| Message ID | 20230627120340.19432-1-dg573847474@gmail.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | 6feb37b3b06e9049e20dcf7e23998f92c9c5be9a |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | sctp: fix potential deadlock on &net->sctp.addr_wq_lock | expand |
On Tue, Jun 27, 2023 at 8:04 AM Chengfeng Ye <dg573847474@gmail.com> wrote: > > As &net->sctp.addr_wq_lock is also acquired by the timer > sctp_addr_wq_timeout_handler() in protocal.c, the same lock acquisition > at sctp_auto_asconf_init() seems should disable irq since it is called > from sctp_accept() under process context. > > Possible deadlock scenario: > sctp_accept() > -> sctp_sock_migrate() > -> sctp_auto_asconf_init() > -> spin_lock(&net->sctp.addr_wq_lock) > <timer interrupt> > -> sctp_addr_wq_timeout_handler() > -> spin_lock_bh(&net->sctp.addr_wq_lock); (deadlock here) > > This flaw was found using an experimental static analysis tool we are > developing for irq-related deadlock. > > The tentative patch fix the potential deadlock by spin_lock_bh(). > > Signed-off-by: Chengfeng Ye <dg573847474@gmail.com> > --- > net/sctp/socket.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/net/sctp/socket.c b/net/sctp/socket.c > index cda8c2874691..b2c7d17ff848 100644 > --- a/net/sctp/socket.c > +++ b/net/sctp/socket.c > @@ -364,9 +364,9 @@ static void sctp_auto_asconf_init(struct sctp_sock *sp) > struct net *net = sock_net(&sp->inet.sk); > > if (net->sctp.default_auto_asconf) { > - spin_lock(&net->sctp.addr_wq_lock); > + spin_lock_bh(&net->sctp.addr_wq_lock); > list_add_tail(&sp->auto_asconf_list, &net->sctp.auto_asconf_splist); > - spin_unlock(&net->sctp.addr_wq_lock); > + spin_unlock_bh(&net->sctp.addr_wq_lock); > sp->do_auto_asconf = 1; > } > } > -- > 2.17.1 > Fixes: 34e5b0118685 ("sctp: delay auto_asconf init until binding the first addr") Acked-by: Xin Long <lucien.xin@gmail.com> Thanks.
On Tue, 2023-06-27 at 12:03 +0000, Chengfeng Ye wrote: > As &net->sctp.addr_wq_lock is also acquired by the timer > sctp_addr_wq_timeout_handler() in protocal.c, the same lock acquisition > at sctp_auto_asconf_init() seems should disable irq since it is called > from sctp_accept() under process context. > > Possible deadlock scenario: > sctp_accept() > -> sctp_sock_migrate() > -> sctp_auto_asconf_init() > -> spin_lock(&net->sctp.addr_wq_lock) > <timer interrupt> > -> sctp_addr_wq_timeout_handler() > -> spin_lock_bh(&net->sctp.addr_wq_lock); (deadlock here) > > This flaw was found using an experimental static analysis tool we are > developing for irq-related deadlock. > > The tentative patch fix the potential deadlock by spin_lock_bh(). Patch LGTM. Please note that the above suggests a possible net-next follow- up/cleanup, replacing the spin_lock_bh() in sctp_addr_wq_timeout_handler() with a simple/faster spin_lock() - since sctp_addr_wq_timeout_handler() runs with BH disabled. Anyhow net-next is closed now, it will have to wait a bit ;) Cheers, Paolo
Hello: This patch was applied to netdev/net.git (main) by Paolo Abeni <pabeni@redhat.com>: On Tue, 27 Jun 2023 12:03:40 +0000 you wrote: > As &net->sctp.addr_wq_lock is also acquired by the timer > sctp_addr_wq_timeout_handler() in protocal.c, the same lock acquisition > at sctp_auto_asconf_init() seems should disable irq since it is called > from sctp_accept() under process context. > > Possible deadlock scenario: > sctp_accept() > -> sctp_sock_migrate() > -> sctp_auto_asconf_init() > -> spin_lock(&net->sctp.addr_wq_lock) > <timer interrupt> > -> sctp_addr_wq_timeout_handler() > -> spin_lock_bh(&net->sctp.addr_wq_lock); (deadlock here) > > [...] Here is the summary with links: - sctp: fix potential deadlock on &net->sctp.addr_wq_lock https://git.kernel.org/netdev/net/c/6feb37b3b06e You are awesome, thank you!
diff --git a/net/sctp/socket.c b/net/sctp/socket.c index cda8c2874691..b2c7d17ff848 100644 --- a/net/sctp/socket.c +++ b/net/sctp/socket.c @@ -364,9 +364,9 @@ static void sctp_auto_asconf_init(struct sctp_sock *sp) struct net *net = sock_net(&sp->inet.sk); if (net->sctp.default_auto_asconf) { - spin_lock(&net->sctp.addr_wq_lock); + spin_lock_bh(&net->sctp.addr_wq_lock); list_add_tail(&sp->auto_asconf_list, &net->sctp.auto_asconf_splist); - spin_unlock(&net->sctp.addr_wq_lock); + spin_unlock_bh(&net->sctp.addr_wq_lock); sp->do_auto_asconf = 1; } }
As &net->sctp.addr_wq_lock is also acquired by the timer sctp_addr_wq_timeout_handler() in protocal.c, the same lock acquisition at sctp_auto_asconf_init() seems should disable irq since it is called from sctp_accept() under process context. Possible deadlock scenario: sctp_accept() -> sctp_sock_migrate() -> sctp_auto_asconf_init() -> spin_lock(&net->sctp.addr_wq_lock) <timer interrupt> -> sctp_addr_wq_timeout_handler() -> spin_lock_bh(&net->sctp.addr_wq_lock); (deadlock here) This flaw was found using an experimental static analysis tool we are developing for irq-related deadlock. The tentative patch fix the potential deadlock by spin_lock_bh(). Signed-off-by: Chengfeng Ye <dg573847474@gmail.com> --- net/sctp/socket.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)