From 570df4e9c23f861aa3f8f2954468c534a033bf1a Mon Sep 17 00:00:00 2001 From: "Yan, Zheng" Date: Wed, 15 Nov 2017 17:39:40 +0800 Subject: ceph: snapshot nfs re-export To support snapshot nfs re-export, we need a way to lookup snapped inode by file handle. For directory inode, snapped metadata are always stored together with head inode. Client just need to pass vinodeno_t to MDS. For non-directory inode, there can be multiple version of snapped inodes and they can be stored in different dirfrags. Besides vinodeno_t, client also need to tell mds from which dirfrag it got the snapped inode. Another problem of supporting snapshot nfs re-export is that there can be multiple paths to access a snapped inode. For example: mkdir -p d1/d2/d3 mkdir d1/.snap/s1 Paths 'd1/.snap/s1/d2/d3', 'd1/d2/.snap/_s1_/d3' and 'd1/d2/d3/.snap/_s1_' are all reference to the same snapped inode. For a given snapped inode, There is no convenient way to get the first form and the second form paths. For simplicity, ceph_get_parent() return snapdir for snapped directory inode. Furthermore, client may access snapshot of deleted directory. For example: mkdir -p d1/d2 mkdir d1/.snap/s1 open d1/.snap/s1/d2 rm -rf d1/d2 The path constucted by ceph_get_parent() and ceph_get_name() is '/.snap/_s1_'. Futher lookup parent of will fail. To workaround this case, this patch uses d_obtain_root() to get dentry for snapdir of deleted directory. snapdir dentry has no DCACHE_DISCONNECTED flag set, reconnect_path() stops when it reaches snapdir dentry. Link: http://tracker.ceph.com/issues/22105 Signed-off-by: "Yan, Zheng" Signed-off-by: Ilya Dryomov --- include/linux/ceph/ceph_fs.h | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'include/linux') diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h index 4903deb0777a..3ac0feaf2b5e 100644 --- a/include/linux/ceph/ceph_fs.h +++ b/include/linux/ceph/ceph_fs.h @@ -436,6 +436,12 @@ union ceph_mds_request_args { __le64 length; /* num bytes to lock from start */ __u8 wait; /* will caller wait for lock to become available? */ } __attribute__ ((packed)) filelock_change; + struct { + __le32 mask; /* CEPH_CAP_* */ + __le64 snapid; + __le64 parent; + __le32 hash; + } __attribute__ ((packed)) lookupino; } __attribute__ ((packed)); #define CEPH_MDS_FLAG_REPLAY 1 /* this is a replayed op */ -- cgit