From 11adfabee1ae69c32429b8e8b861de7bafa60ef7 Mon Sep 17 00:00:00 2001 From: Shifeng Li Date: Fri, 2 Feb 2024 19:53:13 -0800 Subject: [PATCH] RDMA/device: Fix a race between mad_client and cm_client init [ Upstream commit 7a8bccd8b29c321ac181369b42b04fecf05f98e2 ] The mad_client will be initialized in enable_device_and_get(), while the devices_rwsem will be downgraded to a read semaphore. There is a window that leads to the failed initialization for cm_client, since it can not get matched mad port from ib_mad_port_list, and the matched mad port will be added to the list after that. mad_client | cm_client ------------------|-------------------------------------------------------- ib_register_device| enable_device_and_get down_write(&devices_rwsem) xa_set_mark(&devices, DEVICE_REGISTERED) downgrade_write(&devices_rwsem) | |ib_cm_init |ib_register_client(&cm_client) |down_read(&devices_rwsem) |xa_for_each_marked (&devices, DEVICE_REGISTERED) |add_client_context |cm_add_one |ib_register_mad_agent |ib_get_mad_port |__ib_get_mad_port |list_for_each_entry(entry, &ib_mad_port_list, port_list) |return NULL |up_read(&devices_rwsem) | add_client_context| ib_mad_init_device| ib_mad_port_open | list_add_tail(&port_priv->port_list, &ib_mad_port_list) up_read(&devices_rwsem) | Fix it by using down_write(&devices_rwsem) in ib_register_client(). Fixes: d0899892edd0 ("RDMA/device: Provide APIs from the core code to help unregistration") Link: https://lore.kernel.org/r/20240203035313.98991-1-lishifeng@sangfor.com.cn Suggested-by: Jason Gunthorpe Signed-off-by: Shifeng Li Signed-off-by: Jason Gunthorpe Signed-off-by: Sasha Levin --- drivers/infiniband/core/device.c | 37 +++++++++++++++++++------------- 1 file changed, 22 insertions(+), 15 deletions(-) diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c index 3c29fd04b301..94c3bad72cc5 100644 --- a/drivers/infiniband/core/device.c +++ b/drivers/infiniband/core/device.c @@ -1686,7 +1686,7 @@ static int assign_client_id(struct ib_client *client) { int ret; - down_write(&clients_rwsem); + lockdep_assert_held(&clients_rwsem); /* * The add/remove callbacks must be called in FIFO/LIFO order. To * achieve this we assign client_ids so they are sorted in @@ -1695,14 +1695,11 @@ static int assign_client_id(struct ib_client *client) client->client_id = highest_client_id; ret = xa_insert(&clients, client->client_id, client, GFP_KERNEL); if (ret) - goto out; + return ret; highest_client_id++; xa_set_mark(&clients, client->client_id, CLIENT_REGISTERED); - -out: - up_write(&clients_rwsem); - return ret; + return 0; } static void remove_client_id(struct ib_client *client) @@ -1732,25 +1729,35 @@ int ib_register_client(struct ib_client *client) { struct ib_device *device; unsigned long index; + bool need_unreg = false; int ret; refcount_set(&client->uses, 1); init_completion(&client->uses_zero); + + /* + * The devices_rwsem is held in write mode to ensure that a racing + * ib_register_device() sees a consisent view of clients and devices. + */ + down_write(&devices_rwsem); + down_write(&clients_rwsem); ret = assign_client_id(client); if (ret) - return ret; + goto out; - down_read(&devices_rwsem); + need_unreg = true; xa_for_each_marked (&devices, index, device, DEVICE_REGISTERED) { ret = add_client_context(device, client); - if (ret) { - up_read(&devices_rwsem); - ib_unregister_client(client); - return ret; - } + if (ret) + goto out; } - up_read(&devices_rwsem); - return 0; + ret = 0; +out: + up_write(&clients_rwsem); + up_write(&devices_rwsem); + if (need_unreg && ret) + ib_unregister_client(client); + return ret; } EXPORT_SYMBOL(ib_register_client);