mptcp: get rid of msk->subflow
Subsystems (3)
Paolo Abeni
maintainercover mptcp: get rid of msk->subflow
The mptcp protocol maintains an additional struct socket per connection, mainly to be able to use easily tcp-level struct socket operations.
That lead to several ill side effects, beyond the quite unfortunate/ confusing field name
- active and passive sockets behavior is incosistent, as only active ones have not NULL msk->subflow, leading to different error handling (and different error code returned to the user-space) in several places.
- active sockets uses an unneeded, larger amount of memory
- passive sockets can't successfully go through accept()/disconnect() accept()
This series address all the above finally getting rid of the blamed field. The first 2 patches are minor cleanups, in the next 11 patches msk->subflow usage is sistematically removed from the mptcp protocol, replacing it with direct msk->first usage, eventually introducing new core helpers as needed.
The final patch finally dispose the field, and it's the only patch in the series intened to produce functional changes.
Paolo Abeni (14):
mptcp: more accurate NL event generation.
mptcp: avoid unneeded mptcp_token_destroy() calls
mptcp: avoid additional __inet_stream_connect() call
mptcp: avoid subflow socket usage in mptcp_get_port()
net: factor out inet{,6}_bind_sk helpers
mptcp: mptcp: avoid additional indirection in mptcp_bind()
net: factor out __inet_listen_sk() helper
mptcp: avoid additional indirection in mptcp_listen()
mptcp: avoid additional indirection in mptcp_poll()
mptcp: avoid unneeded indirection in mptcp_stream_accept()
mptcp: avoid additional indirection in sockopt
mptcp: avoid ssock usage in mptcp_pm_nl_create_listen_socket()
mptcp: change the mpc check helper to return a sk
mptcp: get rid of msk->subflow
include/net/inet_common.h | 2 + include/net/ipv6.h | 1 + net/ipv4/af_inet.c | 47 ++++++----- net/ipv6/af_inet6.c | 10 ++- net/mptcp/pm_netlink.c | 30 ++++--- net/mptcp/protocol.c | 161 +++++++++++++++++++------------------- net/mptcp/protocol.h | 15 ++-- net/mptcp/sockopt.c | 65 ++++++++------- 8 files changed, 174 insertions(+), 157 deletions(-)
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 1/14 mptcp: more accurate NL event generation.
Currently the mptcp code generate a "new listener" event even if the actual listen() syscall fails. Address the issue moving the event generation call under the successful branch.
---
this could possibly go via -net, but is used by later patches and is not really critical IMHO.
Eventually the additional tag would be:
--- net/mptcp/protocol.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
net/mptcp/protocol.c +1 -2
@@ -3762,10 +3762,9 @@ static int mptcp_listen(struct socket *sock, int backlog)if (!err) {sock_prot_inuse_add (sock_net (sk), sk->sk_prot,1 );mptcp_copy_inaddrs (sk, ssock->sk); +mptcp_event_pm_listener (ssock->sk,MPTCP_EVENT_LISTENER_CREATED ); } -mptcp_event_pm_listener (ssock->sk,MPTCP_EVENT_LISTENER_CREATED ); - unlock:release_sock (sk);return err;
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 2/14 mptcp: avoid unneeded mptcp_token_destroy() calls
In reply to Paolo AbeniThe mptcp protocol currently clears the msk token both at connect() and listen() time. That was necessary before the mptcp protocol gained a full disconnect implmenetation, but after commit b29fcfb54cd7 ("mptcp: full disconnect implementation") such calls are no more necessary and a bit confusing.
Just drop them.
--- net/mptcp/protocol.c | 2 -- 1 file changed, 2 deletions(-)
net/mptcp/protocol.c +0 -2
@@ -3634,7 +3634,6 @@ static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)if (IS_ERR (ssock))return PTR_ERR (ssock); -mptcp_token_destroy (msk);inet_sk_state_store (sk,TCP_SYN_SENT ); subflow =mptcp_subflow_ctx (ssock->sk);#ifdef CONFIG_TCP_MD5SIG
@@ -3753,7 +3752,6 @@ static int mptcp_listen(struct socket *sock, int backlog) goto unlock; } -mptcp_token_destroy (msk);inet_sk_state_store (sk,TCP_LISTEN );sock_set_flag (sk,SOCK_RCU_FREE );
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 3/14 mptcp: avoid additional __inet_stream_connect() call
In reply to Paolo AbeniThe mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is __inet_stream_connect().
We are going to remove the first subflow socket soon, so avoid the addictional indirection via at connect time, calling directly into the sock-level connect() ops.
No functional change intended.
--- net/mptcp/protocol.c | 40 +++++++++++++++++++++++++++++----------- 1 file changed, 29 insertions(+), 11 deletions(-)
net/mptcp/protocol.c +29 -11
@@ -3629,22 +3629,24 @@ static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)struct mptcp_sock *msk =mptcp_sk (sk);struct socket *ssock;int err = -EINVAL ; +struct sock *ssk; ssock =__mptcp_nmpc_socket (msk);if (IS_ERR (ssock))return PTR_ERR (ssock);inet_sk_state_store (sk,TCP_SYN_SENT ); - subflow =mptcp_subflow_ctx (ssock->sk); + ssk = msk->first; + subflow =mptcp_subflow_ctx (ssk);#ifdef CONFIG_TCP_MD5SIG /* no MPTCP if MD5SIG is enabled on this socket or we may run out of * TCP option space. */ -if (rcu_access_pointer (tcp_sk (ssock->sk)->md5sig_info)) +if (rcu_access_pointer (tcp_sk (ssk)->md5sig_info))mptcp_subflow_early_fallback (msk, subflow);#endif -if (subflow->request_mptcp &&mptcp_token_new_connect (ssock->sk)) { -MPTCP_INC_STATS (sock_net (ssock->sk),MPTCP_MIB_TOKENFALLBACKINIT ); +if (subflow->request_mptcp &&mptcp_token_new_connect (ssk)) { +MPTCP_INC_STATS (sock_net (ssk),MPTCP_MIB_TOKENFALLBACKINIT );mptcp_subflow_early_fallback (msk, subflow); }if (likely (!__mptcp_check_fallback (msk)))
@@ -3653,21 +3655,37 @@ static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)/* if reaching here via the fastopen/sendmsg path, the caller already * acquired the subflow socket lock, too. */ -if (msk->fastopening) - err =__inet_stream_connect (ssock, uaddr, addr_len,O_NONBLOCK ,1 ); -else - err =inet_stream_connect (ssock, uaddr, addr_len,O_NONBLOCK ); -inet_sk (sk)->defer_connect =inet_sk (ssock->sk)->defer_connect; +if (!msk->fastopening) +lock_sock (ssk); + +if (ssk->sk_state !=TCP_CLOSE ) + goto out; + +if (BPF_CGROUP_PRE_CONNECT_ENABLED (ssk)) { + err = ssk->sk_prot->pre_connect (ssk, uaddr, addr_len); +if (err) + goto out; + } + + err = ssk->sk_prot->connect (ssk, uaddr, addr_len); +if (err <0 ) + goto out; + +inet_sk (sk)->defer_connect =inet_sk (ssk)->defer_connect; + +out: +if (!msk->fastopening) +release_sock (ssk);/* on successful connect, the msk state will be moved to established by * subflow_finish_connect() */ if (unlikely (err && err != -EINPROGRESS )) { -inet_sk_state_store (sk,inet_sk_state_load (ssock->sk)); +inet_sk_state_store (sk,inet_sk_state_load (ssk));return err; } -mptcp_copy_inaddrs (sk, ssock->sk); +mptcp_copy_inaddrs (sk, ssk);/* silence EINPROGRESS and let the caller inet_stream_connect * handle the connection in progress
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 4/14 mptcp: avoid subflow socket usage in mptcp_get_port()
In reply to Paolo AbeniWe are going to remove the first subflow socket soon, so avoid accessing it in mptcp_get_port(). Instead, access directly the first subflow sock.
--- net/mptcp/protocol.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-)
net/mptcp/protocol.c +3 -5
@@ -3445,14 +3445,12 @@ static void mptcp_unhash(struct sock *sk)static int mptcp_get_port (struct sock *sk,unsigned short snum) {struct mptcp_sock *msk =mptcp_sk (sk); -struct socket *ssock; - ssock = msk->subflow; -pr_debug ("msk=%p, subflow=%p" , msk, ssock); -if (WARN_ON_ONCE (!ssock)) +pr_debug ("msk=%p, ssk=%p" , msk, msk->first); +if (WARN_ON_ONCE (!msk->first))return -EINVAL ; -return inet_csk_get_port (ssock->sk, snum); +return inet_csk_get_port (msk->first, snum); }void mptcp_finish_connect (struct sock *ssk)
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 5/14 net: factor out inet{,6}_bind_sk helpers
In reply to Paolo AbeniThe mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is bind().
Factor out the helpers operating directly on the struct sock, to allow get rid of the above dependency in the next patch without duplicating the existing code.
No functional changes intended.
--- include/net/inet_common.h | 1 + include/net/ipv6.h | 1 + net/ipv4/af_inet.c | 8 ++++++-- net/ipv6/af_inet6.c | 10 +++++++--- 4 files changed, 15 insertions(+), 5 deletions(-)
include/net/inet_common.h +1 -0
@@ -42,6 +42,7 @@ int inet_shutdown(struct socket *sock, int how);int inet_listen (struct socket *sock,int backlog);void inet_sock_destruct (struct sock *sk);int inet_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len); +int inet_bind_sk (struct sock *sk,struct sockaddr *uaddr,int addr_len);/* Don't allocate port at this moment, defer to connect. */ #define BIND_FORCE_ADDRESS_NO_PORT (1 << 0)/* Grab and release socket lock. */
include/net/ipv6.h +1 -0
@@ -1220,6 +1220,7 @@ void inet6_cleanup_sock(struct sock *sk);void inet6_sock_destruct (struct sock *sk);int inet6_release (struct socket *sock);int inet6_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len); +int inet6_bind_sk (struct sock *sk,struct sockaddr *uaddr,int addr_len);int inet6_getname (struct socket *sock,struct sockaddr *uaddr,int peer);int inet6_ioctl (struct socket *sock,unsigned int cmd,unsigned long arg);
net/ipv4/af_inet.c +6 -2
@@ -431,9 +431,8 @@ int inet_release(struct socket *sock) }EXPORT_SYMBOL (inet_release); -int inet_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len) +int inet_bind_sk (struct sock *sk,struct sockaddr *uaddr,int addr_len) { -struct sock *sk = sock->sk;u32 flags =BIND_WITH_LOCK ;int err;
@@ -454,6 +453,11 @@ int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)return __inet_bind (sk, uaddr, addr_len, flags); } + +int inet_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len) +{ +return inet_bind_sk (sock->sk, uaddr, addr_len); +}EXPORT_SYMBOL (inet_bind); int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
net/ipv6/af_inet6.c +7 -3
@@ -435,10 +435,8 @@ static int __inet6_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len, goto out; } -/* bind for INET6 API */ -int inet6_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len) +int inet6_bind_sk (struct sock *sk,struct sockaddr *uaddr,int addr_len) { -struct sock *sk = sock->sk;u32 flags =BIND_WITH_LOCK ;const struct proto *prot;int err =0 ;
@@ -462,6 +460,12 @@ int inet6_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)return __inet6_bind (sk, uaddr, addr_len, flags); } + +/* bind for INET6 API */ +int inet6_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len) +{ +return inet6_bind_sk (sock->sk, uaddr, addr_len); +}EXPORT_SYMBOL (inet6_bind);int inet6_release (struct socket *sock)
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 6/14 mptcp: mptcp: avoid additional indirection in mptcp_bind()
In reply to Paolo AbeniWe are going to remove the first subflow socket soon, so avoid the addictional indirection via at bind() time. Instead call directly the recently introduced helpers on the first subflow sock.
--- net/mptcp/protocol.c | 17 ++++++++++++----- 1 file changed, 12 insertions(+), 5 deletions(-)
net/mptcp/protocol.c +12 -5
@@ -3728,22 +3728,29 @@ static struct proto mptcp_prot = {static int mptcp_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len) {struct mptcp_sock *msk =mptcp_sk (sock->sk); +struct sock *ssk, *sk = sock->sk;struct socket *ssock; -int err; +int err = -EINVAL ; -lock_sock (sock->sk); +lock_sock (sk); ssock =__mptcp_nmpc_socket (msk);if (IS_ERR (ssock)) { err =PTR_ERR (ssock); goto unlock; } - err = ssock->ops->bind (ssock, uaddr, addr_len); + ssk = msk->first; +if (sk->sk_family ==AF_INET ) + err =inet_bind_sk (ssk, uaddr, addr_len); +#if IS_ENABLED (CONFIG_MPTCP_IPV6 ) + else if (sk->sk_family == AF_INET6) + err =inet6_bind_sk (ssk, uaddr, addr_len); +#endif if (!err) -mptcp_copy_inaddrs (sock->sk, ssock->sk); +mptcp_copy_inaddrs (sk, ssk); unlock: -release_sock (sock->sk); +release_sock (sk);return err; }
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 7/14 net: factor out __inet_listen_sk() helper
In reply to Paolo AbeniThe mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is inet_listen().
Factor out an helper operating directly on the (locked) struct sock, to allow get rid of the above dependency in the next patch without duplicating the existing code.
No functional changes intended.
--- include/net/inet_common.h | 1 + net/ipv4/af_inet.c | 39 +++++++++++++++++++++++---------------- 2 files changed, 24 insertions(+), 16 deletions(-)
include/net/inet_common.h +1 -0
@@ -40,6 +40,7 @@ int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, int flags);int inet_shutdown (struct socket *sock,int how);int inet_listen (struct socket *sock,int backlog); +int __inet_listen_sk (struct sock *sk,int backlog);void inet_sock_destruct (struct sock *sk);int inet_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len);int inet_bind_sk (struct sock *sk,struct sockaddr *uaddr,int addr_len);
net/ipv4/af_inet.c +23 -16
@@ -187,24 +187,13 @@ static int inet_autobind(struct sock *sk)return 0 ; } -/* -* Move a socket into listening state. -*/ -int inet_listen (struct socket *sock,int backlog) +int __inet_listen_sk (struct sock *sk,int backlog) { -struct sock *sk = sock->sk; -unsigned char old_state; +unsigned char old_state = sk->sk_state;int err, tcp_fastopen; -lock_sock (sk); - - err = -EINVAL ; -if (sock->state !=SS_UNCONNECTED || sock->type !=SOCK_STREAM ) - goto out; - - old_state = sk->sk_state;if (!((1 << old_state) & (TCPF_CLOSE |TCPF_LISTEN ))) - goto out; +return -EINVAL ;WRITE_ONCE (sk->sk_max_ack_backlog, backlog);/* Really, if the socket is already in listen state
@@ -227,10 +216,28 @@ int inet_listen(struct socket *sock, int backlog) err =inet_csk_listen_start (sk);if (err) - goto out; +return err; +tcp_call_bpf (sk,BPF_SOCK_OPS_TCP_LISTEN_CB ,0 ,NULL ); } - err =0 ; +return 0 ; +} + +/* +* Move a socket into listening state. +*/ +int inet_listen (struct socket *sock,int backlog) +{ +struct sock *sk = sock->sk; +int err; + +lock_sock (sk); + + err = -EINVAL ; +if (sock->state !=SS_UNCONNECTED || sock->type !=SOCK_STREAM ) + goto out; + + err =__inet_listen_sk (sk, backlog); out:release_sock (sk);
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 8/14 mptcp: avoid additional indirection in mptcp_listen()
In reply to Paolo AbeniWe are going to remove the first subflow socket soon, so avoid the addictional indirection via at listen() time. Instead call directly the recently introduced helper on the first subflow sock.
No functional changes intended.
--- net/mptcp/protocol.c | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-)
net/mptcp/protocol.c +10 -4
@@ -3759,6 +3759,7 @@ static int mptcp_listen(struct socket *sock, int backlog)struct mptcp_sock *msk =mptcp_sk (sock->sk);struct sock *sk = sock->sk;struct socket *ssock; +struct sock *ssk;int err;pr_debug ("msk=%p" , msk);
@@ -3775,15 +3776,20 @@ static int mptcp_listen(struct socket *sock, int backlog) goto unlock; } + ssk = msk->first;inet_sk_state_store (sk,TCP_LISTEN );sock_set_flag (sk,SOCK_RCU_FREE ); - err = ssock->ops->listen (ssock, backlog); -inet_sk_state_store (sk,inet_sk_state_load (ssock->sk)); +lock_sock (ssk); + err =__inet_listen_sk (ssk, backlog); +release_sock (ssk); +inet_sk_state_store (sk,inet_sk_state_load (ssk)); +if (!err) { +WRITE_ONCE (sk->sk_max_ack_backlog, backlog);sock_prot_inuse_add (sock_net (sk), sk->sk_prot,1 ); -mptcp_copy_inaddrs (sk, ssock->sk); -mptcp_event_pm_listener (ssock->sk,MPTCP_EVENT_LISTENER_CREATED ); +mptcp_copy_inaddrs (sk, ssk); +mptcp_event_pm_listener (ssk,MPTCP_EVENT_LISTENER_CREATED ); } unlock:
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 9/14 mptcp: avoid additional indirection in mptcp_poll()
In reply to Paolo AbeniWe are going to remove the first subflow socket soon, so avoid the addictional indirection at poll() time. Instead access directly the first subflow sock.
No functional changes intended.
--- net/mptcp/protocol.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
net/mptcp/protocol.c +3 -3
@@ -3884,12 +3884,12 @@ static __poll_t mptcp_poll(struct file *file, struct socket *sock, state =inet_sk_state_load (sk);pr_debug ("msk=%p state=%d flags=%lx" , msk, state, msk->flags);if (state ==TCP_LISTEN ) { -struct socket *ssock =READ_ONCE (msk->subflow); +struct sock *ssk =READ_ONCE (msk->first); -if (WARN_ON_ONCE (!ssock || !ssock->sk)) +if (WARN_ON_ONCE (!ssk))return 0 ; -return inet_csk_listen_poll (ssock->sk); +return inet_csk_listen_poll (ssk); } shutdown =READ_ONCE (sk->sk_shutdown);
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 10/14 mptcp: avoid unneeded indirection in mptcp_stream_accept()
In reply to Paolo AbeniWe are going to remove the first subflow socket soon, so avoid the addictional indirection at accept() time. Instead access directly the first subflow sock.
No functional changes intended.
--- net/mptcp/protocol.c | 17 +++++------------ 1 file changed, 5 insertions(+), 12 deletions(-)
net/mptcp/protocol.c +5 -12
@@ -3220,17 +3220,12 @@ static struct sock *mptcp_accept(struct sock *sk, int flags, int *err, bool kern) {struct mptcp_sock *msk =mptcp_sk (sk); -struct socket *listener; -struct sock *newsk; +struct sock *ssk, *newsk; - listener =READ_ONCE (msk->subflow); -if (WARN_ON_ONCE (!listener)) { - *err = -EINVAL ; -return NULL ; - } + ssk =READ_ONCE (msk->first); -pr_debug ("msk=%p, listener=%p" , msk,mptcp_subflow_ctx (listener->sk)); - newsk =inet_csk_accept (listener->sk, flags, err, kern); +pr_debug ("msk=%p, listener=%p" , msk,mptcp_subflow_ctx (ssk)); + newsk =inet_csk_accept (ssk, flags, err, kern);if (!newsk)return NULL ;
@@ -3801,7 +3796,6 @@ static int mptcp_stream_accept(struct socket *sock, struct socket *newsock, int flags, bool kern) {struct mptcp_sock *msk =mptcp_sk (sock->sk); -struct socket *ssock;struct sock *newsk;int err;
@@ -3810,8 +3804,7 @@ static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,/* Buggy applications can call accept on socket states other then LISTEN * but no need to allocate the first subflow just to error out. */ - ssock =READ_ONCE (msk->subflow); -if (!ssock) +if (!READ_ONCE (msk->first))return -EINVAL ; newsk =mptcp_accept (sock->sk, flags, &err, kern);
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 11/14 mptcp: avoid additional indirection in sockopt
In reply to Paolo AbeniThe mptcp sockopt infrastructure unneedly uses the first subflow socket struct in a few spots. We are going to remove such field soon, so use directly the first subflow sock instead.
No functional changes intended.
--- net/mptcp/sockopt.c | 28 ++++++++++++++++------------ 1 file changed, 16 insertions(+), 12 deletions(-)
net/mptcp/sockopt.c +16 -12
@@ -293,6 +293,7 @@ static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname, {struct sock *sk = (struct sock *)msk;struct socket *ssock; +struct sock *ssk;int ret;switch (optname) {
@@ -307,16 +308,17 @@ static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,return PTR_ERR (ssock); } - ret =sock_setsockopt (ssock,SOL_SOCKET , optname, optval, optlen); + ssk = msk->first; + ret =sk_setsockopt (ssk,SOL_SOCKET , optname, optval, optlen);if (ret ==0 ) {if (optname ==SO_REUSEPORT ) - sk->sk_reuseport = ssock->sk->sk_reuseport; + sk->sk_reuseport = ssk->sk_reuseport;else if (optname ==SO_REUSEADDR ) - sk->sk_reuse = ssock->sk->sk_reuse; + sk->sk_reuse = ssk->sk_reuse;else if (optname ==SO_BINDTODEVICE ) - sk->sk_bound_dev_if = ssock->sk->sk_bound_dev_if; + sk->sk_bound_dev_if = ssk->sk_bound_dev_if;else if (optname ==SO_BINDTOIFINDEX ) - sk->sk_bound_dev_if = ssock->sk->sk_bound_dev_if; + sk->sk_bound_dev_if = ssk->sk_bound_dev_if; }release_sock (sk);return ret;
@@ -391,6 +393,7 @@ static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,struct sock *sk = (struct sock *)msk;int ret = -EOPNOTSUPP ;struct socket *ssock; +struct sock *ssk;switch (optname) {case IPV6_V6ONLY :
@@ -403,7 +406,8 @@ static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,return PTR_ERR (ssock); } - ret =tcp_setsockopt (ssock->sk,SOL_IPV6 , optname, optval, optlen); + ssk = msk->first; + ret =tcp_setsockopt (ssk,SOL_IPV6 , optname, optval, optlen);if (ret !=0 ) {release_sock (sk);return ret;
@@ -413,13 +417,13 @@ static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,switch (optname) {case IPV6_V6ONLY : - sk->sk_ipv6only = ssock->sk->sk_ipv6only; + sk->sk_ipv6only = ssk->sk_ipv6only;break ;case IPV6_TRANSPARENT : -inet_sk (sk)->transparent =inet_sk (ssock->sk)->transparent; +inet_sk (sk)->transparent =inet_sk (ssk)->transparent;break ;case IPV6_FREEBIND : -inet_sk (sk)->freebind =inet_sk (ssock->sk)->freebind; +inet_sk (sk)->freebind =inet_sk (ssk)->freebind;break ; }
@@ -700,7 +704,7 @@ static int mptcp_setsockopt_sol_ip_set_transparent(struct mptcp_sock *msk, int oreturn PTR_ERR (ssock); } - issk =inet_sk (ssock->sk); + issk =inet_sk (msk->first);switch (optname) {case IP_FREEBIND :
@@ -865,8 +869,8 @@ static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int {struct sock *sk = (struct sock *)msk;struct socket *ssock; -int ret;struct sock *ssk; +int ret;lock_sock (sk); ssk = msk->first;
@@ -881,7 +885,7 @@ static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int goto out; } - ret =tcp_getsockopt (ssock->sk, level, optname, optval, optlen); + ret =tcp_getsockopt (ssk, level, optname, optval, optlen); out:release_sock (sk);
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 12/14 mptcp: avoid ssock usage in mptcp_pm_nl_create_listen_socket()
In reply to Paolo AbeniThis is one of the few remaining spots actually manipulating the first subflow socket. We can leverage the recently introduced inet helpers to get rid of ssock there.
No functional changes intended.
--- net/mptcp/pm_netlink.c | 24 +++++++++++++++--------- 1 file changed, 15 insertions(+), 9 deletions(-)
net/mptcp/pm_netlink.c +15 -9
@@ -9,6 +9,7 @@#include <linux/inet.h> #include <linux/kernel.h> #include <net/tcp.h> +#include <net/inet_common.h> #include <net/netns/generic.h> #include <net/mptcp.h> #include <net/genetlink.h>
@@ -1002,8 +1003,8 @@ static int mptcp_pm_nl_create_listen_socket(struct sock *sk,bool is_ipv6 = sk->sk_family ==AF_INET6 ;int addrlen =sizeof (struct sockaddr_in );struct sockaddr_storage addr; +struct sock *newsk, *ssk;struct socket *ssock; -struct sock *newsk;int backlog =1024 ;int err;
@@ -1039,18 +1040,23 @@ static int mptcp_pm_nl_create_listen_socket(struct sock *sk,if (entry->addr.family ==AF_INET6 ) addrlen =sizeof (struct sockaddr_in6 );#endif - err =kernel_bind (ssock, (struct sockaddr *)&addr, addrlen); + ssk =mptcp_sk (newsk)->first; +if (ssk->sk_family ==AF_INET ) + err =inet_bind_sk (ssk, (struct sockaddr *)&addr, addrlen); +#if IS_ENABLED (CONFIG_MPTCP_IPV6 ) + else if (ssk->sk_family == AF_INET6) + err =inet6_bind_sk (ssk, (struct sockaddr *)&addr, addrlen); +#endif if (err)return err;inet_sk_state_store (newsk,TCP_LISTEN ); - err =kernel_listen (ssock, backlog); -if (err) -return err; - -mptcp_event_pm_listener (ssock->sk,MPTCP_EVENT_LISTENER_CREATED ); - -return 0 ; +lock_sock (ssk); + err =__inet_listen_sk (ssk, backlog); +if (!err) +mptcp_event_pm_listener (ssk,MPTCP_EVENT_LISTENER_CREATED ); +release_sock (ssk); +return err; }int mptcp_pm_nl_get_local_id (struct mptcp_sock *msk,struct mptcp_addr_info *skc)
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 13/14 mptcp: change the mpc check helper to return a sk
In reply to Paolo AbeniAfter the previous patch the __mptcp_nmpc_socket helper is used only to ensure that the MPTCP socket is a suitable status - that is, the mptcp capable handshake is not started yet.
Change the return value to the relevant subflow sock, to finally remove the last references to first subflow socket in the MPTCP stack.
As a bonus, we can get rid of a few local variables in different functions.
No functional change intended.
--- net/mptcp/pm_netlink.c | 8 +++----- net/mptcp/protocol.c | 40 +++++++++++++++------------------------ net/mptcp/protocol.h | 2 +- net/mptcp/sockopt.c | 43 +++++++++++++++++++----------------------- 4 files changed, 38 insertions(+), 55 deletions(-)
net/mptcp/pm_netlink.c +3 -5
@@ -1004,7 +1004,6 @@ static int mptcp_pm_nl_create_listen_socket(struct sock *sk,int addrlen =sizeof (struct sockaddr_in );struct sockaddr_storage addr;struct sock *newsk, *ssk; -struct socket *ssock;int backlog =1024 ;int err;
@@ -1030,17 +1029,16 @@ static int mptcp_pm_nl_create_listen_socket(struct sock *sk, &mptcp_keys[is_ipv6]);lock_sock (newsk); - ssock =__mptcp_nmpc_socket (mptcp_sk (newsk)); + ssk =__mptcp_nmpc_sk (mptcp_sk (newsk));release_sock (newsk); -if (IS_ERR (ssock)) -return PTR_ERR (ssock); +if (IS_ERR (ssk)) +return PTR_ERR (ssk);mptcp_info2sockaddr (&entry->addr, &addr, entry->addr.family);#if IS_ENABLED (CONFIG_MPTCP_IPV6 )if (entry->addr.family ==AF_INET6 ) addrlen =sizeof (struct sockaddr_in6 );#endif - ssk =mptcp_sk (newsk)->first;if (ssk->sk_family ==AF_INET ) err =inet_bind_sk (ssk, (struct sockaddr *)&addr, addrlen);#if IS_ENABLED (CONFIG_MPTCP_IPV6 )
net/mptcp/protocol.c +15 -25
@@ -108,7 +108,7 @@ static int __mptcp_socket_create(struct mptcp_sock *msk)/* If the MPC handshake is not started, returns the first subflow, * eventually allocating it. */ -struct socket *__mptcp_nmpc_socket (struct mptcp_sock *msk) +struct sock *__mptcp_nmpc_sk (struct mptcp_sock *msk) {struct sock *sk = (struct sock *)msk;int ret;
@@ -116,10 +116,7 @@ struct socket *__mptcp_nmpc_socket(struct mptcp_sock *msk)if (!((1 << sk->sk_state) & (TCPF_CLOSE |TCPF_LISTEN )))return ERR_PTR (-EINVAL ); -if (!msk->subflow) { -if (msk->first) -return ERR_PTR (-EINVAL ); - +if (!msk->first) { ret =__mptcp_socket_create (msk);if (ret)return ERR_PTR (ret);
@@ -127,7 +124,7 @@ struct socket *__mptcp_nmpc_socket(struct mptcp_sock *msk)mptcp_sockopt_sync (msk, msk->first); } -return msk->subflow; +return msk->first; }static void mptcp_drop (struct sock *sk,struct sk_buff *skb)
@@ -1671,7 +1668,6 @@ static int mptcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, {unsigned int saved_flags = msg->msg_flags;struct mptcp_sock *msk =mptcp_sk (sk); -struct socket *ssock;struct sock *ssk;int ret;
@@ -1682,9 +1678,9 @@ static int mptcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,* fastopen attempt, no need to check for additional subflow status. */ if (msg->msg_flags &MSG_FASTOPEN ) { - ssock =__mptcp_nmpc_socket (msk); -if (IS_ERR (ssock)) -return PTR_ERR (ssock); + ssk =__mptcp_nmpc_sk (msk); +if (IS_ERR (ssk)) +return PTR_ERR (ssk); }if (!msk->first)return -EINVAL ;
@@ -3620,16 +3616,14 @@ static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) {struct mptcp_subflow_context *subflow;struct mptcp_sock *msk =mptcp_sk (sk); -struct socket *ssock;int err = -EINVAL ;struct sock *ssk; - ssock =__mptcp_nmpc_socket (msk); -if (IS_ERR (ssock)) -return PTR_ERR (ssock); + ssk =__mptcp_nmpc_sk (msk); +if (IS_ERR (ssk)) +return PTR_ERR (ssk);inet_sk_state_store (sk,TCP_SYN_SENT ); - ssk = msk->first; subflow =mptcp_subflow_ctx (ssk);#ifdef CONFIG_TCP_MD5SIG /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
@@ -3724,17 +3718,15 @@ static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) {struct mptcp_sock *msk =mptcp_sk (sock->sk);struct sock *ssk, *sk = sock->sk; -struct socket *ssock;int err = -EINVAL ;lock_sock (sk); - ssock =__mptcp_nmpc_socket (msk); -if (IS_ERR (ssock)) { - err =PTR_ERR (ssock); + ssk =__mptcp_nmpc_sk (msk); +if (IS_ERR (ssk)) { + err =PTR_ERR (ssk); goto unlock; } - ssk = msk->first;if (sk->sk_family ==AF_INET ) err =inet_bind_sk (ssk, uaddr, addr_len);#if IS_ENABLED (CONFIG_MPTCP_IPV6 )
@@ -3753,7 +3745,6 @@ static int mptcp_listen(struct socket *sock, int backlog) {struct mptcp_sock *msk =mptcp_sk (sock->sk);struct sock *sk = sock->sk; -struct socket *ssock;struct sock *ssk;int err;
@@ -3765,13 +3756,12 @@ static int mptcp_listen(struct socket *sock, int backlog)if (sock->state !=SS_UNCONNECTED || sock->type !=SOCK_STREAM ) goto unlock; - ssock =__mptcp_nmpc_socket (msk); -if (IS_ERR (ssock)) { - err =PTR_ERR (ssock); + ssk =__mptcp_nmpc_sk (msk); +if (IS_ERR (ssk)) { + err =PTR_ERR (ssk); goto unlock; } - ssk = msk->first;inet_sk_state_store (sk,TCP_LISTEN );sock_set_flag (sk,SOCK_RCU_FREE );
net/mptcp/protocol.h +1 -1
@@ -636,7 +636,7 @@ void __mptcp_subflow_send_ack(struct sock *ssk);void mptcp_subflow_reset (struct sock *ssk);void mptcp_subflow_queue_clean (struct sock *sk,struct sock *ssk);void mptcp_sock_graft (struct sock *sk,struct socket *parent); -struct socket *__mptcp_nmpc_socket (struct mptcp_sock *msk); +struct sock *__mptcp_nmpc_sk (struct mptcp_sock *msk);bool __mptcp_close (struct sock *sk,long timeout);void mptcp_cancel_work (struct sock *sk);void __mptcp_unaccepted_force_close (struct sock *sk);
net/mptcp/sockopt.c +19 -24
113 lines not on this page · raw message
Signature · 1 line
2.41.0
Paolo Abeni
maintainerPATCH 14/14 mptcp: get rid of msk->subflow
In reply to Paolo AbeniSuch field is now unused just as a flag to control the first subflow deletion at close() time. Introduce a new bit flag for that and finally drop the mentioned field.
As an intended side effect, now the first subflow sock is not freed before close() even for passive sockets. The msk has no open/active subflows if the first one is closed and the subflow list is singular, update accordingly the state check in mptcp_stream_accept().
Among other benefits, the subflow removal, reduces the amount of memory used on the client side for each mptcp connection, allows passive sockets to go through successful accept()/disconnect()/connect() and makes return error code consistent for failing both passive and active sockets.
---
Side notes:
- syzkaller will be likely happy about the new code path to possibly exploit
- we could possibly avoid allocating the 'socket' struct at __mptcp_subflow_connect() time, but that will require more invasive helpers creation in inet core.
--- net/mptcp/protocol.c | 22 +++++----------------- net/mptcp/protocol.h | 13 ++++++------- 2 files changed, 11 insertions(+), 24 deletions(-)
net/mptcp/protocol.c +5 -17
@@ -91,7 +91,6 @@ static int __mptcp_socket_create(struct mptcp_sock *msk)return err;WRITE_ONCE (msk->first, ssock->sk); -WRITE_ONCE (msk->subflow, ssock); subflow =mptcp_subflow_ctx (ssock->sk);list_add (&subflow->node, &msk->conn_list);sock_hold (ssock->sk);
@@ -101,6 +100,7 @@ static int __mptcp_socket_create(struct mptcp_sock *msk)/* This is the first subflow, always with id 0 */ subflow->local_id_valid =1 ;mptcp_sock_graft (msk->first, sk->sk_socket); +iput (SOCK_INODE (ssock));return 0 ; }
@@ -2263,14 +2263,6 @@ struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)return min_stale_count >1 ? backup :NULL ; } -static void mptcp_dispose_initial_subflow (struct mptcp_sock *msk) -{ -if (msk->subflow) { -iput (SOCK_INODE (msk->subflow)); -WRITE_ONCE (msk->subflow,NULL ); - } -} -bool __mptcp_retransmit_pending_data (struct sock *sk) {struct mptcp_data_frag *cur, *rtx_head;
@@ -2349,7 +2341,7 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk, goto out_release; } - dispose_it = !msk->subflow || ssk != msk->subflow->sk; + dispose_it = msk->free_first || ssk != msk->first;if (dispose_it)list_del (&subflow->node);
@@ -2370,7 +2362,6 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,* disconnect should never fail */ WARN_ON_ONCE (tcp_disconnect (ssk,0 )); - msk->subflow->state =SS_UNCONNECTED ;mptcp_subflow_ctx_reset (subflow);release_sock (ssk);
@@ -3147,7 +3138,6 @@ struct sock *mptcp_sk_clone_init(const struct sock *sk, msk =mptcp_sk (nsk); msk->local_key = subflow_req->local_key; msk->token = subflow_req->token; -WRITE_ONCE (msk->subflow,NULL ); msk->in_accept_queue =1 ;WRITE_ONCE (msk->fully_established, false);if (mp_opt->suboptions &OPTION_MPTCP_CSUMREQD )
@@ -3285,10 +3275,8 @@ static void mptcp_destroy(struct sock *sk) {struct mptcp_sock *msk =mptcp_sk (sk); -/* clears msk->subflow, allowing the following to close -* even the initial subflow -*/ -mptcp_dispose_initial_subflow (msk); +/* allow the following to close even the initial subflow */ + msk->free_first =1 ;mptcp_destroy_common (msk,0 );sk_sockets_allocated_dec (sk); }
@@ -3828,7 +3816,7 @@ static int mptcp_stream_accept(struct socket *sock, struct socket *newsock, unlikely(inet_sk_state_load(msk->first) == TCP_CLOSE)) {__mptcp_close_ssk (newsk, msk->first,mptcp_subflow_ctx (msk->first),0 ); -if (unlikely (list_empty (&msk->conn_list))) +if (unlikely (list_is_singular (&msk->conn_list)))inet_sk_state_store (newsk,TCP_CLOSE ); } }
net/mptcp/protocol.h +6 -7
@@ -297,7 +297,8 @@ struct mptcp_sock { cork:1, nodelay:1, fastopening:1, - in_accept_queue:1; + in_accept_queue:1, + free_first:1;struct work_struct work;struct sk_buff *ooo_last_skb;struct rb_root out_of_order_queue;
@@ -306,12 +307,10 @@ struct mptcp_sock {struct list_head rtx_queue;struct mptcp_data_frag *first_pending;struct list_head join_list; -struct socket *subflow;/* outgoing connect/listener/!mp_capable -* The mptcp ops can safely dereference, using suitable -* ONCE annotation, the subflow outside the socket -* lock as such sock is freed after close(). -*/ -struct sock *first; +struct sock *first;/* The mptcp ops can safely dereference, using suitable +* ONCE annotation, the subflow outside the socket +* lock as such sock is freed after close(). +*/ struct mptcp_pm_data pm;struct mptcp_sched_ops *sched;struct {
Signature · 1 line
2.41.0
Re: mptcp: get rid of msk->subflow: Tests Results
Hi Paolo,
Thank you for your modifications, that's great!
Our CI did some validations and here is its report:
- KVM Validation: normal (except selftest_mptcp_join):
- Success! ✅:
- Task: https://cirrus-ci.com/task/6440442267435008
- Summary: https://api.cirrus-ci.com/v1/artifact/task/6440442267435008/summary/summary.txt
- KVM Validation: debug (except selftest_mptcp_join):
- Success! ✅:
- Task: https://cirrus-ci.com/task/6158967290724352
- Summary: https://api.cirrus-ci.com/v1/artifact/task/6158967290724352/summary/summary.txt
- KVM Validation: debug (only selftest_mptcp_join):
- Unstable: 1 failed test(s): selftest_mptcp_join 🔴:
- Task: https://cirrus-ci.com/task/5596017337303040
- Summary: https://api.cirrus-ci.com/v1/artifact/task/5596017337303040/summary/summary.txt
- KVM Validation: normal (only selftest_mptcp_join):
- Success! ✅:
- Task: https://cirrus-ci.com/task/5033067383881728
- Summary: https://api.cirrus-ci.com/v1/artifact/task/5033067383881728/summary/summary.txt
Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/917c23c6ef4f
If there are some issues, you can reproduce them using the same environment as the one used by the CI thanks to a docker image, e.g.:
$ cd [kernel source code]
$ docker run -v "${PWD}:${PWD}:rw" -w "${PWD}" --privileged --rm -it \
--pull always mptcp/mptcp-upstream-virtme-docker:latest \
auto-debugFor more details:
https://github.com/multipath-tcp/mptcp-upstream-virtme-docker
Please note that despite all the efforts that have been already done to have a stable tests suite when executed on a public CI like here, it is possible some reported issues are not due to your modifications. Still, do not hesitate to help us improve that ;-)
Cheers,
MPTCP GH Action bot
Bot operated by Matthieu Baerts (Tessares)
Mat Martineau
maintainerRe: [PATCH mptcp-next 03/14] mptcp: avoid additional __inet_stream_connect() call
In reply to Paolo AbeniQuoted 4 lines · Paolo Abeni
The mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is
__inet_stream_connect().
Hi Paolo -
Thanks for the series! It will be good to get rid of the confusing subflow pointer.
Quoted 2 lines
We are going to remove the first subflow socket soon, so avoid the addictional indirection via at connect time, calling directly
^^^^^^^^^^^
"additional" is misspelled this way in many of the series commit messages.
Quoted 58 lines
into the sock-level
connect()ops.No functional change intended.
Signed-off-by Paolo Abeni <[email protected]>---
net/mptcp/protocol.c| 40 +++++++++++++++++++++++++++++-----------1 file changed, 29 insertions(+), 11 deletions(-)net/mptcp/protocol.c +13
-9@@ -3629,22 +3629,24 @@ static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)struct mptcp_sock *msk =mptcp_sk (sk);struct socket *ssock;int err = -EINVAL ; +struct sock *ssk; ssock =__mptcp_nmpc_socket (msk);if (IS_ERR (ssock))return PTR_ERR (ssock);inet_sk_state_store (sk,TCP_SYN_SENT ); - subflow =mptcp_subflow_ctx (ssock->sk); + ssk = msk->first; + subflow =mptcp_subflow_ctx (ssk);#ifdef CONFIG_TCP_MD5SIG /* no MPTCP if MD5SIG is enabled on this socket or we may run out of * TCP option space. */ -if (rcu_access_pointer (tcp_sk (ssock->sk)->md5sig_info)) +if (rcu_access_pointer (tcp_sk (ssk)->md5sig_info))mptcp_subflow_early_fallback (msk, subflow);#endif -if (subflow->request_mptcp &&mptcp_token_new_connect (ssock->sk)) { -MPTCP_INC_STATS (sock_net (ssock->sk),MPTCP_MIB_TOKENFALLBACKINIT ); +if (subflow->request_mptcp &&mptcp_token_new_connect (ssk)) { +MPTCP_INC_STATS (sock_net (ssk),MPTCP_MIB_TOKENFALLBACKINIT );mptcp_subflow_early_fallback (msk, subflow); }if (likely (!__mptcp_check_fallback (msk)))@@ -3653,21 +3655,37 @@ static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)/* if reaching here via the fastopen/sendmsg path, the caller already * acquired the subflow socket lock, too. */ -if (msk->fastopening) - err =__inet_stream_connect (ssock, uaddr, addr_len,O_NONBLOCK ,1 ); -else - err =inet_stream_connect (ssock, uaddr, addr_len,O_NONBLOCK ); -inet_sk (sk)->defer_connect =inet_sk (ssock->sk)->defer_connect; +if (!msk->fastopening) +lock_sock (ssk); + +if (ssk->sk_state !=TCP_CLOSE ) + goto out; + +if (BPF_CGROUP_PRE_CONNECT_ENABLED (ssk)) {
net/mptcp/protocol.c static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) · line 3674
Quoted 10 lines
net/mptcp/protocol.c +10
-0+ err = ssk->sk_prot->pre_connect (ssk, uaddr, addr_len); +if (err) + goto out; + } + + err = ssk->sk_prot->connect (ssk, uaddr, addr_len); +if (err <0 ) + goto out; + +inet_sk (sk)->defer_connect =inet_sk (ssk)->defer_connect;
The above code doesn't do everything __inet_stream_connect() does. Is that code omitted here because the caller of this function handles the timeouts and msk-level socket states already?
- Mat
Quoted 24 lines
net/mptcp/protocol.c +6
-2+ +out: +if (!msk->fastopening) +release_sock (ssk);/* on successful connect, the msk state will be moved to established by * subflow_finish_connect() */ if (unlikely (err && err != -EINPROGRESS )) { -inet_sk_state_store (sk,inet_sk_state_load (ssock->sk)); +inet_sk_state_store (sk,inet_sk_state_load (ssk));return err; } -mptcp_copy_inaddrs (sk, ssock->sk); +mptcp_copy_inaddrs (sk, ssk);/* silence EINPROGRESS and let the caller inet_stream_connect * handle the connection in progress Signature · 1 line
2.41.0
Mat Martineau
maintainerRe: [PATCH mptcp-next 07/14] net: factor out __inet_listen_sk() helper
In reply to Paolo AbeniQuoted 74 lines · Paolo Abeni
The mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is
inet_listen().Factor out an helper operating directly on the (locked) struct sock, to allow get rid of the above dependency in the next patch without duplicating the existing code.
No functional changes intended.
Signed-off-by Paolo Abeni <[email protected]>---
include/net/inet_common.h| 1 +net/ipv4/af_inet.c| 39 +++++++++++++++++++++++----------------2 files changed, 24 insertions(+), 16 deletions(-)include/net/inet_common.h +1
-0@@ -40,6 +40,7 @@ int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, int flags);int inet_shutdown (struct socket *sock,int how);int inet_listen (struct socket *sock,int backlog); +int __inet_listen_sk (struct sock *sk,int backlog);void inet_sock_destruct (struct sock *sk);int inet_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len);int inet_bind_sk (struct sock *sk,struct sockaddr *uaddr,int addr_len);net/ipv4/af_inet.c +9
-16@@ -187,24 +187,13 @@ static int inet_autobind(struct sock *sk)return 0 ; } -/* -* Move a socket into listening state. -*/ -int inet_listen (struct socket *sock,int backlog) +int __inet_listen_sk (struct sock *sk,int backlog) { -struct sock *sk = sock->sk; -unsigned char old_state; +unsigned char old_state = sk->sk_state;int err, tcp_fastopen; -lock_sock (sk); - - err = -EINVAL ; -if (sock->state !=SS_UNCONNECTED || sock->type !=SOCK_STREAM ) - goto out; - - old_state = sk->sk_state;if (!((1 << old_state) & (TCPF_CLOSE |TCPF_LISTEN ))) - goto out; +return -EINVAL ;WRITE_ONCE (sk->sk_max_ack_backlog, backlog);/* Really, if the socket is already in listen state @@ -227,10 +216,28 @@ int inet_listen(struct socket *sock, int backlog) err =inet_csk_listen_start (sk);if (err) - goto out; +return err; +tcp_call_bpf (sk,BPF_SOCK_OPS_TCP_LISTEN_CB ,0 ,NULL ); } - err =0 ; +return 0 ; +} + +/*
net/ipv4/af_inet.c int inet_listen(struct socket *sock, int backlog) · line 236
Quoted 10 lines
net/ipv4/af_inet.c +10
-0+* Move a socket into listening state. +*/ +int inet_listen (struct socket *sock,int backlog) +{ +struct sock *sk = sock->sk; +int err; + +lock_sock (sk); + + err = -EINVAL ;
Minor tweak: can initialize in the declaration above.
Quoted 12 lines
net/ipv4/af_inet.c +4
-0+if (sock->state !=SS_UNCONNECTED || sock->type !=SOCK_STREAM ) + goto out; + + err =__inet_listen_sk (sk, backlog);out:
release_sock(sk);Signature · 1 line
2.41.0
Mat Martineau
maintainerRe: [PATCH mptcp-next 08/14] mptcp: avoid additional indirection in mptcp_listen()
In reply to Paolo AbeniQuoted 30 lines · Paolo Abeni
We are going to remove the first subflow socket soon, so avoid the addictional indirection via at
listen()time. Instead call directly the recently introduced helper on the first subflow sock.No functional changes intended.
Signed-off-by Paolo Abeni <[email protected]>---
net/mptcp/protocol.c| 14 ++++++++++----1 file changed, 10 insertions(+), 4 deletions(-)net/mptcp/protocol.c +2
-0@@ -3759,6 +3759,7 @@ static int mptcp_listen(struct socket *sock, int backlog)struct mptcp_sock *msk =mptcp_sk (sock->sk);struct sock *sk = sock->sk;struct socket *ssock; +struct sock *ssk;int err;pr_debug ("msk=%p" , msk);@@ -3775,15 +3776,20 @@ static int mptcp_listen(struct socket *sock, int backlog) goto unlock; } + ssk = msk->first;inet_sk_state_store (sk,TCP_LISTEN );sock_set_flag (sk,SOCK_RCU_FREE );
net/mptcp/protocol.c static int mptcp_listen(struct socket *sock, int backlog) · line 3789
Quoted 10 lines
net/mptcp/protocol.c +6
-2- err = ssock->ops->listen (ssock, backlog); -inet_sk_state_store (sk,inet_sk_state_load (ssock->sk)); +lock_sock (ssk); + err =__inet_listen_sk (ssk, backlog); +release_sock (ssk); +inet_sk_state_store (sk,inet_sk_state_load (ssk)); +if (!err) { +WRITE_ONCE (sk->sk_max_ack_backlog, backlog);
This seems like new functionality that isn't mentioned in the commit message. Do the changes here expose an issue with making the msk-level sk_max_ack_backlog match the subflow, or is this an unrelated fix?
Quoted 13 lines
net/mptcp/protocol.c +2
-2sock_prot_inuse_add (sock_net (sk), sk->sk_prot,1 ); -mptcp_copy_inaddrs (sk, ssock->sk); -mptcp_event_pm_listener (ssock->sk,MPTCP_EVENT_LISTENER_CREATED ); +mptcp_copy_inaddrs (sk, ssk); +mptcp_event_pm_listener (ssk,MPTCP_EVENT_LISTENER_CREATED ); }unlock:
Signature · 1 line
2.41.0
Mat Martineau
maintainerRe: [PATCH mptcp-next 10/14] mptcp: avoid unneeded indirection in mptcp_stream_accept()
In reply to Paolo AbeniQuoted 19 lines · Paolo Abeni
We are going to remove the first subflow socket soon, so avoid the addictional indirection at
accept()time. Instead access directly the first subflow sock.No functional changes intended.
Signed-off-by Paolo Abeni <[email protected]>---
net/mptcp/protocol.c| 17 +++++------------1 file changed, 5 insertions(+), 12 deletions(-)net/mptcp/protocol.c +0
-0@@ -3220,17 +3220,12 @@ static struct sock *mptcp_accept(struct sock *sk, int flags, int *err, bool kern) {struct mptcp_sock *msk =mptcp_sk (sk);
net/mptcp/protocol.c static struct sock *mptcp_accept(struct sock *sk, int flags, int *err, · line 3225
Quoted 10 lines
net/mptcp/protocol.c +2
-7-struct socket *listener; -struct sock *newsk; +struct sock *ssk, *newsk; - listener =READ_ONCE (msk->subflow); -if (WARN_ON_ONCE (!listener)) { - *err = -EINVAL ; -return NULL ; - } + ssk =READ_ONCE (msk->first);
There's a check for NULL msk->first in mptcp_stream_accept() below, but the analogous NULL check in this function has been removed. Should this be checking? (not sure if we have test coverage for out-of-sequence accept() calls)
Quoted 31 lines
net/mptcp/protocol.c +3
-5-pr_debug ("msk=%p, listener=%p" , msk,mptcp_subflow_ctx (listener->sk)); - newsk =inet_csk_accept (listener->sk, flags, err, kern); +pr_debug ("msk=%p, listener=%p" , msk,mptcp_subflow_ctx (ssk)); + newsk =inet_csk_accept (ssk, flags, err, kern);if (!newsk)return NULL ;@@ -3801,7 +3796,6 @@ static int mptcp_stream_accept(struct socket *sock, struct socket *newsock, int flags, bool kern) {struct mptcp_sock *msk =mptcp_sk (sock->sk); -struct socket *ssock;struct sock *newsk;int err;@@ -3810,8 +3804,7 @@ static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,/* Buggy applications can call accept on socket states other then LISTEN * but no need to allocate the first subflow just to error out. */ - ssock =READ_ONCE (msk->subflow); -if (!ssock) +if (!READ_ONCE (msk->first))return -EINVAL ; newsk =mptcp_accept (sock->sk, flags, &err, kern);Signature · 1 line
2.41.0
Mat Martineau
maintainerQuoted 101 lines · Paolo Abeni
Such field is now unused just as a flag to control the first subflow deletion at
close()time. Introduce a new bit flag for that and finally drop the mentioned field.As an intended side effect, now the first subflow sock is not freed before
close()even for passive sockets. The msk has no open/active subflows if the first one is closed and the subflow list is singular, update accordingly the state check inmptcp_stream_accept().Among other benefits, the subflow removal, reduces the amount of memory used on the client side for each mptcp connection, allows passive sockets to go through successful accept()/disconnect()/connect() and makes return error code consistent for failing both passive and active sockets.
Signed-off-by Paolo Abeni <[email protected]>---
Side notes:
- syzkaller will be likely happy about the new code path to possibly exploit
- we could possibly avoid allocating the 'socket' struct at__mptcp_subflow_connect()time, but that will require more invasive helpers creation in inet core.
---net/mptcp/protocol.c| 22 +++++-----------------net/mptcp/protocol.h| 13 ++++++-------2 files changed, 11 insertions(+), 24 deletions(-)net/mptcp/protocol.c +4
-16@@ -91,7 +91,6 @@ static int __mptcp_socket_create(struct mptcp_sock *msk)return err;WRITE_ONCE (msk->first, ssock->sk); -WRITE_ONCE (msk->subflow, ssock); subflow =mptcp_subflow_ctx (ssock->sk);list_add (&subflow->node, &msk->conn_list);sock_hold (ssock->sk);@@ -101,6 +100,7 @@ static int __mptcp_socket_create(struct mptcp_sock *msk)/* This is the first subflow, always with id 0 */ subflow->local_id_valid =1 ;mptcp_sock_graft (msk->first, sk->sk_socket); +iput (SOCK_INODE (ssock));return 0 ; }@@ -2263,14 +2263,6 @@ struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)return min_stale_count >1 ? backup :NULL ; } -static void mptcp_dispose_initial_subflow (struct mptcp_sock *msk) -{ -if (msk->subflow) { -iput (SOCK_INODE (msk->subflow)); -WRITE_ONCE (msk->subflow,NULL ); - } -} -bool __mptcp_retransmit_pending_data (struct sock *sk) {struct mptcp_data_frag *cur, *rtx_head;@@ -2349,7 +2341,7 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk, goto out_release; } - dispose_it = !msk->subflow || ssk != msk->subflow->sk; + dispose_it = msk->free_first || ssk != msk->first;if (dispose_it)list_del (&subflow->node);@@ -2370,7 +2362,6 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,* disconnect should never fail */ WARN_ON_ONCE (tcp_disconnect (ssk,0 )); - msk->subflow->state =SS_UNCONNECTED ;mptcp_subflow_ctx_reset (subflow);release_sock (ssk);@@ -3147,7 +3138,6 @@ struct sock *mptcp_sk_clone_init(const struct sock *sk, msk =mptcp_sk (nsk); msk->local_key = subflow_req->local_key; msk->token = subflow_req->token; -WRITE_ONCE (msk->subflow,NULL ); msk->in_accept_queue =1 ;WRITE_ONCE (msk->fully_established, false);if (mp_opt->suboptions &OPTION_MPTCP_CSUMREQD )@@ -3285,10 +3275,8 @@ static void mptcp_destroy(struct sock *sk) {struct mptcp_sock *msk =mptcp_sk (sk); -/* clears msk->subflow, allowing the following to close -* even the initial subflow -*/ -mptcp_dispose_initial_subflow (msk); +/* allow the following to close even the initial subflow */ + msk->free_first =1 ;mptcp_destroy_common (msk,0 );sk_sockets_allocated_dec (sk); }
net/mptcp/protocol.c static int mptcp_stream_accept(struct socket *sock, struct socket *newsock, · line 3819
Quoted 5 lines
net/mptcp/protocol.c +1
-1@@ -3828,7 +3816,7 @@ static int mptcp_stream_accept(struct socket *sock, struct socket *newsock, unlikely(inet_sk_state_load(msk->first) == TCP_CLOSE)) {__mptcp_close_ssk (newsk, msk->first,mptcp_subflow_ctx (msk->first),0 ); -if (unlikely (list_empty (&msk->conn_list))) +if (unlikely (list_is_singular (&msk->conn_list)))
There are other checks for an empty conn_list in the PM code. Do we need a new helper to check for "no open/active subflows", if list_empty() is no longer the correct thing?
Quoted 39 lines
net/mptcp/protocol.c +0
-0inet_sk_state_store (newsk,TCP_CLOSE ); } }net/mptcp/protocol.h +6
-7@@ -297,7 +297,8 @@ struct mptcp_sock { cork:1, nodelay:1, fastopening:1, - in_accept_queue:1; + in_accept_queue:1, + free_first:1;struct work_struct work;struct sk_buff *ooo_last_skb;struct rb_root out_of_order_queue;@@ -306,12 +307,10 @@ struct mptcp_sock {struct list_head rtx_queue;struct mptcp_data_frag *first_pending;struct list_head join_list; -struct socket *subflow;/* outgoing connect/listener/!mp_capable -* The mptcp ops can safely dereference, using suitable -* ONCE annotation, the subflow outside the socket -* lock as such sock is freed after close(). -*/ -struct sock *first; +struct sock *first;/* The mptcp ops can safely dereference, using suitable +* ONCE annotation, the subflow outside the socket +* lock as such sock is freed after close(). +*/ struct mptcp_pm_data pm;struct mptcp_sched_ops *sched;struct {Signature · 1 line
2.41.0
Re: [PATCH mptcp-next 03/14] mptcp: avoid additional __inet_stream_connect() call
In reply to Mat MartineauHi Paolo, Mat,
Thank you for the series and the reviews!
Quoted 5 lines · Mat Martineau
Quoted 3 lines · Paolo Abeni
The mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is
__inet_stream_connect().
(...)
Quoted 24 lines
Quoted 24 lines
net/mptcp/protocol.c +13
-5@@ -3653,21 +3655,37 @@ static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) /* if reaching here via the fastopen/sendmsg path, the caller already* acquired the subflow socket lock, too. */ - if (msk->fastopening) - err = __inet_stream_connect(ssock, uaddr, addr_len, O_NONBLOCK, 1); - else - err = inet_stream_connect(ssock, uaddr, addr_len, O_NONBLOCK); - inet_sk(sk)->defer_connect = inet_sk(ssock->sk)->defer_connect; + if (!msk->fastopening) + lock_sock(ssk); + + if (ssk->sk_state != TCP_CLOSE) + goto out; + + if (BPF_CGROUP_PRE_CONNECT_ENABLED(ssk)) { + err = ssk->sk_prot->pre_connect(ssk, uaddr, addr_len); + if (err) + goto out; + } + + err = ssk->sk_prot->connect(ssk, uaddr, addr_len);
Quoted 8 lines
net/mptcp/protocol.c static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) · line 3674
Quoted 4 lines
net/mptcp/protocol.c +4
-0+ if (err < 0) + goto out; + + inet_sk(sk)->defer_connect = inet_sk(ssk)->defer_connect;The above code doesn't do everything
__inet_stream_connect()does. Is that code omitted here because the caller of this function handles the timeouts and msk-level socket states already?
Is it not possible to extract the code we need from __inet_stream_connect() to a new function in af_inet.c? Then we can call this new function from __inet_stream_connect() and here.
I'm always a bit worry when we duplicate code from upper layers because we could miss future modifications done in the original code (we had that a few times with the fork and it is hard to track and maintain). I would say that the minimum is at least to add a comment stating this code is a duplication from somewhere else, just in case to make it clear where we need to look at in case of issues there.
Cheers,
Matt
Signature · 2 lines
Tessares | Belgium | Hybrid Access Solutions www.tessares.net
Re: [PATCH mptcp-next 01/14] mptcp: more accurate NL event generation.
In reply to Paolo AbeniHi Paolo, Mat,
Quoted 3 lines · Paolo Abeni
Currently the mptcp code generate a "new listener" event even if the actual
listen()syscall fails. Address the issue moving the event generation call under the successful branch.
Good catch!
Quoted 6 lines
Signed-off-by Paolo Abeni <[email protected]>---
this could possibly go via -net, but is used by later patches and is not really critical IMHO.
Eventually the additional tag would be:Fixes f8c9dfbd875b ("mptcp: add pm listener events")
I think it makes sense to send this to -net. If it is OK for Mat, I can already apply this patch.
Cheers,
Matt
Signature · 2 lines
Tessares | Belgium | Hybrid Access Solutions www.tessares.net
Mat Martineau
maintainerRe: [PATCH mptcp-next 01/14] mptcp: more accurate NL event generation.
Quoted 16 lines · Matthieu Baerts
Hi Paolo, Mat,
Quoted 3 lines · Paolo Abeni
Currently the mptcp code generate a "new listener" event even if the actual
listen()syscall fails. Address the issue moving the event generation call under the successful branch.Good catch!
Quoted 6 lines
Signed-off-by Paolo Abeni <[email protected]>---
this could possibly go via -net, but is used by later patches and is not really critical IMHO.
Eventually the additional tag would be:Fixes f8c9dfbd875b ("mptcp: add pm listener events")
Quoted 2 lines
I think it makes sense to send this to -net. If it is OK for Mat, I can already apply this patch.
Sure, fine to apply this one.
- Mat
Re: [PATCH mptcp-next 01/14] mptcp: more accurate NL event generation.
In reply to Paolo AbeniQuoted 11 lines · Paolo Abeni
Currently the mptcp code generate a "new listener" event even if the actual
listen()syscall fails. Address the issue moving the event generation call under the successful branch.Signed-off-by Paolo Abeni <[email protected]>---
this could possibly go via -net, but is used by later patches and is not really critical IMHO.
Eventually the additional tag would be:Fixes f8c9dfbd875b ("mptcp: add pm listener events")
I replied deeper in the thread about applying this patch to -net, but I need to tag it too:
Quoted 25 lines
---
net/mptcp/protocol.c| 3 +--1 file changed, 1 insertion(+), 2 deletions(-)net/mptcp/protocol.c +1
-2@@ -3762,10 +3762,9 @@ static int mptcp_listen(struct socket *sock, int backlog)if (!err) {sock_prot_inuse_add (sock_net (sk), sk->sk_prot,1 );mptcp_copy_inaddrs (sk, ssock->sk); +mptcp_event_pm_listener (ssock->sk,MPTCP_EVENT_LISTENER_CREATED ); } -mptcp_event_pm_listener (ssock->sk,MPTCP_EVENT_LISTENER_CREATED ); -unlock:
release_sock(sk); return err;Signature · 1 line
2.41.0
Re: mptcp: get rid of msk->subflow: Tests Results
In reply to Paolo AbeniHi Paolo,
Thank you for your modifications, that's great!
Our CI did some validations and here is its report:
- KVM Validation: normal (except selftest_mptcp_join):
- Success! ✅:
- Task: https://cirrus-ci.com/task/6122328535334912
- Summary: https://api.cirrus-ci.com/v1/artifact/task/6122328535334912/summary/summary.txt
- KVM Validation: normal (only selftest_mptcp_join):
- Unstable: 1 failed test(s): selftest_mptcp_join 🔴:
- Task: https://cirrus-ci.com/task/4589019657404416
- Summary: https://api.cirrus-ci.com/v1/artifact/task/4589019657404416/summary/summary.txt
- KVM Validation: debug (except selftest_mptcp_join):
- Success! ✅:
- Task: https://cirrus-ci.com/task/6685278488756224
- Summary: https://api.cirrus-ci.com/v1/artifact/task/6685278488756224/summary/summary.txt
- KVM Validation: debug (only selftest_mptcp_join):
- Success! ✅:
- Task: https://cirrus-ci.com/task/6048767019843584
- Summary: https://api.cirrus-ci.com/v1/artifact/task/6048767019843584/summary/summary.txt
Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/6f392039115d
If there are some issues, you can reproduce them using the same environment as the one used by the CI thanks to a docker image, e.g.:
$ cd [kernel source code]
$ docker run -v "${PWD}:${PWD}:rw" -w "${PWD}" --privileged --rm -it \
--pull always mptcp/mptcp-upstream-virtme-docker:latest \
auto-debugFor more details:
https://github.com/multipath-tcp/mptcp-upstream-virtme-docker
Please note that despite all the efforts that have been already done to have a stable tests suite when executed on a public CI like here, it is possible some reported issues are not due to your modifications. Still, do not hesitate to help us improve that ;-)
Cheers,
MPTCP GH Action bot
Bot operated by Matthieu Baerts (Tessares)
Re: [PATCH mptcp-next 01/14] mptcp: more accurate NL event generation.
In reply to Mat MartineauHi Paolo, Mat,
Quoted 14 lines · Mat Martineau
Quoted 11 lines · Paolo Abeni
Currently the mptcp code generate a "new listener" event even if the actual
listen()syscall fails. Address the issue moving the event generation call under the successful branch.Signed-off-by Paolo Abeni <[email protected]>---
this could possibly go via -net, but is used by later patches and is not really critical IMHO.
Eventually the additional tag would be:Fixes f8c9dfbd875b ("mptcp: add pm listener events")
Quoted 4 lines
I replied deeper in the thread about applying this patch to -net, but I need to tag it too:
Reviewed-by Mat Martineau <[email protected]>
Thank you for the patch and the review!
Now in our tree (fixes for -net), only patch 1/14, with the Fixes tag:
New patches for t/upstream-net and t/upstream:
- 3e47bbe23f1c: mptcp: more accurate NL event generation
- Results: a99957c74d24..2f898c1f1eda (export-net)
- Results: 3ee02a7d38ae..14584832110b (export)
Tests are now in progress:
https://cirrus-ci.com/github/multipath-tcp/mptcp_net-next/export-net/20230713T191154
https://cirrus-ci.com/github/multipath-tcp/mptcp_net-next/export/20230713T191154
Cheers,
Matt
Signature · 2 lines
Tessares | Belgium | Hybrid Access Solutions www.tessares.net
Paolo Abeni
maintainerRe: [PATCH mptcp-next 03/14] mptcp: avoid additional __inet_stream_connect() call
In reply to Mat MartineauQuoted 9 lines · Mat Martineau
Quoted 4 lines · Paolo Abeni
The mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is
__inet_stream_connect().Hi Paolo -
Quoted 8 lines
Thanks for the series! It will be good to get rid of the confusing subflow pointer.
Quoted 2 lines
We are going to remove the first subflow socket soon, so avoid the addictional indirection via at connect time, calling directly
^^^^^^^^^^^"additional" is misspelled this way in many of the series commit messages.
Oops, I guess that the "many" above burns my usual "is a typo" excuse;)
Quoted 51 lines
Quoted 51 lines
net/mptcp/protocol.c +19
-9@@ -3629,22 +3629,24 @@ static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)struct mptcp_sock *msk =mptcp_sk (sk);struct socket *ssock;int err = -EINVAL ; +struct sock *ssk; ssock =__mptcp_nmpc_socket (msk);if (IS_ERR (ssock))return PTR_ERR (ssock);inet_sk_state_store (sk,TCP_SYN_SENT ); - subflow =mptcp_subflow_ctx (ssock->sk); + ssk = msk->first; + subflow =mptcp_subflow_ctx (ssk);#ifdef CONFIG_TCP_MD5SIG /* no MPTCP if MD5SIG is enabled on this socket or we may run out of * TCP option space. */ -if (rcu_access_pointer (tcp_sk (ssock->sk)->md5sig_info)) +if (rcu_access_pointer (tcp_sk (ssk)->md5sig_info))mptcp_subflow_early_fallback (msk, subflow);#endif -if (subflow->request_mptcp &&mptcp_token_new_connect (ssock->sk)) { -MPTCP_INC_STATS (sock_net (ssock->sk),MPTCP_MIB_TOKENFALLBACKINIT ); +if (subflow->request_mptcp &&mptcp_token_new_connect (ssk)) { +MPTCP_INC_STATS (sock_net (ssk),MPTCP_MIB_TOKENFALLBACKINIT );mptcp_subflow_early_fallback (msk, subflow); }if (likely (!__mptcp_check_fallback (msk)))@@ -3653,21 +3655,37 @@ static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)/* if reaching here via the fastopen/sendmsg path, the caller already * acquired the subflow socket lock, too. */ -if (msk->fastopening) - err =__inet_stream_connect (ssock, uaddr, addr_len,O_NONBLOCK ,1 ); -else - err =inet_stream_connect (ssock, uaddr, addr_len,O_NONBLOCK ); -inet_sk (sk)->defer_connect =inet_sk (ssock->sk)->defer_connect; +if (!msk->fastopening) +lock_sock (ssk); + +if (ssk->sk_state !=TCP_CLOSE ) + goto out; + +if (BPF_CGROUP_PRE_CONNECT_ENABLED (ssk)) { + err = ssk->sk_prot->pre_connect (ssk, uaddr, addr_len); +if (err) + goto out; + } + + err = ssk->sk_prot->connect (ssk, uaddr, addr_len);
Quoted 8 lines
net/mptcp/protocol.c static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) · line 3674
Quoted 4 lines
net/mptcp/protocol.c +4
-0+if (err <0 ) + goto out; + +inet_sk (sk)->defer_connect =inet_sk (ssk)->defer_connect;The above code doesn't do everything
__inet_stream_connect()does. Is that code omitted here because the caller of this function handles the timeouts and msk-level socket states already?
Yes. The main point is that we had an inet_stream_connect() call in the existing/prior code because we did not have access to a more specific/constrained helper. Maintaining the struct socket state for the ssock will be useless soon - we are going to drop such struct. So the above should be all we really need.
Cheers,
Paolo
Paolo Abeni
maintainerRe: [PATCH mptcp-next 07/14] net: factor out __inet_listen_sk() helper
In reply to Mat MartineauQuoted 80 lines · Mat Martineau
Quoted 78 lines · Paolo Abeni
The mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is
inet_listen().Factor out an helper operating directly on the (locked) struct sock, to allow get rid of the above dependency in the next patch without duplicating the existing code.
No functional changes intended.
Signed-off-by Paolo Abeni <[email protected]>---
include/net/inet_common.h| 1 +net/ipv4/af_inet.c| 39 +++++++++++++++++++++++----------------2 files changed, 24 insertions(+), 16 deletions(-)include/net/inet_common.h +1
-0@@ -40,6 +40,7 @@ int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, int flags);int inet_shutdown (struct socket *sock,int how);int inet_listen (struct socket *sock,int backlog); +int __inet_listen_sk (struct sock *sk,int backlog);void inet_sock_destruct (struct sock *sk);int inet_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len);int inet_bind_sk (struct sock *sk,struct sockaddr *uaddr,int addr_len);net/ipv4/af_inet.c +13
-16@@ -187,24 +187,13 @@ static int inet_autobind(struct sock *sk)return 0 ; } -/* -* Move a socket into listening state. -*/ -int inet_listen (struct socket *sock,int backlog) +int __inet_listen_sk (struct sock *sk,int backlog) { -struct sock *sk = sock->sk; -unsigned char old_state; +unsigned char old_state = sk->sk_state;int err, tcp_fastopen; -lock_sock (sk); - - err = -EINVAL ; -if (sock->state !=SS_UNCONNECTED || sock->type !=SOCK_STREAM ) - goto out; - - old_state = sk->sk_state;if (!((1 << old_state) & (TCPF_CLOSE |TCPF_LISTEN ))) - goto out; +return -EINVAL ;WRITE_ONCE (sk->sk_max_ack_backlog, backlog);/* Really, if the socket is already in listen state @@ -227,10 +216,28 @@ int inet_listen(struct socket *sock, int backlog) err =inet_csk_listen_start (sk);if (err) - goto out; +return err; +tcp_call_bpf (sk,BPF_SOCK_OPS_TCP_LISTEN_CB ,0 ,NULL ); } - err =0 ; +return 0 ; +} + +/* +* Move a socket into listening state. +*/ +int inet_listen (struct socket *sock,int backlog) +{
Quoted 8 lines
net/ipv4/af_inet.c int inet_listen(struct socket *sock, int backlog) · line 236
Quoted 6 lines
net/ipv4/af_inet.c +6
-0+struct sock *sk = sock->sk; +int err; + +lock_sock (sk); + + err = -EINVAL ;Minor tweak: can initialize in the declaration above.
Right you are! Not sure how I missed it. I'll do in v2.
Thanks!
Paolo
Paolo Abeni
maintainerRe: [PATCH mptcp-next 08/14] mptcp: avoid additional indirection in mptcp_listen()
In reply to Mat MartineauQuoted 33 lines · Mat Martineau
Quoted 31 lines · Paolo Abeni
We are going to remove the first subflow socket soon, so avoid the addictional indirection via at
listen()time. Instead call directly the recently introduced helper on the first subflow sock.No functional changes intended.
Signed-off-by Paolo Abeni <[email protected]>---
net/mptcp/protocol.c| 14 ++++++++++----1 file changed, 10 insertions(+), 4 deletions(-)net/mptcp/protocol.c +2
-0@@ -3759,6 +3759,7 @@ static int mptcp_listen(struct socket *sock, int backlog)struct mptcp_sock *msk =mptcp_sk (sock->sk);struct sock *sk = sock->sk;struct socket *ssock; +struct sock *ssk;int err;pr_debug ("msk=%p" , msk);@@ -3775,15 +3776,20 @@ static int mptcp_listen(struct socket *sock, int backlog) goto unlock; } + ssk = msk->first;inet_sk_state_store (sk,TCP_LISTEN );sock_set_flag (sk,SOCK_RCU_FREE );
Quoted 13 lines
net/mptcp/protocol.c static int mptcp_listen(struct socket *sock, int backlog) · line 3789
Quoted 9 lines
net/mptcp/protocol.c +6
-2- err = ssock->ops->listen (ssock, backlog); -inet_sk_state_store (sk,inet_sk_state_load (ssock->sk)); +lock_sock (ssk); + err =__inet_listen_sk (ssk, backlog); +release_sock (ssk); +inet_sk_state_store (sk,inet_sk_state_load (ssk)); +if (!err) { +WRITE_ONCE (sk->sk_max_ack_backlog, backlog);This seems like new functionality that isn't mentioned in the commit message. Do the changes here expose an issue with making the msk-level
sk_max_ack_backlogmatch the subflow, or is this an unrelated fix?
You are right, I should have move this to a different patch.
I added this just for added consistency, but the field is never used and not even exposed to the user-space (diag reads the first subflow sk_max_ack_backlog). I'll drop the chunk in v2.
Thanks!
Paolo
Paolo Abeni
maintainerRe: [PATCH mptcp-next 03/14] mptcp: avoid additional __inet_stream_connect() call
In reply to Matthieu BaertsQuoted 46 lines · Matthieu Baerts
Hi Paolo, Mat,
Thank you for the series and the reviews!
Quoted 5 lines · Mat Martineau
Quoted 3 lines · Paolo Abeni
The mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is
__inet_stream_connect().(...)
Quoted 24 lines
Quoted 24 lines
net/mptcp/protocol.c +13
-5@@ -3653,21 +3655,37 @@ static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) /* if reaching here via the fastopen/sendmsg path, the caller already* acquired the subflow socket lock, too. */ - if (msk->fastopening) - err = __inet_stream_connect(ssock, uaddr, addr_len, O_NONBLOCK, 1); - else - err = inet_stream_connect(ssock, uaddr, addr_len, O_NONBLOCK); - inet_sk(sk)->defer_connect = inet_sk(ssock->sk)->defer_connect; + if (!msk->fastopening) + lock_sock(ssk); + + if (ssk->sk_state != TCP_CLOSE) + goto out; + + if (BPF_CGROUP_PRE_CONNECT_ENABLED(ssk)) { + err = ssk->sk_prot->pre_connect(ssk, uaddr, addr_len); + if (err) + goto out; + } + + err = ssk->sk_prot->connect(ssk, uaddr, addr_len);Quoted 8 lines
net/mptcp/protocol.c static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) · line 3674
Quoted 4 lines
net/mptcp/protocol.c +4
-0+ if (err < 0) + goto out; + + inet_sk(sk)->defer_connect = inet_sk(ssk)->defer_connect;The above code doesn't do everything
__inet_stream_connect()does. Is that code omitted here because the caller of this function handles the timeouts and msk-level socket states already?
Quoted 7 lines
Is it not possible to extract the code we need from
__inet_stream_connect()to a new function inaf_inet.c? Then we can call this new function from__inet_stream_connect()and here.I'm always a bit worry when we duplicate code from upper layers because we could miss future modifications done in the original code (we had that a few times with the fork and it is hard to track and maintain).
Well, that kind of problems becomes order of magnitude worse with out- of-tree code, because nobody is aware nor interested of such code needs.
Chances of issues here are really low:
* mptcp code is in tree
* maintainers are aware ;)
* bpf hooks should not change
* [most important topic] the dup code is extremely simple and small
All in all, it looks reasonably safe to me.
Quoted 4 lines
I
would say that the minimum is at least to add a comment stating this code is a duplication from somewhere else, just in case to make it clear where we need to look at in case of issues there.
I'll add the comment in v2.
Thanks,
Paolo
Paolo Abeni
maintainerRe: [PATCH mptcp-next 10/14] mptcp: avoid unneeded indirection in mptcp_stream_accept()
In reply to Mat MartineauQuoted 25 lines · Mat Martineau
Quoted 23 lines · Paolo Abeni
We are going to remove the first subflow socket soon, so avoid the addictional indirection at
accept()time. Instead access directly the first subflow sock.No functional changes intended.
Signed-off-by Paolo Abeni <[email protected]>---
net/mptcp/protocol.c| 17 +++++------------1 file changed, 5 insertions(+), 12 deletions(-)net/mptcp/protocol.c +1
-2@@ -3220,17 +3220,12 @@ static struct sock *mptcp_accept(struct sock *sk, int flags, int *err, bool kern) {struct mptcp_sock *msk =mptcp_sk (sk); -struct socket *listener; -struct sock *newsk; +struct sock *ssk, *newsk;
Quoted 11 lines
net/mptcp/protocol.c static struct sock *mptcp_accept(struct sock *sk, int flags, int *err, · line 3225
Quoted 6 lines
net/mptcp/protocol.c +1
-5- listener =READ_ONCE (msk->subflow); -if (WARN_ON_ONCE (!listener)) { - *err = -EINVAL ; -return NULL ; - } + ssk =READ_ONCE (msk->first);There's a check for NULL
msk->firstinmptcp_stream_accept()below, but the analogous NULL check in this function has been removed. Should this be checking? (not sure if we have test coverage for out-of-sequenceaccept()calls)
The existing check above was really a pedantic one. The caller (mptcp_stream_accept) already checked the same field under the same lock scope, so only random memory corruption could change it. Note that mptcp_accept() can be invoked only by mptcp_stream_accept().
I'll a note in the commit message to explain why we can drop the check.
Thanks,
Paolo
Paolo Abeni
maintainerQuoted 104 lines · Mat Martineau
Quoted 102 lines · Paolo Abeni
Such field is now unused just as a flag to control the first subflow deletion at
close()time. Introduce a new bit flag for that and finally drop the mentioned field.As an intended side effect, now the first subflow sock is not freed before
close()even for passive sockets. The msk has no open/active subflows if the first one is closed and the subflow list is singular, update accordingly the state check inmptcp_stream_accept().Among other benefits, the subflow removal, reduces the amount of memory used on the client side for each mptcp connection, allows passive sockets to go through successful accept()/disconnect()/connect() and makes return error code consistent for failing both passive and active sockets.
Signed-off-by Paolo Abeni <[email protected]>---
Side notes:
- syzkaller will be likely happy about the new code path to possibly exploit
- we could possibly avoid allocating the 'socket' struct at__mptcp_subflow_connect()time, but that will require more invasive helpers creation in inet core.
---net/mptcp/protocol.c| 22 +++++-----------------net/mptcp/protocol.h| 13 ++++++-------2 files changed, 11 insertions(+), 24 deletions(-)net/mptcp/protocol.c +4
-16@@ -91,7 +91,6 @@ static int __mptcp_socket_create(struct mptcp_sock *msk)return err;WRITE_ONCE (msk->first, ssock->sk); -WRITE_ONCE (msk->subflow, ssock); subflow =mptcp_subflow_ctx (ssock->sk);list_add (&subflow->node, &msk->conn_list);sock_hold (ssock->sk);@@ -101,6 +100,7 @@ static int __mptcp_socket_create(struct mptcp_sock *msk)/* This is the first subflow, always with id 0 */ subflow->local_id_valid =1 ;mptcp_sock_graft (msk->first, sk->sk_socket); +iput (SOCK_INODE (ssock));return 0 ; }@@ -2263,14 +2263,6 @@ struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)return min_stale_count >1 ? backup :NULL ; } -static void mptcp_dispose_initial_subflow (struct mptcp_sock *msk) -{ -if (msk->subflow) { -iput (SOCK_INODE (msk->subflow)); -WRITE_ONCE (msk->subflow,NULL ); - } -} -bool __mptcp_retransmit_pending_data (struct sock *sk) {struct mptcp_data_frag *cur, *rtx_head;@@ -2349,7 +2341,7 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk, goto out_release; } - dispose_it = !msk->subflow || ssk != msk->subflow->sk; + dispose_it = msk->free_first || ssk != msk->first;if (dispose_it)list_del (&subflow->node);@@ -2370,7 +2362,6 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,* disconnect should never fail */ WARN_ON_ONCE (tcp_disconnect (ssk,0 )); - msk->subflow->state =SS_UNCONNECTED ;mptcp_subflow_ctx_reset (subflow);release_sock (ssk);@@ -3147,7 +3138,6 @@ struct sock *mptcp_sk_clone_init(const struct sock *sk, msk =mptcp_sk (nsk); msk->local_key = subflow_req->local_key; msk->token = subflow_req->token; -WRITE_ONCE (msk->subflow,NULL ); msk->in_accept_queue =1 ;WRITE_ONCE (msk->fully_established, false);if (mp_opt->suboptions &OPTION_MPTCP_CSUMREQD )@@ -3285,10 +3275,8 @@ static void mptcp_destroy(struct sock *sk) {struct mptcp_sock *msk =mptcp_sk (sk); -/* clears msk->subflow, allowing the following to close -* even the initial subflow -*/ -mptcp_dispose_initial_subflow (msk); +/* allow the following to close even the initial subflow */ + msk->free_first =1 ;mptcp_destroy_common (msk,0 );sk_sockets_allocated_dec (sk); }@@ -3828,7 +3816,7 @@ static int mptcp_stream_accept(struct socket *sock, struct socket *newsock, unlikely(inet_sk_state_load(msk->first) == TCP_CLOSE)) {
Quoted 8 lines
net/mptcp/protocol.c static int mptcp_stream_accept(struct socket *sock, struct socket *newsock, · line 3819
Quoted 4 lines
net/mptcp/protocol.c +1
-1__mptcp_close_ssk (newsk, msk->first,mptcp_subflow_ctx (msk->first),0 ); -if (unlikely (list_empty (&msk->conn_list))) +if (unlikely (list_is_singular (&msk->conn_list)))There are other checks for an empty
conn_listin the PM code. Do we need a new helper to check for "no open/active subflows", iflist_empty()is no longer the correct thing?
Addendum: I audited the code and all the other places use list_empty() as an optimization to avoid acquiring additional locks: this is the only place we need to update.
Note that the generic helper will be less trivial:
bool mptcp_no_open_subflow(struct mptcp_sock *msk)
{
if (!msk->first)
return false;
if (inet_sk_state_load(msk->first) != TCP_CLOSE)
return false;
return list_is_singular(&msk->conn_list);
}but here we already did the first 2 checks. I'll postpone the helper creation to some other use-case.
Cheers,
Paolo
Re: [PATCH mptcp-next 03/14] mptcp: avoid additional __inet_stream_connect() call
In reply to Paolo AbeniHi Paolo,
Quoted 59 lines · Paolo Abeni
Quoted 46 lines · Matthieu Baerts
Hi Paolo, Mat,
Thank you for the series and the reviews!
Quoted 5 lines · Mat Martineau
Quoted 3 lines · Paolo Abeni
The mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is
__inet_stream_connect().(...)
Quoted 24 lines
Quoted 24 lines
net/mptcp/protocol.c +13
-5@@ -3653,21 +3655,37 @@ static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) /* if reaching here via the fastopen/sendmsg path, the caller already* acquired the subflow socket lock, too. */ - if (msk->fastopening) - err = __inet_stream_connect(ssock, uaddr, addr_len, O_NONBLOCK, 1); - else - err = inet_stream_connect(ssock, uaddr, addr_len, O_NONBLOCK); - inet_sk(sk)->defer_connect = inet_sk(ssock->sk)->defer_connect; + if (!msk->fastopening) + lock_sock(ssk); + + if (ssk->sk_state != TCP_CLOSE) + goto out; + + if (BPF_CGROUP_PRE_CONNECT_ENABLED(ssk)) { + err = ssk->sk_prot->pre_connect(ssk, uaddr, addr_len); + if (err) + goto out; + } + + err = ssk->sk_prot->connect(ssk, uaddr, addr_len);Quoted 8 lines
net/mptcp/protocol.c static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) · line 3674
Quoted 4 lines
net/mptcp/protocol.c +4
-0+ if (err < 0) + goto out; + + inet_sk(sk)->defer_connect = inet_sk(ssk)->defer_connect;The above code doesn't do everything
__inet_stream_connect()does. Is that code omitted here because the caller of this function handles the timeouts and msk-level socket states already?Quoted 7 lines
Is it not possible to extract the code we need from
__inet_stream_connect()to a new function inaf_inet.c? Then we can call this new function from__inet_stream_connect()and here.I'm always a bit worry when we duplicate code from upper layers because we could miss future modifications done in the original code (we had that a few times with the fork and it is hard to track and maintain).
Well, that kind of problems becomes order of magnitude worse with out- of-tree code, because nobody is aware nor interested of such code needs.
Quoted 7 lines
Chances of issues here are really low:
* mptcp code is in tree
* maintainers are aware ;)
* bpf hooks should not change
* [most important topic] the dup code is extremely simple and smallAll in all, it looks reasonably safe to me.
Thank you for the explanations!
If you think this is safe for you, that's fine for me :)
Cheers,
Matt
Signature · 2 lines
Tessares | Belgium | Hybrid Access Solutions www.tessares.net