[PATCH bpf-next 1/8] bpf: offload: don't require rtnl for dev list manipulation

[Jakub Kicinski]

We only need to hold rtnl_lock() around ndo calls.  The device
offload initialization doesn't require it.  Neither will soon-
-to-come querying the offload info.  Use struct rw_semaphore
because map offload will require sleeping with the semaphore
held for read.

Suggested-by: Kirill Tkhai <ktk...@virtuozzo.com>
Signed-off-by: Jakub Kicinski <jakub.kicin...@netronome.com>
Reviewed-by: Quentin Monnet <quentin.mon...@netronome.com>
---
 kernel/bpf/offload.c | 21 +++++++++++++++++----
 1 file changed, 17 insertions(+), 4 deletions(-)

diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c
index 8455b89d1bbf..b88e5ebdc61d 100644
--- a/kernel/bpf/offload.c
+++ b/kernel/bpf/offload.c
@@ -20,8 +20,12 @@
 #include <linux/netdevice.h>
 #include <linux/printk.h>
 #include <linux/rtnetlink.h>
+#include <linux/rwsem.h>
 
-/* protected by RTNL */
+/* Protects bpf_prog_offload_devs and offload members of all progs.
+ * RTNL lock cannot be taken when holding this lock.
+ */
+static struct rw_semaphore bpf_devs_lock;
 static LIST_HEAD(bpf_prog_offload_devs);
 
 int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr)
@@ -43,17 +47,21 @@ int bpf_prog_offload_init(struct bpf_prog *prog, union 
bpf_attr *attr)
        offload->prog = prog;
        init_waitqueue_head(&offload->verifier_done);
 
-       rtnl_lock();
+       /* Our UNREGISTER notifier will grab bpf_devs_lock, so we are safe
+        * to assume the netdev doesn't get unregistered as long as we hold
+        * bpf_devs_lock.
+        */
+       down_write(&bpf_devs_lock);
        offload->netdev = __dev_get_by_index(net, attr->prog_ifindex);
        if (!offload->netdev) {
-               rtnl_unlock();
+               up_write(&bpf_devs_lock);
                kfree(offload);
                return -EINVAL;
        }
 
        prog->aux->offload = offload;
        list_add_tail(&offload->offloads, &bpf_prog_offload_devs);
-       rtnl_unlock();
+       up_write(&bpf_devs_lock);
 
        return 0;
 }
@@ -126,7 +134,9 @@ void bpf_prog_offload_destroy(struct bpf_prog *prog)
        wake_up(&offload->verifier_done);
 
        rtnl_lock();
+       down_write(&bpf_devs_lock);
        __bpf_prog_offload_destroy(prog);
+       up_write(&bpf_devs_lock);
        rtnl_unlock();
 
        kfree(offload);
@@ -181,11 +191,13 @@ static int bpf_offload_notification(struct notifier_block 
*notifier,
                if (netdev->reg_state != NETREG_UNREGISTERING)
                        break;
 
+               down_write(&bpf_devs_lock);
                list_for_each_entry_safe(offload, tmp, &bpf_prog_offload_devs,
                                         offloads) {
                        if (offload->netdev == netdev)
                                __bpf_prog_offload_destroy(offload->prog);
                }
+               up_write(&bpf_devs_lock);
                break;
        default:
                break;
@@ -199,6 +211,7 @@ static struct notifier_block bpf_offload_notifier = {
 
 static int __init bpf_offload_init(void)
 {
+       init_rwsem(&bpf_devs_lock);
        register_netdevice_notifier(&bpf_offload_notifier);
        return 0;
 }
-- 
2.15.1