RDMA/srp: Rework the srp_add_port() error path

device_register() always calls device_initialize() so calling device_del()
is safe even if device_register() fails. Implement the following advice
from the comment block above device_register(): "NOTE: _Never_ directly free
@dev after calling this function, even if it returned an error! Always use
put_device() to give up the reference initialized in this function instead."
Keep the kfree() call in the error path since srp_release_dev() does not
free the host.

Link: https://lore.kernel.org/r/20220825213900.864587-2-bvanassche@acm.org
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Leon Romanovsky <leon@kernel.org>

diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c
index 7720ea2..8fd6a88 100644 (file)
--- a/drivers/infiniband/ulp/srp/ib_srp.c
+++ b/drivers/infiniband/ulp/srp/ib_srp.c
@@ -3909,20 +3909,19 @@ static struct srp_host *srp_add_port(struct srp_device *device, u8 port)
                     port);
 
        if (device_register(&host->dev))
-               goto free_host;
+               goto put_host;
        if (device_create_file(&host->dev, &dev_attr_add_target))
-               goto err_class;
+               goto put_host;
        if (device_create_file(&host->dev, &dev_attr_ibdev))
-               goto err_class;
+               goto put_host;
        if (device_create_file(&host->dev, &dev_attr_port))
-               goto err_class;
+               goto put_host;
 
        return host;
 
-err_class:
-       device_unregister(&host->dev);
-
-free_host:
+put_host:
+       device_del(&host->dev);
+       put_device(&host->dev);
        kfree(host);
 
        return NULL;