On 2020/1/6 23:35, Jeff Layton wrote:
Currently, we just assume that it will stick around by virtue of the
submitter's reference, but later patches will allow the syscall to
return early and we can't rely on that reference at that point.
Take an extra reference to the inode when setting r_parent and release
it when releasing the request.
Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx>
---
fs/ceph/mds_client.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 94cce2ab92c4..b7122f682678 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -708,8 +708,10 @@ void ceph_mdsc_release_request(struct kref *kref)
/* avoid calling iput_final() in mds dispatch threads */
ceph_async_iput(req->r_inode);
}
- if (req->r_parent)
+ if (req->r_parent) {
ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
+ ceph_async_iput(req->r_parent);
+ }
ceph_async_iput(req->r_target_inode);
if (req->r_dentry)
dput(req->r_dentry);
@@ -2706,8 +2708,10 @@ int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
/* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
if (req->r_inode)
ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
- if (req->r_parent)
+ if (req->r_parent) {
ceph_get_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
+ ihold(req->r_parent);
+ }
This might also fix another issue when the mdsc request is timedout and
returns to the vfs, then the r_parent maybe released in vfs. And then if
we reference it again in mdsc handle_reply() -->
ceph_mdsc_release_request(), some unknown issues may happen later ??
if (req->r_old_dentry_dir)
ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
CEPH_CAP_PIN);
