netlink: fix locking around NETLINK_LIST_MEMBERSHIPS

Currently, NETLINK_LIST_MEMBERSHIPS grabs the netlink table while copying
the membership state to user-space. However, grabing the netlink table is
effectively a write_lock_irq(), and as such we should not be triggering
page-faults in the critical section.

This can be easily reproduced by the following snippet:
    int s = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
    void *p = mmap(0, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANON, -1, 0);
    int r = getsockopt(s, 0x10e, 9, p, (void*)((char*)p + 4092));

This should work just fine, but currently triggers EFAULT and a possible
WARN_ON below handle_mm_fault().

Fix this by reducing locking of NETLINK_LIST_MEMBERSHIPS to a read-side
lock. The write-lock was overkill in the first place, and the read-lock
allows page-faults just fine.

Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index 0a49a8c7c56432989de924e01563f692cf10e545..fafe33bdb61989e680dc4b26dbe99dcc1d4064b5 100644 (file)
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -2371,7 +2371,7 @@ static int netlink_getsockopt(struct socket *sock, int level, int optname,
                int pos, idx, shift;
 
                err = 0;
-               netlink_table_grab();
+               netlink_lock_table();
                for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
                        if (len - pos < sizeof(u32))
                                break;
@@ -2386,7 +2386,7 @@ static int netlink_getsockopt(struct socket *sock, int level, int optname,
                }
                if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
                        err = -EFAULT;
-               netlink_table_ungrab();
+               netlink_unlock_table();
                break;
        }
        case NETLINK_CAP_ACK: