Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client

Pull Ceph fixes from Sage Weil:
 "There is a lifecycle fix in the auth code, a fix for a narrow race
  condition on map, and a helpful message in the log when there is a
  feature mismatch (which happens frequently now that the default
  server-side options have changed)"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
  rbd: report unsupported features to syslog
  rbd: fix rbd map vs notify races
  libceph: make authorizer destruction independent of ceph_auth_client

10 files changed, 87 insertions, 92 deletions

diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c
index 94a1843b0426d..0ede6d7e25686 100644
--- a/drivers/block/rbd.c
+++ b/drivers/block/rbd.c
@@ -538,7 +538,6 @@ static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
 				u8 *order, u64 *snap_size);
 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
 		u64 *snap_features);
-static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name);
 
 static int rbd_open(struct block_device *bdev, fmode_t mode)
 {
@@ -3127,9 +3126,6 @@ static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
 	struct rbd_device *rbd_dev = (struct rbd_device *)data;
 	int ret;
 
-	if (!rbd_dev)
-		return;
-
 	dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
 		rbd_dev->header_name, (unsigned long long)notify_id,
 		(unsigned int)opcode);
@@ -3263,6 +3259,9 @@ static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev)
 
 	ceph_osdc_cancel_event(rbd_dev->watch_event);
 	rbd_dev->watch_event = NULL;
+
+	dout("%s flushing notifies\n", __func__);
+	ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
 }
 
 /*
@@ -3642,21 +3641,14 @@ static void rbd_exists_validate(struct rbd_device *rbd_dev)
 static void rbd_dev_update_size(struct rbd_device *rbd_dev)
 {
 	sector_t size;
-	bool removing;
 
 	/*
-	 * Don't hold the lock while doing disk operations,
-	 * or lock ordering will conflict with the bdev mutex via:
-	 * rbd_add() -> blkdev_get() -> rbd_open()
+	 * If EXISTS is not set, rbd_dev->disk may be NULL, so don't
+	 * try to update its size.  If REMOVING is set, updating size
+	 * is just useless work since the device can't be opened.
 	 */
-	spin_lock_irq(&rbd_dev->lock);
-	removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
-	spin_unlock_irq(&rbd_dev->lock);
-	/*
-	 * If the device is being removed, rbd_dev->disk has
-	 * been destroyed, so don't try to update its size
-	 */
-	if (!removing) {
+	if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) &&
+	    !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) {
 		size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
 		dout("setting size to %llu sectors", (unsigned long long)size);
 		set_capacity(rbd_dev->disk, size);
@@ -4191,7 +4183,7 @@ static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
 		__le64 features;
 		__le64 incompat;
 	} __attribute__ ((packed)) features_buf = { 0 };
-	u64 incompat;
+	u64 unsup;
 	int ret;
 
 	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
@@ -4204,9 +4196,12 @@ static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
 	if (ret < sizeof (features_buf))
 		return -ERANGE;
 
-	incompat = le64_to_cpu(features_buf.incompat);
-	if (incompat & ~RBD_FEATURES_SUPPORTED)
+	unsup = le64_to_cpu(features_buf.incompat) & ~RBD_FEATURES_SUPPORTED;
+	if (unsup) {
+		rbd_warn(rbd_dev, "image uses unsupported features: 0x%llx",
+			 unsup);
 		return -ENXIO;
+	}
 
 	*snap_features = le64_to_cpu(features_buf.features);
 
@@ -5187,6 +5182,10 @@ out_err:
 	return ret;
 }
 
+/*
+ * rbd_dev->header_rwsem must be locked for write and will be unlocked
+ * upon return.
+ */
 static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
 {
 	int ret;
@@ -5195,7 +5194,7 @@ static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
 
 	ret = rbd_dev_id_get(rbd_dev);
 	if (ret)
-		return ret;
+		goto err_out_unlock;
 
 	BUILD_BUG_ON(DEV_NAME_LEN
 			< sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
@@ -5236,8 +5235,9 @@ static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
 	/* Everything's ready.  Announce the disk to the world. */
 
 	set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
-	add_disk(rbd_dev->disk);
+	up_write(&rbd_dev->header_rwsem);
 
+	add_disk(rbd_dev->disk);
 	pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
 		(unsigned long long) rbd_dev->mapping.size);
 
@@ -5252,6 +5252,8 @@ err_out_blkdev:
 		unregister_blkdev(rbd_dev->major, rbd_dev->name);
 err_out_id:
 	rbd_dev_id_put(rbd_dev);
+err_out_unlock:
+	up_write(&rbd_dev->header_rwsem);
 	return ret;
 }
 
@@ -5442,6 +5444,7 @@ static ssize_t do_rbd_add(struct bus_type *bus,
 	spec = NULL;		/* rbd_dev now owns this */
 	rbd_opts = NULL;	/* rbd_dev now owns this */
 
+	down_write(&rbd_dev->header_rwsem);
 	rc = rbd_dev_image_probe(rbd_dev, 0);
 	if (rc < 0)
 		goto err_out_rbd_dev;
@@ -5471,6 +5474,7 @@ out:
 	return rc;
 
 err_out_rbd_dev:
+	up_write(&rbd_dev->header_rwsem);
 	rbd_dev_destroy(rbd_dev);
 err_out_client:
 	rbd_put_client(rbdc);
@@ -5577,12 +5581,6 @@ static ssize_t do_rbd_remove(struct bus_type *bus,
 		return ret;
 
 	rbd_dev_header_unwatch_sync(rbd_dev);
-	/*
-	 * flush remaining watch callbacks - these must be complete
-	 * before the osd_client is shutdown
-	 */
-	dout("%s: flushing notifies", __func__);
-	ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
 
 	/*
 	 * Don't free anything from rbd_dev->disk until after all
diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 541ead4d89650..85b8517f17a09 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -386,9 +386,7 @@ void ceph_put_mds_session(struct ceph_mds_session *s)
 	     atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
 	if (atomic_dec_and_test(&s->s_ref)) {
 		if (s->s_auth.authorizer)
-			ceph_auth_destroy_authorizer(
-				s->s_mdsc->fsc->client->monc.auth,
-				s->s_auth.authorizer);
+			ceph_auth_destroy_authorizer(s->s_auth.authorizer);
 		kfree(s);
 	}
 }
@@ -3900,7 +3898,7 @@ static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
 	struct ceph_auth_handshake *auth = &s->s_auth;
 
 	if (force_new && auth->authorizer) {
-		ceph_auth_destroy_authorizer(ac, auth->authorizer);
+		ceph_auth_destroy_authorizer(auth->authorizer);
 		auth->authorizer = NULL;
 	}
 	if (!auth->authorizer) {
diff --git a/include/linux/ceph/auth.h b/include/linux/ceph/auth.h
index 260d78b587c48..1563265d20974 100644
--- a/include/linux/ceph/auth.h
+++ b/include/linux/ceph/auth.h
@@ -12,9 +12,12 @@
  */
 
 struct ceph_auth_client;
-struct ceph_authorizer;
 struct ceph_msg;
 
+struct ceph_authorizer {
+	void (*destroy)(struct ceph_authorizer *);
+};
+
 struct ceph_auth_handshake {
 	struct ceph_authorizer *authorizer;
 	void *authorizer_buf;
@@ -62,8 +65,6 @@ struct ceph_auth_client_ops {
 				 struct ceph_auth_handshake *auth);
 	int (*verify_authorizer_reply)(struct ceph_auth_client *ac,
 				       struct ceph_authorizer *a, size_t len);
-	void (*destroy_authorizer)(struct ceph_auth_client *ac,
-				   struct ceph_authorizer *a);
 	void (*invalidate_authorizer)(struct ceph_auth_client *ac,
 				      int peer_type);
 
@@ -112,8 +113,7 @@ extern int ceph_auth_is_authenticated(struct ceph_auth_client *ac);
 extern int ceph_auth_create_authorizer(struct ceph_auth_client *ac,
 				       int peer_type,
 				       struct ceph_auth_handshake *auth);
-extern void ceph_auth_destroy_authorizer(struct ceph_auth_client *ac,
-					 struct ceph_authorizer *a);
+void ceph_auth_destroy_authorizer(struct ceph_authorizer *a);
 extern int ceph_auth_update_authorizer(struct ceph_auth_client *ac,
 				       int peer_type,
 				       struct ceph_auth_handshake *a);
diff --git a/include/linux/ceph/osd_client.h b/include/linux/ceph/osd_client.h
index 4343df8067101..cbf460927c424 100644
--- a/include/linux/ceph/osd_client.h
+++ b/include/linux/ceph/osd_client.h
@@ -16,7 +16,6 @@ struct ceph_msg;
 struct ceph_snap_context;
 struct ceph_osd_request;
 struct ceph_osd_client;
-struct ceph_authorizer;
 
 /*
  * completion callback for async writepages
diff --git a/net/ceph/auth.c b/net/ceph/auth.c
index 6b923bcaa2a48..2bc5965fdd1ec 100644
--- a/net/ceph/auth.c
+++ b/net/ceph/auth.c
@@ -293,13 +293,9 @@ int ceph_auth_create_authorizer(struct ceph_auth_client *ac,
 }
 EXPORT_SYMBOL(ceph_auth_create_authorizer);
 
-void ceph_auth_destroy_authorizer(struct ceph_auth_client *ac,
-				  struct ceph_authorizer *a)
+void ceph_auth_destroy_authorizer(struct ceph_authorizer *a)
 {
-	mutex_lock(&ac->mutex);
-	if (ac->ops && ac->ops->destroy_authorizer)
-		ac->ops->destroy_authorizer(ac, a);
-	mutex_unlock(&ac->mutex);
+	a->destroy(a);
 }
 EXPORT_SYMBOL(ceph_auth_destroy_authorizer);
 
diff --git a/net/ceph/auth_none.c b/net/ceph/auth_none.c
index 8c93fa8d81bc4..5f836f02ae36c 100644
--- a/net/ceph/auth_none.c
+++ b/net/ceph/auth_none.c
@@ -16,7 +16,6 @@ static void reset(struct ceph_auth_client *ac)
 	struct ceph_auth_none_info *xi = ac->private;
 
 	xi->starting = true;
-	xi->built_authorizer = false;
 }
 
 static void destroy(struct ceph_auth_client *ac)
@@ -39,6 +38,27 @@ static int should_authenticate(struct ceph_auth_client *ac)
 	return xi->starting;
 }
 
+static int ceph_auth_none_build_authorizer(struct ceph_auth_client *ac,
+					   struct ceph_none_authorizer *au)
+{
+	void *p = au->buf;
+	void *const end = p + sizeof(au->buf);
+	int ret;
+
+	ceph_encode_8_safe(&p, end, 1, e_range);
+	ret = ceph_entity_name_encode(ac->name, &p, end);
+	if (ret < 0)
+		return ret;
+
+	ceph_encode_64_safe(&p, end, ac->global_id, e_range);
+	au->buf_len = p - (void *)au->buf;
+	dout("%s built authorizer len %d\n", __func__, au->buf_len);
+	return 0;
+
+e_range:
+	return -ERANGE;
+}
+
 static int build_request(struct ceph_auth_client *ac, void *buf, void *end)
 {
 	return 0;
@@ -57,32 +77,32 @@ static int handle_reply(struct ceph_auth_client *ac, int result,
 	return result;
 }
 
+static void ceph_auth_none_destroy_authorizer(struct ceph_authorizer *a)
+{
+	kfree(a);
+}
+
 /*
- * build an 'authorizer' with our entity_name and global_id.  we can
- * reuse a single static copy since it is identical for all services
- * we connect to.
+ * build an 'authorizer' with our entity_name and global_id.  it is
+ * identical for all services we connect to.
  */
 static int ceph_auth_none_create_authorizer(
 	struct ceph_auth_client *ac, int peer_type,
 	struct ceph_auth_handshake *auth)
 {
-	struct ceph_auth_none_info *ai = ac->private;
-	struct ceph_none_authorizer *au = &ai->au;
-	void *p, *end;
+	struct ceph_none_authorizer *au;
 	int ret;
 
-	if (!ai->built_authorizer) {
-		p = au->buf;
-		end = p + sizeof(au->buf);
-		ceph_encode_8(&p, 1);
-		ret = ceph_entity_name_encode(ac->name, &p, end - 8);
-		if (ret < 0)
-			goto bad;
-		ceph_decode_need(&p, end, sizeof(u64), bad2);
-		ceph_encode_64(&p, ac->global_id);
-		au->buf_len = p - (void *)au->buf;
-		ai->built_authorizer = true;
-		dout("built authorizer len %d\n", au->buf_len);
+	au = kmalloc(sizeof(*au), GFP_NOFS);
+	if (!au)
+		return -ENOMEM;
+
+	au->base.destroy = ceph_auth_none_destroy_authorizer;
+
+	ret = ceph_auth_none_build_authorizer(ac, au);
+	if (ret) {
+		kfree(au);
+		return ret;
 	}
 
 	auth->authorizer = (struct ceph_authorizer *) au;
@@ -92,17 +112,6 @@ static int ceph_auth_none_create_authorizer(
 	auth->authorizer_reply_buf_len = sizeof (au->reply_buf);
 
 	return 0;
-
-bad2:
-	ret = -ERANGE;
-bad:
-	return ret;
-}
-
-static void ceph_auth_none_destroy_authorizer(struct ceph_auth_client *ac,
-				      struct ceph_authorizer *a)
-{
-	/* nothing to do */
 }
 
 static const struct ceph_auth_client_ops ceph_auth_none_ops = {
@@ -114,7 +123,6 @@ static const struct ceph_auth_client_ops ceph_auth_none_ops = {
 	.build_request = build_request,
 	.handle_reply = handle_reply,
 	.create_authorizer = ceph_auth_none_create_authorizer,
-	.destroy_authorizer = ceph_auth_none_destroy_authorizer,
 };
 
 int ceph_auth_none_init(struct ceph_auth_client *ac)
@@ -127,7 +135,6 @@ int ceph_auth_none_init(struct ceph_auth_client *ac)
 		return -ENOMEM;
 
 	xi->starting = true;
-	xi->built_authorizer = false;
 
 	ac->protocol = CEPH_AUTH_NONE;
 	ac->private = xi;
diff --git a/net/ceph/auth_none.h b/net/ceph/auth_none.h
index 059a3ce4b53f4..62021535ae4a8 100644
--- a/net/ceph/auth_none.h
+++ b/net/ceph/auth_none.h
@@ -12,6 +12,7 @@
  */
 
 struct ceph_none_authorizer {
+	struct ceph_authorizer base;
 	char buf[128];
 	int buf_len;
 	char reply_buf[0];
@@ -19,8 +20,6 @@ struct ceph_none_authorizer {
 
 struct ceph_auth_none_info {
 	bool starting;
-	bool built_authorizer;
-	struct ceph_none_authorizer au;   /* we only need one; it's static */
 };
 
 int ceph_auth_none_init(struct ceph_auth_client *ac);
diff --git a/net/ceph/auth_x.c b/net/ceph/auth_x.c
index 9e43a315e6622..a0905f04bd13f 100644
--- a/net/ceph/auth_x.c
+++ b/net/ceph/auth_x.c
@@ -565,6 +565,14 @@ static int ceph_x_handle_reply(struct ceph_auth_client *ac, int result,
 	return -EAGAIN;
 }
 
+static void ceph_x_destroy_authorizer(struct ceph_authorizer *a)
+{
+	struct ceph_x_authorizer *au = (void *)a;
+
+	ceph_x_authorizer_cleanup(au);
+	kfree(au);
+}
+
 static int ceph_x_create_authorizer(
 	struct ceph_auth_client *ac, int peer_type,
 	struct ceph_auth_handshake *auth)
@@ -581,6 +589,8 @@ static int ceph_x_create_authorizer(
 	if (!au)
 		return -ENOMEM;
 
+	au->base.destroy = ceph_x_destroy_authorizer;
+
 	ret = ceph_x_build_authorizer(ac, th, au);
 	if (ret) {
 		kfree(au);
@@ -643,16 +653,6 @@ static int ceph_x_verify_authorizer_reply(struct ceph_auth_client *ac,
 	return ret;
 }
 
-static void ceph_x_destroy_authorizer(struct ceph_auth_client *ac,
-				      struct ceph_authorizer *a)
-{
-	struct ceph_x_authorizer *au = (void *)a;
-
-	ceph_x_authorizer_cleanup(au);
-	kfree(au);
-}
-
-
 static void ceph_x_reset(struct ceph_auth_client *ac)
 {
 	struct ceph_x_info *xi = ac->private;
@@ -770,7 +770,6 @@ static const struct ceph_auth_client_ops ceph_x_ops = {
 	.create_authorizer = ceph_x_create_authorizer,
 	.update_authorizer = ceph_x_update_authorizer,
 	.verify_authorizer_reply = ceph_x_verify_authorizer_reply,
-	.destroy_authorizer = ceph_x_destroy_authorizer,
 	.invalidate_authorizer = ceph_x_invalidate_authorizer,
 	.reset =  ceph_x_reset,
 	.destroy = ceph_x_destroy,
diff --git a/net/ceph/auth_x.h b/net/ceph/auth_x.h
index 40b1a3cf73973..21a5af904bae7 100644
--- a/net/ceph/auth_x.h
+++ b/net/ceph/auth_x.h
@@ -26,6 +26,7 @@ struct ceph_x_ticket_handler {
 
 
 struct ceph_x_authorizer {
+	struct ceph_authorizer base;
 	struct ceph_crypto_key session_key;
 	struct ceph_buffer *buf;
 	unsigned int service;
diff --git a/net/ceph/osd_client.c b/net/ceph/osd_client.c
index 32355d9d0103a..40a53a70efdff 100644
--- a/net/ceph/osd_client.c
+++ b/net/ceph/osd_client.c
@@ -1087,10 +1087,8 @@ static void put_osd(struct ceph_osd *osd)
 	dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
 	     atomic_read(&osd->o_ref) - 1);
 	if (atomic_dec_and_test(&osd->o_ref)) {
-		struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
-
 		if (osd->o_auth.authorizer)
-			ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
+			ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
 		kfree(osd);
 	}
 }
@@ -2984,7 +2982,7 @@ static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
 	struct ceph_auth_handshake *auth = &o->o_auth;
 
 	if (force_new && auth->authorizer) {
-		ceph_auth_destroy_authorizer(ac, auth->authorizer);
+		ceph_auth_destroy_authorizer(auth->authorizer);
 		auth->authorizer = NULL;
 	}
 	if (!auth->authorizer) {