ocfs2: only take lock if dio entry when recover orphans
authorJoseph Qi <joseph.qi@huawei.com>
Fri, 6 Nov 2015 02:44:04 +0000 (18:44 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 6 Nov 2015 03:34:48 +0000 (19:34 -0800)
We have no need to take inode mutex, rw and inode lock if it is not dio
entry when recover orphans.  Optimize it by adding a flag
OCFS2_INODE_DIO_ORPHAN_ENTRY to ocfs2_inode_info to reduce contention.

Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/ocfs2/inode.h
fs/ocfs2/journal.c

index ca3431ee7f2493fb999cdb6ced2908bdb87fbe17..aac8b86f312e41009ed673f2725b276691c66961 100644 (file)
@@ -112,6 +112,8 @@ struct ocfs2_inode_info
 #define OCFS2_INODE_OPEN_DIRECT                0x00000020
 /* Tell the inode wipe code it's not in orphan dir */
 #define OCFS2_INODE_SKIP_ORPHAN_DIR     0x00000040
+/* Entry in orphan dir with 'dio-' prefix */
+#define OCFS2_INODE_DIO_ORPHAN_ENTRY   0x00000080
 
 static inline struct ocfs2_inode_info *OCFS2_I(struct inode *inode)
 {
index 4bac6007837dbbb1b1e094e6fcbc22ba432b7258..04d6303f39a74b1768d812a9fdc7952722a49df5 100644 (file)
@@ -2049,6 +2049,10 @@ static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name,
        if (IS_ERR(iter))
                return 0;
 
+       if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
+                       OCFS2_DIO_ORPHAN_PREFIX_LEN))
+               OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY;
+
        /* Skip inodes which are already added to recover list, since dio may
         * happen concurrently with unlink/rename */
        if (OCFS2_I(iter)->ip_next_orphan) {
@@ -2195,25 +2199,51 @@ static int ocfs2_recover_orphans(struct ocfs2_super *osb,
                iter = oi->ip_next_orphan;
                oi->ip_next_orphan = NULL;
 
-               mutex_lock(&inode->i_mutex);
-               ret = ocfs2_rw_lock(inode, 1);
-               if (ret < 0) {
-                       mlog_errno(ret);
-                       goto next;
-               }
-               /*
-                * We need to take and drop the inode lock to
-                * force read inode from disk.
-                */
-               ret = ocfs2_inode_lock(inode, &di_bh, 1);
-               if (ret) {
-                       mlog_errno(ret);
-                       goto unlock_rw;
-               }
+               if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) {
+                       mutex_lock(&inode->i_mutex);
+                       ret = ocfs2_rw_lock(inode, 1);
+                       if (ret < 0) {
+                               mlog_errno(ret);
+                               goto unlock_mutex;
+                       }
+                       /*
+                        * We need to take and drop the inode lock to
+                        * force read inode from disk.
+                        */
+                       ret = ocfs2_inode_lock(inode, &di_bh, 1);
+                       if (ret) {
+                               mlog_errno(ret);
+                               goto unlock_rw;
+                       }
 
-               di = (struct ocfs2_dinode *)di_bh->b_data;
+                       di = (struct ocfs2_dinode *)di_bh->b_data;
 
-               if (inode->i_nlink == 0) {
+                       if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) {
+                               ret = ocfs2_truncate_file(inode, di_bh,
+                                               i_size_read(inode));
+                               if (ret < 0) {
+                                       if (ret != -ENOSPC)
+                                               mlog_errno(ret);
+                                       goto unlock_inode;
+                               }
+
+                               ret = ocfs2_del_inode_from_orphan(osb, inode,
+                                               di_bh, 0, 0);
+                               if (ret)
+                                       mlog_errno(ret);
+                       }
+unlock_inode:
+                       ocfs2_inode_unlock(inode, 1);
+                       brelse(di_bh);
+                       di_bh = NULL;
+unlock_rw:
+                       ocfs2_rw_unlock(inode, 1);
+unlock_mutex:
+                       mutex_unlock(&inode->i_mutex);
+
+                       /* clear dio flag in ocfs2_inode_info */
+                       oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY;
+               } else {
                        spin_lock(&oi->ip_lock);
                        /* Set the proper information to get us going into
                         * ocfs2_delete_inode. */
@@ -2221,28 +2251,6 @@ static int ocfs2_recover_orphans(struct ocfs2_super *osb,
                        spin_unlock(&oi->ip_lock);
                }
 
-               if ((orphan_reco_type == ORPHAN_NEED_TRUNCATE) &&
-                               (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
-                       ret = ocfs2_truncate_file(inode, di_bh,
-                                       i_size_read(inode));
-                       if (ret < 0) {
-                               if (ret != -ENOSPC)
-                                       mlog_errno(ret);
-                               goto unlock_inode;
-                       }
-
-                       ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
-                       if (ret)
-                               mlog_errno(ret);
-               } /* else if ORPHAN_NO_NEED_TRUNCATE, do nothing */
-unlock_inode:
-               ocfs2_inode_unlock(inode, 1);
-               brelse(di_bh);
-               di_bh = NULL;
-unlock_rw:
-               ocfs2_rw_unlock(inode, 1);
-next:
-               mutex_unlock(&inode->i_mutex);
                iput(inode);
                inode = iter;
        }