[Ocfs2-devel] [PATCH 11/39] ocfs2: Add support of decrementing refcount for delete.
Tao Ma
tao.ma at oracle.com
Wed Apr 29 15:58:23 PDT 2009
Given a physical cpos and length, decrement the refcount
in the tree. If the refcount for any portion of the extent goes
to zero, that portion is queued for freeing.
Signed-off-by: Tao Ma <tao.ma at oracle.com>
---
fs/ocfs2/refcounttree.c | 227 ++++++++++++++++++++++++++++++++++++++++++++++-
fs/ocfs2/refcounttree.h | 4 +
2 files changed, 229 insertions(+), 2 deletions(-)
diff --git a/fs/ocfs2/refcounttree.c b/fs/ocfs2/refcounttree.c
index bc4bfb8..58b2c12 100644
--- a/fs/ocfs2/refcounttree.c
+++ b/fs/ocfs2/refcounttree.c
@@ -603,6 +603,10 @@ static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
ocfs2_rotate_refcount_rec_left(rb, index);
}
+/*
+ * Change the refcount indexed by "index" in ref_bh.
+ * If refcount reaches 0, remove it.
+ */
static int ocfs2_change_refcount_rec(handle_t *handle,
struct ocfs2_caching_info *ci,
struct buffer_head *ref_leaf_bh,
@@ -611,7 +615,8 @@ static int ocfs2_change_refcount_rec(handle_t *handle,
int ret;
struct ocfs2_refcount_block *rb =
(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
- struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
+ struct ocfs2_refcount_list *rl = &rb->rf_records;
+ struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
OCFS2_JOURNAL_ACCESS_WRITE);
@@ -624,7 +629,18 @@ static int ocfs2_change_refcount_rec(handle_t *handle,
le32_to_cpu(rec->r_refcount), change);
le32_add_cpu(&rec->r_refcount, change);
- ocfs2_refcount_rec_merge(rb, index);
+ if (!rec->r_refcount) {
+ if (index != le16_to_cpu(rl->rl_used) - 1) {
+ memmove(rec, rec + 1,
+ (le16_to_cpu(rl->rl_used) - index - 1) *
+ sizeof(struct ocfs2_refcount_rec));
+ memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
+ 0, sizeof(struct ocfs2_refcount_rec));
+ }
+
+ le16_add_cpu(&rl->rl_used, -1);
+ } else
+ ocfs2_refcount_rec_merge(rb, index);
ret = ocfs2_journal_dirty(handle, ref_leaf_bh);
if (ret)
@@ -1288,3 +1304,210 @@ out:
brelse(ref_leaf_bh);
return ret;
}
+
+static int ocfs2_remove_refcount_extent(handle_t *handle,
+ struct ocfs2_caching_info *ci,
+ struct buffer_head *ref_root_bh,
+ struct buffer_head *ref_leaf_bh,
+ struct ocfs2_alloc_context *meta_ac,
+ struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+ int ret;
+ struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+ struct ocfs2_refcount_block *rb =
+ (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+ struct ocfs2_extent_tree et;
+
+ BUG_ON(rb->rf_records.rl_used);
+
+ ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
+ ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
+ 1, meta_ac, dealloc);
+ if (ret) {
+ mlog_errno(ret);
+ goto out;
+ }
+
+ ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
+ OCFS2_JOURNAL_ACCESS_WRITE);
+ if (ret) {
+ mlog_errno(ret);
+ goto out;
+ }
+
+ rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+
+ le32_add_cpu(&rb->rf_clusters, -1);
+
+ /*
+ * check whether we need to restore the root refcount block if
+ * there is no leaf extent block at atll.
+ */
+ if (!rb->rf_list.l_next_free_rec) {
+ BUG_ON(rb->rf_clusters);
+
+ mlog(0, "reset refcount block root %llu to a leaf block.\n",
+ (unsigned long long)ref_root_bh->b_blocknr);
+
+ memset(&rb->rf_records, 0, sb->s_blocksize -
+ offsetof(struct ocfs2_refcount_block, rf_records));
+ rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
+ rb->rf_records.rl_count =
+ cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
+ }
+
+ ocfs2_journal_dirty(handle, ref_root_bh);
+
+out:
+ return ret;
+}
+
+static int ocfs2_decrease_refcount_rec(handle_t *handle,
+ struct ocfs2_caching_info *ci,
+ struct buffer_head *ref_root_bh,
+ struct buffer_head *ref_leaf_bh,
+ int index, u64 cpos, unsigned int len,
+ struct ocfs2_alloc_context *meta_ac,
+ struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+ int ret;
+ struct ocfs2_refcount_block *rb =
+ (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+ struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
+
+ BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
+ BUG_ON(cpos + len >
+ le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
+
+ if (cpos == le64_to_cpu(rec->r_cpos) && cpos + len ==
+ le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters))
+ ret = ocfs2_change_refcount_rec(handle, ci,
+ ref_leaf_bh, index, -1);
+ else {
+ struct ocfs2_refcount_rec split = *rec;
+ split.r_cpos = cpu_to_le64(cpos);
+ split.r_clusters = cpu_to_le32(len);
+
+ le32_add_cpu(&split.r_refcount, -1);
+
+ mlog(0, "split refcount rec, start %llu, "
+ "len %u, count %u, original start %llu, len %u\n",
+ (unsigned long long)le64_to_cpu(split.r_cpos),
+ len, le32_to_cpu(split.r_refcount),
+ (unsigned long long)le64_to_cpu(rec->r_cpos),
+ le32_to_cpu(rec->r_clusters));
+ ret = ocfs2_split_refcount_rec(handle, ci,
+ ref_root_bh, ref_leaf_bh,
+ &split, index,
+ meta_ac, dealloc);
+ }
+
+ if (ret) {
+ mlog_errno(ret);
+ goto out;
+ }
+
+ /* Remove the leaf refcount block if it contains no refcount record. */
+ if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
+ ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
+ ref_leaf_bh, meta_ac,
+ dealloc);
+ if (ret)
+ mlog_errno(ret);
+ }
+
+out:
+ return ret;
+}
+
+static int __ocfs2_decrease_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)
+{
+ int ret = 0, index = 0;
+ struct ocfs2_refcount_rec rec;
+ unsigned int r_count = 0, r_len;
+ struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+ struct buffer_head *ref_leaf_bh = NULL;
+
+ mlog(0, "Tree owner %llu, decrease refcount start %llu, len %u\n",
+ (unsigned long long)ocfs2_metadata_cache_owner(ci),
+ (unsigned long long)cpos, len);
+
+ while (len) {
+ ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
+ cpos, len, &rec, &index,
+ &ref_leaf_bh);
+ if (ret) {
+ mlog_errno(ret);
+ goto out;
+ }
+
+ r_count = le32_to_cpu(rec.r_refcount);
+ BUG_ON(r_count == 0);
+
+ r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
+ le32_to_cpu(rec.r_clusters)) - cpos;
+
+ ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
+ ref_leaf_bh, index,
+ cpos, r_len,
+ meta_ac, dealloc);
+ if (ret) {
+ mlog_errno(ret);
+ goto out;
+ }
+
+ if (le32_to_cpu(rec.r_refcount) == 1) {
+ ret = ocfs2_cache_cluster_dealloc(dealloc,
+ ocfs2_clusters_to_blocks(sb, cpos),
+ r_len);
+ if (ret) {
+ mlog_errno(ret);
+ goto out;
+ }
+ }
+
+ cpos += r_len;
+ len -= r_len;
+ brelse(ref_leaf_bh);
+ ref_leaf_bh = NULL;
+ }
+
+out:
+ brelse(ref_leaf_bh);
+ return ret;
+}
+
+int ocfs2_decrease_refcount(struct inode *inode, struct buffer_head *di_bh,
+ handle_t *handle, u32 cpos, u32 len,
+ struct ocfs2_alloc_context *meta_ac,
+ struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+ int ret;
+ struct ocfs2_inode_info *oi = OCFS2_I(inode);
+ struct buffer_head *ref_root_bh = NULL;
+ struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+
+ BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+ BUG_ON(!di->i_refcount_loc);
+
+ ret = ocfs2_read_refcount_block(INODE_CACHE(inode),
+ le64_to_cpu(di->i_refcount_loc),
+ &ref_root_bh);
+ if (ret) {
+ mlog_errno(ret);
+ goto out;
+ }
+
+ ret = __ocfs2_decrease_refcount(handle, INODE_CACHE(inode), ref_root_bh,
+ cpos, len, meta_ac, dealloc);
+ if (ret)
+ mlog_errno(ret);
+out:
+ brelse(ref_root_bh);
+ return ret;
+}
diff --git a/fs/ocfs2/refcounttree.h b/fs/ocfs2/refcounttree.h
index 9f4bdac..92fe116 100644
--- a/fs/ocfs2/refcounttree.h
+++ b/fs/ocfs2/refcounttree.h
@@ -23,4 +23,8 @@ int ocfs2_set_refcount_tree(struct inode *inode,
u64 blkno);
int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh);
+int ocfs2_decrease_refcount(struct inode *inode, struct buffer_head *di_bh,
+ handle_t *handle, u32 cpos, u32 len,
+ struct ocfs2_alloc_context *meta_ac,
+ struct ocfs2_cached_dealloc_ctxt *dealloc);
#endif /* OCFS2_REFCOUNTTREE_H */
--
1.6.2.rc2.16.gf474c
More information about the Ocfs2-devel
mailing list