Merge tag 'ceph-for-6.4-rc6' of https://github.com/ceph/ceph-client

Pull ceph fixes from Ilya Dryomov:
 "A fix for a potential data corruption in differential backup and
  snapshot-based mirroring scenarios in RBD and a reference counting
  fixup to avoid use-after-free in CephFS, all marked for stable"

* tag 'ceph-for-6.4-rc6' of https://github.com/ceph/ceph-client:
  ceph: fix use-after-free bug for inodes when flushing capsnaps
  rbd: get snapshot context after exclusive lock is ensured to be held
  rbd: move RBD_OBJ_FLAG_COPYUP_ENABLED flag setting

3 files changed, 53 insertions, 19 deletions

diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c
index 84ad3b17956f67..632751ddb28708 100644
--- a/drivers/block/rbd.c
+++ b/drivers/block/rbd.c
@@ -1334,14 +1334,30 @@ static bool rbd_obj_is_tail(struct rbd_obj_request *obj_req)
 /*
  * Must be called after rbd_obj_calc_img_extents().
  */
-static bool rbd_obj_copyup_enabled(struct rbd_obj_request *obj_req)
+static void rbd_obj_set_copyup_enabled(struct rbd_obj_request *obj_req)
 {
-	if (!obj_req->num_img_extents ||
-	    (rbd_obj_is_entire(obj_req) &&
-	     !obj_req->img_request->snapc->num_snaps))
-		return false;
+	rbd_assert(obj_req->img_request->snapc);
 
-	return true;
+	if (obj_req->img_request->op_type == OBJ_OP_DISCARD) {
+		dout("%s %p objno %llu discard\n", __func__, obj_req,
+		     obj_req->ex.oe_objno);
+		return;
+	}
+
+	if (!obj_req->num_img_extents) {
+		dout("%s %p objno %llu not overlapping\n", __func__, obj_req,
+		     obj_req->ex.oe_objno);
+		return;
+	}
+
+	if (rbd_obj_is_entire(obj_req) &&
+	    !obj_req->img_request->snapc->num_snaps) {
+		dout("%s %p objno %llu entire\n", __func__, obj_req,
+		     obj_req->ex.oe_objno);
+		return;
+	}
+
+	obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
 }
 
 static u64 rbd_obj_img_extents_bytes(struct rbd_obj_request *obj_req)
@@ -1442,6 +1458,7 @@ __rbd_obj_add_osd_request(struct rbd_obj_request *obj_req,
 static struct ceph_osd_request *
 rbd_obj_add_osd_request(struct rbd_obj_request *obj_req, int num_ops)
 {
+	rbd_assert(obj_req->img_request->snapc);
 	return __rbd_obj_add_osd_request(obj_req, obj_req->img_request->snapc,
 					 num_ops);
 }
@@ -1578,15 +1595,18 @@ static void rbd_img_request_init(struct rbd_img_request *img_request,
 	mutex_init(&img_request->state_mutex);
 }
 
+/*
+ * Only snap_id is captured here, for reads.  For writes, snapshot
+ * context is captured in rbd_img_object_requests() after exclusive
+ * lock is ensured to be held.
+ */
 static void rbd_img_capture_header(struct rbd_img_request *img_req)
 {
 	struct rbd_device *rbd_dev = img_req->rbd_dev;
 
 	lockdep_assert_held(&rbd_dev->header_rwsem);
 
-	if (rbd_img_is_write(img_req))
-		img_req->snapc = ceph_get_snap_context(rbd_dev->header.snapc);
-	else
+	if (!rbd_img_is_write(img_req))
 		img_req->snap_id = rbd_dev->spec->snap_id;
 
 	if (rbd_dev_parent_get(rbd_dev))
@@ -2233,9 +2253,6 @@ static int rbd_obj_init_write(struct rbd_obj_request *obj_req)
 	if (ret)
 		return ret;
 
-	if (rbd_obj_copyup_enabled(obj_req))
-		obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
-
 	obj_req->write_state = RBD_OBJ_WRITE_START;
 	return 0;
 }
@@ -2341,8 +2358,6 @@ static int rbd_obj_init_zeroout(struct rbd_obj_request *obj_req)
 	if (ret)
 		return ret;
 
-	if (rbd_obj_copyup_enabled(obj_req))
-		obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
 	if (!obj_req->num_img_extents) {
 		obj_req->flags |= RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT;
 		if (rbd_obj_is_entire(obj_req))
@@ -3286,6 +3301,7 @@ again:
 	case RBD_OBJ_WRITE_START:
 		rbd_assert(!*result);
 
+		rbd_obj_set_copyup_enabled(obj_req);
 		if (rbd_obj_write_is_noop(obj_req))
 			return true;
 
@@ -3472,9 +3488,19 @@ static int rbd_img_exclusive_lock(struct rbd_img_request *img_req)
 
 static void rbd_img_object_requests(struct rbd_img_request *img_req)
 {
+	struct rbd_device *rbd_dev = img_req->rbd_dev;
 	struct rbd_obj_request *obj_req;
 
 	rbd_assert(!img_req->pending.result && !img_req->pending.num_pending);
+	rbd_assert(!need_exclusive_lock(img_req) ||
+		   __rbd_is_lock_owner(rbd_dev));
+
+	if (rbd_img_is_write(img_req)) {
+		rbd_assert(!img_req->snapc);
+		down_read(&rbd_dev->header_rwsem);
+		img_req->snapc = ceph_get_snap_context(rbd_dev->header.snapc);
+		up_read(&rbd_dev->header_rwsem);
+	}
 
 	for_each_obj_request(img_req, obj_req) {
 		int result = 0;
@@ -3492,7 +3518,6 @@ static void rbd_img_object_requests(struct rbd_img_request *img_req)
 
 static bool rbd_img_advance(struct rbd_img_request *img_req, int *result)
 {
-	struct rbd_device *rbd_dev = img_req->rbd_dev;
 	int ret;
 
 again:
@@ -3513,9 +3538,6 @@ again:
 		if (*result)
 			return true;
 
-		rbd_assert(!need_exclusive_lock(img_req) ||
-			   __rbd_is_lock_owner(rbd_dev));
-
 		rbd_img_object_requests(img_req);
 		if (!img_req->pending.num_pending) {
 			*result = img_req->pending.result;
@@ -3977,6 +3999,10 @@ static int rbd_post_acquire_action(struct rbd_device *rbd_dev)
 {
 	int ret;
 
+	ret = rbd_dev_refresh(rbd_dev);
+	if (ret)
+		return ret;
+
 	if (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP) {
 		ret = rbd_object_map_open(rbd_dev);
 		if (ret)
diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index 789be30d6ee220..2321e5ddb664de 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -1627,6 +1627,7 @@ void ceph_flush_snaps(struct ceph_inode_info *ci,
 	struct inode *inode = &ci->netfs.inode;
 	struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 	struct ceph_mds_session *session = NULL;
+	bool need_put = false;
 	int mds;
 
 	dout("ceph_flush_snaps %p\n", inode);
@@ -1671,8 +1672,13 @@ out:
 		ceph_put_mds_session(session);
 	/* we flushed them all; remove this inode from the queue */
 	spin_lock(&mdsc->snap_flush_lock);
+	if (!list_empty(&ci->i_snap_flush_item))
+		need_put = true;
 	list_del_init(&ci->i_snap_flush_item);
 	spin_unlock(&mdsc->snap_flush_lock);
+
+	if (need_put)
+		iput(inode);
 }
 
 /*
diff --git a/fs/ceph/snap.c b/fs/ceph/snap.c
index 0b236ebd989fca..2e73ba62bd7aa7 100644
--- a/fs/ceph/snap.c
+++ b/fs/ceph/snap.c
@@ -693,8 +693,10 @@ int __ceph_finish_cap_snap(struct ceph_inode_info *ci,
 	     capsnap->size);
 
 	spin_lock(&mdsc->snap_flush_lock);
-	if (list_empty(&ci->i_snap_flush_item))
+	if (list_empty(&ci->i_snap_flush_item)) {
+		ihold(inode);
 		list_add_tail(&ci->i_snap_flush_item, &mdsc->snap_flush_list);
+	}
 	spin_unlock(&mdsc->snap_flush_lock);
 	return 1;  /* caller may want to ceph_flush_snaps */
 }