       error = __ocfs2_reflink(old_dentry, old_bh,
+                               new_orphan_inode, preserve);
+       up_write(&OCFS2_I(inode)->ip_alloc_sem);
+       up_write(&OCFS2_I(inode)->ip_xattr_sem);
+
+       ocfs2_inode_unlock(inode, 1);
+       brelse(old_bh);
+
+       if (error) {
+               mlog_errno(error);
+               goto out;
+       }
+
+       /* If the security isn't preserved, we need to re-initialize them. */
+       if (!preserve) {
+               error = ocfs2_init_security_and_acl(dir, new_orphan_inode);
+               if (error)
+                       mlog_errno(error);
+       }
+out:
+       if (!error) {
+               error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
+                                                      new_dentry);
+               if (error)
+                       mlog_errno(error);
+       }
+
+       if (new_orphan_inode) {
+               /*
+                * We need to open_unlock the inode no matter whether we
+                * succeed or not, so that other nodes can delete it later.
+                */
+               ocfs2_open_unlock(new_orphan_inode);
+               if (error)
+                       iput(new_orphan_inode);
+       }
+
+       return error;
+}
+
+/*
+ * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
+ * sys_reflink().  This will go away when vfs_reflink() exists in
+ * fs/namei.c.
+ */
+
+/* copied from may_create in VFS. */
+static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
+{
+       if (child->d_inode)
+               return -EEXIST;
+       if (IS_DEADDIR(dir))
+               return -ENOENT;
+       return inode_permission(dir, MAY_WRITE | MAY_EXEC);
+}
+
+/* copied from user_path_parent. */
+static int ocfs2_user_path_parent(const char __user *path,
+                                 struct nameidata *nd, char **name)
+{
+       char *s = getname(path);
+       int error;
+
+       if (IS_ERR(s))
+               return PTR_ERR(s);
+
+       error = path_lookup(s, LOOKUP_PARENT, nd);
+       if (error)
+               putname(s);
+       else
+               *name = s;
+
+       return error;
+}
+
+/**
+ * ocfs2_vfs_reflink - Create a reference-counted link
+ *
+ * @old_dentry:        source dentry + inode
+ * @dir:       directory to create the target
+ * @new_dentry:        target dentry
+ * @preserve:  if true, preserve all file attributes
+ */
+int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
+                     struct dentry *new_dentry, bool preserve)
+{
+       struct inode *inode = old_dentry->d_inode;
+       int error;
+
+       if (!inode)
+               return -ENOENT;
+
+       error = ocfs2_may_create(dir, new_dentry);
+       if (error)
+               return error;
+
+       if (dir->i_sb != inode->i_sb)
+               return -EXDEV;
+
+       /*
+        * A reflink to an append-only or immutable file cannot be created.
+        */
+       if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
+               return -EPERM;
+
+       /* Only regular files can be reflinked. */
+       if (!S_ISREG(inode->i_mode))
+               return -EPERM;
+
+       /*
+        * If the caller wants to preserve ownership, they require the
+        * rights to do so.
+        */
+       if (preserve) {
+               if ((current_fsuid() != inode->i_uid) && !capable(CAP_CHOWN))
+                       return -EPERM;
+               if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
+                       return -EPERM;
+       }
+
+       /*
+        * If the caller is modifying any aspect of the attributes, they
+        * are not creating a snapshot.  They need read permission on the
+        * file.
+        */
+       if (!preserve) {
+               error = inode_permission(inode, MAY_READ);
+               if (error)
+                       return error;
+       }
+
+       mutex_lock(&inode->i_mutex);
+       vfs_dq_init(dir);
+       error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
+       mutex_unlock(&inode->i_mutex);
+       if (!error)
+               fsnotify_create(dir, new_dentry);
+       return error;
+}
+/*
+ * Most codes are copied from sys_linkat.
+ */
+int ocfs2_reflink_ioctl(struct inode *inode,
+                       const char __user *oldname,
+                       const char __user *newname,
+                       bool preserve)
+{
+       struct dentry *new_dentry;
+       struct nameidata nd;
+       struct path old_path;
+       int error;
+       char *to = NULL;
+
+       if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
+               return -EOPNOTSUPP;
+
+       error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
+       if (error) {
+               mlog_errno(error);
+               return error;
+       }
+
+       error = ocfs2_user_path_parent(newname, &nd, &to);
+       if (error) {
+               mlog_errno(error);
+               goto out;
+       }
+
+       error = -EXDEV;
+       if (old_path.mnt != nd.path.mnt)
+               goto out_release;
+       new_dentry = lookup_create(&nd, 0);
+       error = PTR_ERR(new_dentry);
+       if (IS_ERR(new_dentry)) {
+               mlog_errno(error);
+               goto out_unlock;
+       }
+
+       error = mnt_want_write(nd.path.mnt);
+       if (error) {
+               mlog_errno(error);
+               goto out_dput;
+       }
+
+       error = ocfs2_vfs_reflink(old_path.dentry,
+                                 nd.path.dentry->d_inode,
+                                 new_dentry, preserve);
+       mnt_drop_write(nd.path.mnt);
+out_dput:
+       dput(new_dentry);
+out_unlock:
+       mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
+out_release:
+       path_put(&nd.path);
+       putname(to);
+out:
+       path_put(&old_path);
+
+       return error;
+}