setsockopt(mptcp_fd, SOL_SOCKET, ...)... appears to work (returns 0),
but it has no effect -- this is because the MPTCP layer never has a
chance to copy the settings to the subflow socket.
Skip the generic handling for the mptcp case and instead call the
mptcp specific handler instead for SOL_SOCKET too.
Next patch adds more specific handling for SOL_SOCKET to mptcp.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
When mptcp is used, userspace doesn't read from the tcp (subflow)
socket but from the parent (mptcp) socket receive queue.
skbs are moved from the subflow socket to the mptcp rx queue either from
'data_ready' callback (if mptcp socket can be locked), a work queue, or
the socket receive function.
This means tcp_rcv_space_adjust() is never called and thus no receive
buffer size auto-tuning is done.
An earlier (not merged) patch added tcp_rcv_space_adjust() calls to the
function that moves skbs from subflow to mptcp socket.
While this enabled autotuning, it also meant tuning was done even if
userspace was reading the mptcp socket very slowly.
This adds mptcp_rcv_space_adjust() and calls it after userspace has
read data from the mptcp socket rx queue.
Its very similar to tcp_rcv_space_adjust, with two differences:
1. The rtt estimate is the largest one observed on a subflow
2. The rcvbuf size and window clamp of all subflows is adjusted
to the mptcp-level rcvbuf.
Otherwise, we get spurious drops at tcp (subflow) socket level if
the skbs are not moved to the mptcp socket fast enough.
Before:
time mptcp_connect.sh -t -f $((4*1024*1024)) -d 300 -l 0.01% -r 0 -e "" -m mmap
[..]
ns4 MPTCP -> ns3 (10.0.3.2:10108 ) MPTCP (duration 40823ms) [ OK ]
ns4 MPTCP -> ns3 (10.0.3.2:10109 ) TCP (duration 23119ms) [ OK ]
ns4 TCP -> ns3 (10.0.3.2:10110 ) MPTCP (duration 5421ms) [ OK ]
ns4 MPTCP -> ns3 (dead:beef:3::2:10111) MPTCP (duration 41446ms) [ OK ]
ns4 MPTCP -> ns3 (dead:beef:3::2:10112) TCP (duration 23427ms) [ OK ]
ns4 TCP -> ns3 (dead:beef:3::2:10113) MPTCP (duration 5426ms) [ OK ]
Time: 1396 seconds
After:
ns4 MPTCP -> ns3 (10.0.3.2:10108 ) MPTCP (duration 5417ms) [ OK ]
ns4 MPTCP -> ns3 (10.0.3.2:10109 ) TCP (duration 5427ms) [ OK ]
ns4 TCP -> ns3 (10.0.3.2:10110 ) MPTCP (duration 5422ms) [ OK ]
ns4 MPTCP -> ns3 (dead:beef:3::2:10111) MPTCP (duration 5415ms) [ OK ]
ns4 MPTCP -> ns3 (dead:beef:3::2:10112) TCP (duration 5422ms) [ OK ]
ns4 TCP -> ns3 (dead:beef:3::2:10113) MPTCP (duration 5423ms) [ OK ]
Time: 296 seconds
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
mptcp_poll always return POLLOUT for unblocking
connect(), ensure that the socket is a suitable
state.
The MPTCP_DATA_READY bit is never cleared on accept:
ensure we don't leave mptcp_accept() with an empty
accept queue and such bit set.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently __mptcp_tcp_fallback() always return NULL
on incoming connections, because MPTCP does not create
the additional socket for the first subflow.
Since the previous commit no __mptcp_tcp_fallback()
caller needs a struct socket, so let __mptcp_tcp_fallback()
return the first subflow sock and cope correctly even with
incoming connections.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This cleans the code a bit and makes the behavior more consistent.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This cleanup the code a bit and avoid corrupted states
on weird syscall sequence (accept(), connect()).
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Keep using MPTCP sockets and a use "dummy mapping" in case of fallback
to regular TCP. When fallback is triggered, skip addition of the MPTCP
option on send.
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/11
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/22
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace the radix tree with a hash table allocated
at boot time. The radix tree has some shortcoming:
a single lock is contented by all the mptcp operation,
the lookup currently use such lock, and traversing
all the items would require a lock, too.
With hash table instead we trade a little memory to
address all the above - a per bucket lock is used.
To hash the MPTCP sockets, we re-use the msk' sk_node
entry: the MPTCP sockets are never hashed by the stack.
Replace the existing hash proto callbacks with a dummy
implementation, annotating the above constraint.
Additionally refactor the token creation to code to:
- limit the number of consecutive attempts to a fixed
maximum. Hitting a hash bucket with a long chain is
considered a failed attempt
- accept() no longer can fail to token management.
- if token creation fails at connect() time, we do
fallback to TCP (before the connection was closed)
v1 -> v2:
- fix "no newline at end of file" - Jakub
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the missing annotation in some setup-only
functions.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The msk sk_shutdown flag is set by a workqueue, possibly
introducing some delay in user-space notification. If the last
subflow carries some data with the fin packet, the user space
can wake-up before RCV_SHUTDOWN is set. If it executes unblocking
recvmsg(), it may return with an error instead of eof.
Address the issue explicitly checking for eof in recvmsg(), when
no data is found.
Fixes: 59832e2465 ("mptcp: subflow: check parent mptcp socket on subflow state change")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
xdp_umem.c had overlapping changes between the 64-bit math fix
for the calculation of npgs and the removal of the zerocopy
memory type which got rid of the chunk_size_nohdr member.
The mlx5 Kconfig conflict is a case where we just take the
net-next copy of the Kconfig entry dependency as it takes on
the ESWITCH dependency by one level of indirection which is
what the 'net' conflicting change is trying to ensure.
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently we remote the msk from the token container only
via mptcp_close(). The MPTCP master socket can be destroyed
also via other paths (e.g. if not yet accepted, when shutting
down the listener socket). When we hit the latter scenario,
dangling msk references are left into the token container,
leading to memory corruption and/or UaF.
This change addresses the issue by moving the token removal
into the msk destructor.
Fixes: 79c0949e9a ("mptcp: Add key generation and token tree")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If a MP_JOIN subflow completes the 3whs while another
CPU is closing the master msk, we can hit the
following race:
CPU1 CPU2
close()
mptcp_close
subflow_syn_recv_sock
mptcp_token_get_sock
mptcp_finish_join
inet_sk_state_load
mptcp_token_destroy
inet_sk_state_store(TCP_CLOSE)
__mptcp_flush_join_list()
mptcp_sock_graft
list_add_tail
sk_common_release
sock_orphan()
<socket free>
The MP_JOIN socket will be leaked. Additionally we can hit
UaF for the msk 'struct socket' referenced via the 'conn'
field.
This change try to address the issue introducing some
synchronization between the MP_JOIN 3whs and mptcp_close
via the join_list spinlock. If we detect the msk is closing
the MP_JOIN socket is closed, too.
Fixes: f296234c98 ("mptcp: Add handling of incoming MP_JOIN requests")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently unblocking connect() on MPTCP sockets fails frequently.
If mptcp_stream_connect() is invoked to complete a previously
attempted unblocking connection, it will still try to create
the first subflow via __mptcp_socket_create(). If the 3whs is
completed and the 'can_ack' flag is already set, the latter
will fail with -EINVAL.
This change addresses the issue checking for pending connect and
delegating the completion to the first subflow. Additionally
do msk addresses and sk_state changes only when needed.
Fixes: 2303f994b3 ("mptcp: Associate MPTCP context with TCP socket")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can try to coalesce skbs we take from the subflows rx queue with the
tail of the mptcp rx queue.
If successful, the skb head can be discarded early.
We can also free the skb extensions, we do not access them after this.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
To prepare removing the global routing_ioctl hack start lifting the code
into a newly added ipv6 ->compat_ioctl handler.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
mptcp calls this from the transmit side, from process context.
Allow a sleeping allocation instead of unconditional GFP_ATOMIC.
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
previous patches made sure we only call into this function
when these prerequisites are met, so no need to wait on the
subflow socket anymore.
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/7
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The mptcp_sendmsg_frag helper contains a loop that will wait on the
subflow sk.
It seems preferrable to only wait in mptcp_sendmsg() when blocking io is
requested. mptcp_sendmsg already has such a wait loop that is used when
no subflow socket is available for transmission.
This is another preparation patch that makes sure we call
mptcp_sendmsg_frag only if the page frag cache has been refilled.
Followup patch will remove the wait loop from mptcp_sendmsg_frag().
The retransmit worker doesn't need to do this refill as it won't
transmit new mptcp-level data.
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The mptcp_sendmsg_frag helper contains a loop that will wait on the
subflow sk.
It seems preferrable to only wait in mptcp_sendmsg() when blocking io is
requested. mptcp_sendmsg already has such a wait loop that is used when
no subflow socket is available for transmission.
This is a preparation patch that makes sure we call
mptcp_sendmsg_frag only if a skb extension has been allocated.
Moreover, such allocation currently uses GFP_ATOMIC while it
could use sleeping allocation instead.
Followup patches will remove the wait loop from mptcp_sendmsg_frag()
and will allow to do a sleeping allocation for the extension.
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The transmit loop continues to xmit new data until an error is returned
or all data was transmitted.
For the blocking i/o case, this means that tcp_sendpages() may block on
the subflow until more space becomes available, i.e. we end up sleeping
with the mptcp socket lock held.
Instead we should check if a different subflow is ready to be used.
This restarts the subflow sk lookup when the tx operation succeeded
and the tcp subflow can't accept more data or if tcp_sendpages
indicates -EAGAIN on a blocking mptcp socket.
In that case we also need to set the NOSPACE bit to make sure we get
notified once memory becomes available.
In case all subflows are busy, the existing logic will wait until a
subflow is ready, releasing the mptcp socket lock while doing so.
The mptcp worker already sets DONTWAIT, so no need to make changes there.
v2:
* set NOSPACE bit
* add a comment to clarify that mptcp-sk sndbuf limits need to
be checked as well.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Its not enough to check for available tcp send space.
We also hold on to transmitted data for mptcp-level retransmits.
Right now we will send more and more data if the peer can ack data
at the tcp level fast enough, since that frees up tcp send buffer space.
But we also need to check that data was acked and reclaimed at the mptcp
level.
Therefore add needed check in mptcp_sendmsg, flush tcp data and
wait until more mptcp snd space becomes available if we are over the
limit. Before we wait for more data, also make sure we start the
retransmit timer if we ran out of sndbuf space.
Otherwise there is a very small chance that we wait forever:
* receiver is waiting for data
* sender is blocked because mptcp socket buffer is full
* at tcp level, all data was acked
* mptcp-level snd_una was not updated, because last ack
that acknowledged the last data packet carried an older
MPTCP-ack.
Restarting the retransmit timer avoids this problem: if TCP
subflow is idle, data is retransmitted from the RTX queue.
New data will make the peer send a new, updated MPTCP-Ack.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Paolo noticed that ssk_check_wmem() has same pattern, so add/use
common helper for both places.
Suggested-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the bpf verifier trace check into the new switch statement in
HEAD.
Resolve the overlapping changes in hinic, where bug fixes overlap
the addition of VF support.
Signed-off-by: David S. Miller <davem@davemloft.net>
MP_JOIN subflows must not land into the accept queue.
Currently tcp_check_req() calls an mptcp specific helper
to detect such scenario.
Such helper leverages the subflow context to check for
MP_JOIN subflows. We need to deal also with MP JOIN
failures, even when the subflow context is not available
due allocation failure.
A possible solution would be changing the syn_recv_sock()
signature to allow returning a more descriptive action/
error code and deal with that in tcp_check_req().
Since the above need is MPTCP specific, this patch instead
uses a TCP request socket hole to add a MPTCP specific flag.
Such flag is used by the MPTCP syn_recv_sock() to tell
tcp_check_req() how to deal with the request socket.
This change is a no-op for !MPTCP build, and makes the
MPTCP code simpler. It allows also the next patch to deal
correctly with MP JOIN failure.
v1 -> v2:
- be more conservative on drop_req initialization (Mat)
RFC -> v1:
- move the drop_req bit inside tcp_request_sock (Eric)
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Reviewed-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the other MPTCP-peer uses 32-bit data-sequence numbers, we rely on
map_seq to indicate how to expand to a 64-bit data-sequence number in
expand_seq() when receiving data.
For new subflows, this field is not initialized, thus results in an
"invalid" mapping being discarded.
Fix this by initializing map_seq upon subflow establishment time.
Fixes: f296234c98 ("mptcp: Add handling of incoming MP_JOIN requests")
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Reviewed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Paolo points out that mptcp_disconnect is bogus:
"lock_sock(sk);
looks suspicious (lock should be already held by the caller)
And call to: tcp_disconnect(sk, flags); too, sk is not a tcp
socket".
->disconnect() gets called from e.g. inet_stream_connect when
one tries to disassociate a connected socket again (to re-connect
without closing the socket first).
MPTCP however uses mptcp_stream_connect, not inet_stream_connect,
for the mptcp-socket connect call.
inet_stream_connect only gets called indirectly, for the tcp socket,
so any ->disconnect() calls end up calling tcp_disconnect for that
tcp subflow sk.
This also explains why syzkaller has not yet reported a problem
here. So for now replace this with a stub that doesn't do anything.
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/14
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
We don't need them, as we can use the current ingress opt
data instead. Setting them in syn_recv_sock() may causes
inconsistent mptcp socket status, as per previous commit.
Fixes: cc7972ea19 ("mptcp: parse and emit MP_CAPABLE option according to v1 spec")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Following splat can occur during self test:
BUG: KASAN: use-after-free in subflow_data_ready+0x156/0x160
Read of size 8 at addr ffff888100c35c28 by task mptcp_connect/4808
subflow_data_ready+0x156/0x160
tcp_child_process+0x6a3/0xb30
tcp_v4_rcv+0x2231/0x3730
ip_protocol_deliver_rcu+0x5c/0x860
ip_local_deliver_finish+0x220/0x360
ip_local_deliver+0x1c8/0x4e0
ip_rcv_finish+0x1da/0x2f0
ip_rcv+0xd0/0x3c0
__netif_receive_skb_one_core+0xf5/0x160
__netif_receive_skb+0x27/0x1c0
process_backlog+0x21e/0x780
net_rx_action+0x35f/0xe90
do_softirq+0x4c/0x50
[..]
This occurs when accessing subflow_ctx->conn.
Problem is that tcp_child_process() calls listen sockets'
sk_data_ready() notification, but it doesn't hold the listener
lock. Another cpu calling close() on the listener will then cause
transition of refcount to 0.
Fixes: 58b0991962 ("mptcp: create msk early")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We need to set sk_state to CLOSED, else we will get following:
IPv4: Attempt to release TCP socket in state 3 00000000b95f109e
IPv4: Attempt to release TCP socket in state 10 00000000b95f109e
First one is from inet_sock_destruct(), second one from
mptcp_sk_clone failure handling. Setting sk_state to CLOSED isn't
enough, we also need to orphan sk so it has DEAD flag set.
Otherwise, a very similar warning is printed from inet_sock_destruct().
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Following snippet (replicated from syzkaller reproducer) generates
warning: "IPv4: Attempt to release TCP socket in state 1".
int main(void) {
struct sockaddr_in sin1 = { .sin_family = 2, .sin_port = 0x4e20,
.sin_addr.s_addr = 0x010000e0, };
struct sockaddr_in sin2 = { .sin_family = 2,
.sin_addr.s_addr = 0x0100007f, };
struct sockaddr_in sin3 = { .sin_family = 2, .sin_port = 0x4e20,
.sin_addr.s_addr = 0x0100007f, };
int r0 = socket(0x2, 0x1, 0x106);
int r1 = socket(0x2, 0x1, 0x106);
bind(r1, (void *)&sin1, sizeof(sin1));
connect(r1, (void *)&sin2, sizeof(sin2));
listen(r1, 3);
return connect(r0, (void *)&sin3, 0x4d);
}
Reason is that the newly generated mptcp socket is closed via the ulp
release of the tcp listener socket when its accept backlog gets purged.
To fix this, delay setting the ESTABLISHED state until after userspace
calls accept and via mptcp specific destructor.
Fixes: 58b0991962 ("mptcp: create msk early")
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/9
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
mptcp_connect/28740 is trying to release lock (sk_lock-AF_INET) at:
[<ffffffff82c15869>] mptcp_poll+0xb9/0x550
but there are no more locks to release!
Call Trace:
lock_release+0x50f/0x750
release_sock+0x171/0x1b0
mptcp_poll+0xb9/0x550
sock_poll+0x157/0x470
? get_net_ns+0xb0/0xb0
do_sys_poll+0x63c/0xdd0
Problem is that __mptcp_tcp_fallback() releases the mptcp socket lock,
but after recent change it doesn't do this in all of its return paths.
To fix this, remove the unlock from __mptcp_tcp_fallback() and
always do the unlock in the caller.
Also add a small comment as to why we have this
__mptcp_needs_tcp_fallback().
Fixes: 0b4f33def7 ("mptcp: fix tcp fallback crash")
Reported-by: syzbot+e56606435b7bfeea8cf5@syzkaller.appspotmail.com
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
mptcp_subflow_data_available() is commonly called via
ssk->sk_data_ready(), in this case the mptcp socket lock
cannot be acquired.
Therefore, while we can safely discard subflow data that
was already received up to msk->ack_seq, we cannot be sure
that 'subflow->data_avail' will still be valid at the time
userspace wants to read the data -- a previous read on a
different subflow might have carried this data already.
In that (unlikely) event, msk->ack_seq will have been updated
and will be ahead of the subflow dsn.
We can check for this condition and skip/resync to the expected
sequence number.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is needed at least until proper MPTCP-Level fin/reset
signalling gets added:
We wake parent when a subflow changes, but we should do this only
when all subflows have closed, not just one.
Schedule the mptcp worker and tell it to check eof state on all
subflows.
Only flag mptcp socket as closed and wake userspace processes blocking
in poll if all subflows have closed.
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Christoph Paasch reports following crash:
general protection fault [..]
CPU: 0 PID: 2874 Comm: syz-executor072 Not tainted 5.6.0-rc5 #62
RIP: 0010:__pv_queued_spin_lock_slowpath kernel/locking/qspinlock.c:471
[..]
queued_spin_lock_slowpath arch/x86/include/asm/qspinlock.h:50 [inline]
do_raw_spin_lock include/linux/spinlock.h:181 [inline]
spin_lock_bh include/linux/spinlock.h:343 [inline]
__mptcp_flush_join_list+0x44/0xb0 net/mptcp/protocol.c:278
mptcp_shutdown+0xb3/0x230 net/mptcp/protocol.c:1882
[..]
Problem is that mptcp_shutdown() socket isn't an mptcp socket,
its a plain tcp_sk. Thus, trying to access mptcp_sk specific
members accesses garbage.
Root cause is that accept() returns a fallback (tcp) socket, not an mptcp
one. There is code in getpeername to detect this and override the sockets
stream_ops. But this will only run when accept() caller provided a
sockaddr struct. "accept(fd, NULL, 0)" will therefore result in
mptcp stream ops, but with sock->sk pointing at a tcp_sk.
Update the existing fallback handling to detect this as well.
Moreover, mptcp_shutdown did not have fallback handling, and
mptcp_poll did it too late so add that there as well.
Reported-by: Christoph Paasch <cpaasch@apple.com>
Tested-by: Christoph Paasch <cpaasch@apple.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Exported via same /proc file as the Linux TCP MIB counters, so "netstat -s"
or "nstat" will show them automatically.
The MPTCP MIB counters are allocated in a distinct pcpu area in order to
avoid bloating/wasting TCP pcpu memory.
Counters are allocated once the first MPTCP socket is created in a
network namespace and free'd on exit.
If no sockets have been allocated, all-zero mptcp counters are shown.
The MIB counter list is taken from the multipath-tcp.org kernel, but
only a few counters have been picked up so far. The counter list can
be increased at any time later on.
v2 -> v3:
- remove 'inline' in foo.c files (David S. Miller)
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On timeout event, schedule a work queue to do the retransmission.
Retransmission code closely resembles the sendmsg() implementation and
re-uses mptcp_sendmsg_frag, providing a dummy msghdr - for flags'
sake - and peeking the relevant dfrag from the rtx head.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This will simplify mptcp-level retransmission implementation
in the next patch. If dfrag is provided by the caller, skip
kernel space memory allocation and use data and metadata
provided by the dfrag itself.
Because a peer could ack data at TCP level but refrain from
sending mptcp-level ACKs, we could grow the mptcp socket
backlog indefinitely.
We should thus block mptcp_sendmsg until the peer has acked some of the
sent data.
In order to be able to do so, increment the mptcp socket wmem_queued
counter on memory allocation and decrement it when releasing the memory
on mptcp-level ack reception.
Because TCP performns sndbuf auto-tuning up to tcp_wmem_max[2], make
this the mptcp sk_sndbuf limit.
In the future we could add experiment with autotuning as TCP does in
tcp_sndbuf_expand().
v2 -> v3:
- remove 'inline' in foo.c files (David S. Miller)
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After adding wmem accounting for the mptcp socket we could get
into a situation where the mptcp socket can't transmit more data,
and mptcp_clean_una doesn't reduce wmem even if snd_una has advanced
because it currently will only remove entire dfrags.
Allow advancing the dfrag head sequence and reduce wmem,
even though this isn't correct (as we can't release the page).
Because we will soon block on mptcp sk in case wmem is too large,
call sk_stream_write_space() in case we reduced the backlog so
userspace task blocked in sendmsg or poll will be woken up.
This isn't an issue if the send buffer is large, but it is when
SO_SNDBUF is used to reduce it to a lower value.
Note we can still get a deadlock for low SO_SNDBUF values in
case both sides of the connection write to the socket: both could
be blocked due to wmem being too small -- and current mptcp stack
will only increment mptcp ack_seq on recv.
This doesn't happen with the selftest as it uses poll() and
will always call recv if there is data to read.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Charge the data on the rtx queue to the master MPTCP socket, too.
Such memory in uncharged when the data is acked/dequeued.
Also account mptcp sockets inuse via a protocol specific pcpu
counter.
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The timer will be used to schedule retransmission. It's
frequency is based on the current subflow RTO estimation and
is reset on every una_seq update
The timer is clearer for good by __mptcp_clear_xmit()
Also clean MPTCP rtx queue before each transmission.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Keep the send page fragment on an MPTCP level retransmission queue.
The queue entries are allocated inside the page frag allocator,
acquiring an additional reference to the page for each list entry.
Also switch to a custom page frag refill function, to ensure that
the current page fragment can always host an MPTCP rtx queue entry.
The MPTCP rtx queue is flushed at disconnect() and close() time
Note that now we need to call __mptcp_init_sock() regardless of mptcp
enable status, as the destructor will try to walk the rtx_queue.
v2 -> v3:
- remove 'inline' in foo.c files (David S. Miller)
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
So that we keep per unacked sequence number consistent; since
we update per msk data, use an atomic64 cmpxchg() to protect
against concurrent updates from multiple subflows.
Initialize the snd_una at connect()/accept() time.
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fill in more path manager functionality by adding a worker function and
modifying the related stub functions to schedule the worker.
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Subflow creation may be initiated by the path manager when
the primary connection is fully established and a remote
address has been received via ADD_ADDR.
Create an in-kernel sock and use kernel_connect() to
initiate connection.
Passive sockets can't acquire the mptcp socket lock at
subflow creation time, so an additional list protected by
a new spinlock is used to track the MPJ subflows.
Such list is spliced into conn_list tail every time the msk
socket lock is acquired, so that it will not interfere
with data flow on the original connection.
Data flow and connection failover not addressed by this commit.
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Process the MP_JOIN option in a SYN packet with the same flow
as MP_CAPABLE but when the third ACK is received add the
subflow to the MPTCP socket subflow list instead of adding it to
the TCP socket accept queue.
The subflow is added at the end of the subflow list so it will not
interfere with the existing subflows operation and no data is
expected to be transmitted on it.
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add enough of a path manager interface to allow sending of ADD_ADDR
when an incoming MPTCP connection is created. Capable of sending only
a single IPv4 ADD_ADDR option. The 'pm_data' element of the connection
sock will need to be expanded to handle multiple interfaces and IPv6.
Partial processing of the incoming ADD_ADDR is included so the path
manager notification of that event happens at the proper time, which
involves validating the incoming address information.
This is a skeleton interface definition for events generated by
MPTCP.
Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Co-developed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
it's still possible for packetdrill to hang in mptcp_sendmsg(), when the
MPTCP socket falls back to regular TCP (e.g. after receiving unsupported
flags/version during the three-way handshake). Adjust MPTCP socket state
earlier, to ensure correct functionality of mptcp_sendmsg() even in case
of TCP fallback.
Fixes: 767d3ded5f ("net: mptcp: don't hang before sending 'MP capable with data'")
Fixes: 1954b86016 ("mptcp: Check connection state before attempting send")
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
After commit 58b0991962 ("mptcp: create msk early"), the
msk socket is already available at subflow_syn_recv_sock()
time. Let's move there the state update, to mirror more
closely the first subflow state.
The above will also help multiple subflow supports.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This change moves the mptcp socket allocation from mptcp_accept() to
subflow_syn_recv_sock(), so that subflow->conn is now always set
for the non fallback scenario.
It allows cleaning up a bit mptcp_accept() reducing the additional
locking and will allow fourther cleanup in the next patch.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
The mptcp rcvbuf size is adjusted according to the subflow rcvbuf size.
This should not be done if userspace did set a fixed value.
Fixes: 600911ff5f ("mptcp: add rmem queue accounting")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of reading the MPTCP-level sequence number when sending DATA_FIN,
store the data in the subflow so it can be safely accessed when the
subflow TCP headers are written to the packet without the MPTCP-level
lock held. This also allows the MPTCP-level socket to close individual
subflows without closing the MPTCP connection.
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
MPTCP should wait for an active connection or skip sending depending on
the connection state, as TCP does. This happens before the possible
passthrough to a regular TCP sendmsg because the subflow's socket type
(MPTCP or TCP fallback) is not known until the connection is
complete. This is also relevent at disconnect time, where data should
not be sent in certain MPTCP-level connection states.
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The mptcp conflict was overlapping additions.
The SMC conflict was an additional and removal happening at the same
time.
Signed-off-by: David S. Miller <davem@davemloft.net>
Don't schedule the work queue right away, instead defer this
to the lock release callback.
This has the advantage that it will give recv path a chance to
complete -- this might have moved all pending packets from the
subflow to the mptcp receive queue, which allows to avoid the
schedule_work().
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can't lock_sock() the mptcp socket from the subflow data_ready callback,
it would result in ABBA deadlock with the subflow socket lock.
We can however grab the spinlock: if that succeeds and the mptcp socket
is not owned at the moment, we can process the new skbs right away
without deferring this to the work queue.
This avoids the schedule_work and hence the small delay until the
work item is processed.
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Only used to discard stale data from the subflow, so move
it where needed.
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If userspace never drains the receive buffers we must stop draining
the subflow socket(s) at some point.
This adds the needed rmem accouting for this.
If the threshold is reached, we stop draining the subflows.
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If userspace is not reading data, all the mptcp-level acks contain the
ack_seq from the last time userspace read data rather than the most
recent in-sequence value.
This causes pointless retransmissions for data that is already queued.
The reason for this is that all the mptcp protocol level processing
happens at mptcp_recv time.
This adds work queue to move skbs from the subflow sockets receive
queue on the mptcp socket receive queue (which was not used so far).
This allows us to announce the correct mptcp ack sequence in a timely
fashion, even when the application does not call recv() on the mptcp socket
for some time.
We still wake userspace tasks waiting for POLLIN immediately:
If the mptcp level receive queue is empty (because the work queue is
still pending) it can be filled from in-sequence subflow sockets at
recv time without a need to wait for the worker.
The skb_orphan when moving skbs from subflow to mptcp level is needed,
because the destructor (sock_rfree) relies on skb->sk (ssk!) lock
being taken.
A followup patch will add needed rmem accouting for the moved skbs.
Other problem: In case application behaves as expected, and calls
recv() as soon as mptcp socket becomes readable, the work queue will
only waste cpu cycles. This will also be addressed in followup patches.
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Will be extended with functionality in followup patches.
Initial user is moving skbs from subflows receive queue to
the mptcp-level receive queue.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
allows us to schedule the work queue to drain the ssk receive queue in
a followup patch.
This is needed to avoid sending all-to-pessimistic mptcp-level
acknowledgements. At this time, the ack_seq is what was last read by
userspace instead of the highest in-sequence number queued for reading.
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Userspace should not be able to directly manipulate subflow socket
options before a connection is established since it is not yet known if
it will be an MPTCP subflow or a TCP fallback subflow. TCP fallback
subflows can be more directly controlled by userspace because they are
regular TCP connections, while MPTCP subflow sockets need to be
configured for the specific needs of MPTCP. Use the same logic as
sendmsg/recvmsg to ensure that socket option calls are only passed
through to known TCP fallback subflows.
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix the following sparse warning:
net/mptcp/protocol.c:646:13: warning: symbol 'mptcp_sk_clone_lock' was not declared. Should it be static?
Fixes: b0519de8b3 ("mptcp: fix use-after-free for ipv6")
Signed-off-by: Chen Wandun <chenwandun@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Turns out that when we accept a new subflow, the newly created
inet_sk(tcp_sk)->pinet6 points at the ipv6_pinfo structure of the
listener socket.
This wasn't caught by the selftest because it closes the accepted fd
before the listening one.
adding a close(listenfd) after accept returns is enough:
BUG: KASAN: use-after-free in inet6_getname+0x6ba/0x790
Read of size 1 at addr ffff88810e310866 by task mptcp_connect/2518
Call Trace:
inet6_getname+0x6ba/0x790
__sys_getpeername+0x10b/0x250
__x64_sys_getpeername+0x6f/0xb0
also alter test program to exercise this.
Reported-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
When an mptcp socket connects to a tcp peer or when a middlebox interferes
with tcp options, mptcp needs to fall back to plain tcp.
Problem is that mptcp is trying to be too clever in this case:
It attempts to close the mptcp meta sk and transparently replace it with
the (only) subflow tcp sk.
Unfortunately, this is racy -- the socket is already exposed to userspace.
Any parallel calls to send/recv/setsockopt etc. can cause use-after-free:
BUG: KASAN: use-after-free in atomic_try_cmpxchg include/asm-generic/atomic-instrumented.h:693 [inline]
CPU: 1 PID: 2083 Comm: syz-executor.1 Not tainted 5.5.0 #2
atomic_try_cmpxchg include/asm-generic/atomic-instrumented.h:693 [inline]
queued_spin_lock include/asm-generic/qspinlock.h:78 [inline]
do_raw_spin_lock include/linux/spinlock.h:181 [inline]
__raw_spin_lock_bh include/linux/spinlock_api_smp.h:136 [inline]
_raw_spin_lock_bh+0x71/0xd0 kernel/locking/spinlock.c:175
spin_lock_bh include/linux/spinlock.h:343 [inline]
__lock_sock+0x105/0x190 net/core/sock.c:2414
lock_sock_nested+0x10f/0x140 net/core/sock.c:2938
lock_sock include/net/sock.h:1516 [inline]
mptcp_setsockopt+0x2f/0x1f0 net/mptcp/protocol.c:800
__sys_setsockopt+0x152/0x240 net/socket.c:2130
__do_sys_setsockopt net/socket.c:2146 [inline]
__se_sys_setsockopt net/socket.c:2143 [inline]
__x64_sys_setsockopt+0xba/0x150 net/socket.c:2143
do_syscall_64+0xb7/0x3d0 arch/x86/entry/common.c:294
entry_SYSCALL_64_after_hwframe+0x44/0xa9
While the use-after-free can be resolved, there is another problem:
sock->ops and sock->sk assignments are not atomic, i.e. we may get calls
into mptcp functions with sock->sk already pointing at the subflow socket,
or calls into tcp functions with a mptcp meta sk.
Remove the fallback code and call the relevant functions for the (only)
subflow in case the mptcp socket is connected to tcp peer.
Reported-by: Christoph Paasch <cpaasch@apple.com>
Diagnosed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Tested-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If CONFIG_MPTCP=y, CONFIG_MPTCP_IPV6=n, and CONFIG_IPV6=m:
net/mptcp/protocol.o: In function `__mptcp_tcp_fallback':
protocol.c:(.text+0x786): undefined reference to `inet6_stream_ops'
Fix this by checking for CONFIG_MPTCP_IPV6 instead of CONFIG_IPV6, like
is done in all other places in the mptcp code.
Fixes: 8ab183deb2 ("mptcp: cope with later TCP fallback")
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
syzbot triggered following lockdep splat:
ffffffff82d2cd40 (rtnl_mutex){+.+.}, at: ip_mc_drop_socket+0x52/0x180
but task is already holding lock:
ffff8881187a2310 (sk_lock-AF_INET){+.+.}, at: mptcp_close+0x18/0x30
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #1 (sk_lock-AF_INET){+.+.}:
lock_acquire+0xee/0x230
lock_sock_nested+0x89/0xc0
do_ip_setsockopt.isra.0+0x335/0x22f0
ip_setsockopt+0x35/0x60
tcp_setsockopt+0x5d/0x90
__sys_setsockopt+0xf3/0x190
__x64_sys_setsockopt+0x61/0x70
do_syscall_64+0x72/0x300
entry_SYSCALL_64_after_hwframe+0x49/0xbe
-> #0 (rtnl_mutex){+.+.}:
check_prevs_add+0x2b7/0x1210
__lock_acquire+0x10b6/0x1400
lock_acquire+0xee/0x230
__mutex_lock+0x120/0xc70
ip_mc_drop_socket+0x52/0x180
inet_release+0x36/0xe0
__sock_release+0xfd/0x130
__mptcp_close+0xa8/0x1f0
inet_release+0x7f/0xe0
__sock_release+0x69/0x130
sock_close+0x18/0x20
__fput+0x179/0x400
task_work_run+0xd5/0x110
do_exit+0x685/0x1510
do_group_exit+0x7e/0x170
__x64_sys_exit_group+0x28/0x30
do_syscall_64+0x72/0x300
entry_SYSCALL_64_after_hwframe+0x49/0xbe
The trigger is:
socket(AF_INET, SOCK_STREAM, 0x106 /* IPPROTO_MPTCP */) = 4
setsockopt(4, SOL_IP, MCAST_JOIN_GROUP, {gr_interface=7, gr_group={sa_family=AF_INET, sin_port=htons(20003), sin_addr=inet_addr("224.0.0.2")}}, 136) = 0
exit(0)
Which results in a call to rtnl_lock while we are holding
the parent mptcp socket lock via
mptcp_close -> lock_sock(msk) -> inet_release -> ip_mc_drop_socket -> rtnl_lock().
>From lockdep point of view we thus have both
'rtnl_lock; lock_sock' and 'lock_sock; rtnl_lock'.
Fix this by stealing the msk conn_list and doing the subflow close
without holding the msk lock.
Fixes: cec37a6e41 ("mptcp: Handle MP_CAPABLE options for outgoing connections")
Reported-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Its not possible to call the kernel_(s|g)etsockopt functions here,
the address points to user memory:
General protection fault in user access. Non-canonical address?
WARNING: CPU: 1 PID: 5352 at arch/x86/mm/extable.c:77 ex_handler_uaccess+0xba/0xe0 arch/x86/mm/extable.c:77
Kernel panic - not syncing: panic_on_warn set ...
[..]
Call Trace:
fixup_exception+0x9d/0xcd arch/x86/mm/extable.c:178
general_protection+0x2d/0x40 arch/x86/entry/entry_64.S:1202
do_ip_getsockopt+0x1f6/0x1860 net/ipv4/ip_sockglue.c:1323
ip_getsockopt+0x87/0x1c0 net/ipv4/ip_sockglue.c:1561
tcp_getsockopt net/ipv4/tcp.c:3691 [inline]
tcp_getsockopt+0x8c/0xd0 net/ipv4/tcp.c:3685
kernel_getsockopt+0x121/0x1f0 net/socket.c:3736
mptcp_getsockopt+0x69/0x90 net/mptcp/protocol.c:830
__sys_getsockopt+0x13a/0x220 net/socket.c:2175
We can call tcp_get/setsockopt functions instead. Doing so fixes
crashing, but still leaves rtnl related lockdep splat:
WARNING: possible circular locking dependency detected
5.5.0-rc6 #2 Not tainted
------------------------------------------------------
syz-executor.0/16334 is trying to acquire lock:
ffffffff84f7a080 (rtnl_mutex){+.+.}, at: do_ip_setsockopt.isra.0+0x277/0x3820 net/ipv4/ip_sockglue.c:644
but task is already holding lock:
ffff888116503b90 (sk_lock-AF_INET){+.+.}, at: lock_sock include/net/sock.h:1516 [inline]
ffff888116503b90 (sk_lock-AF_INET){+.+.}, at: mptcp_setsockopt+0x28/0x90 net/mptcp/protocol.c:1284
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #1 (sk_lock-AF_INET){+.+.}:
lock_sock_nested+0xca/0x120 net/core/sock.c:2944
lock_sock include/net/sock.h:1516 [inline]
do_ip_setsockopt.isra.0+0x281/0x3820 net/ipv4/ip_sockglue.c:645
ip_setsockopt+0x44/0xf0 net/ipv4/ip_sockglue.c:1248
udp_setsockopt+0x5d/0xa0 net/ipv4/udp.c:2639
__sys_setsockopt+0x152/0x240 net/socket.c:2130
__do_sys_setsockopt net/socket.c:2146 [inline]
__se_sys_setsockopt net/socket.c:2143 [inline]
__x64_sys_setsockopt+0xba/0x150 net/socket.c:2143
do_syscall_64+0xbd/0x5b0 arch/x86/entry/common.c:294
entry_SYSCALL_64_after_hwframe+0x49/0xbe
-> #0 (rtnl_mutex){+.+.}:
check_prev_add kernel/locking/lockdep.c:2475 [inline]
check_prevs_add kernel/locking/lockdep.c:2580 [inline]
validate_chain kernel/locking/lockdep.c:2970 [inline]
__lock_acquire+0x1fb2/0x4680 kernel/locking/lockdep.c:3954
lock_acquire+0x127/0x330 kernel/locking/lockdep.c:4484
__mutex_lock_common kernel/locking/mutex.c:956 [inline]
__mutex_lock+0x158/0x1340 kernel/locking/mutex.c:1103
do_ip_setsockopt.isra.0+0x277/0x3820 net/ipv4/ip_sockglue.c:644
ip_setsockopt+0x44/0xf0 net/ipv4/ip_sockglue.c:1248
tcp_setsockopt net/ipv4/tcp.c:3159 [inline]
tcp_setsockopt+0x8c/0xd0 net/ipv4/tcp.c:3153
kernel_setsockopt+0x121/0x1f0 net/socket.c:3767
mptcp_setsockopt+0x69/0x90 net/mptcp/protocol.c:1288
__sys_setsockopt+0x152/0x240 net/socket.c:2130
__do_sys_setsockopt net/socket.c:2146 [inline]
__se_sys_setsockopt net/socket.c:2143 [inline]
__x64_sys_setsockopt+0xba/0x150 net/socket.c:2143
do_syscall_64+0xbd/0x5b0 arch/x86/entry/common.c:294
entry_SYSCALL_64_after_hwframe+0x49/0xbe
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0 CPU1
---- ----
lock(sk_lock-AF_INET);
lock(rtnl_mutex);
lock(sk_lock-AF_INET);
lock(rtnl_mutex);
The lockdep complaint is because we hold mptcp socket lock when calling
the sk_prot get/setsockopt handler, and those might need to acquire the
rtnl mutex. Normally, order is:
rtnl_lock(sk) -> lock_sock
Whereas for mptcp the order is
lock_sock(mptcp_sk) rtnl_lock -> lock_sock(subflow_sk)
We can avoid this by releasing the mptcp socket lock early, but, as Paolo
points out, we need to get/put the subflow socket refcount before doing so
to avoid race with concurrent close().
Fixes: 717e79c867 ("mptcp: Add setsockopt()/getsockopt() socket operations")
Reported-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
access to msk->cached_ext is only legal if the msk is locked or all
concurrent accesses are impossible.
Furthermore, once we start to tear down, we must make sure nothing else
can step in and allocate a new cached ext.
So place this code in the destroy callback where it belongs.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
checkpatch.pl had a few complaints in the last set of MPTCP patches:
ERROR: code indent should use tabs where possible
+^I subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);$
CHECK: Comparison to NULL could be written "!new_ctx"
+ if (new_ctx == NULL) {
ERROR: "foo * bar" should be "foo *bar"
+static const struct proto_ops * tcp_proto_ops(struct sock *sk)
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We need to initialise the struct ourselves, else we expose tcp-specific
callbacks such as tcp_splice_read which will then trigger splat because
the socket is an mptcp one:
BUG: KASAN: slab-out-of-bounds in tcp_mstamp_refresh+0x80/0xa0 net/ipv4/tcp_output.c:57
Write of size 8 at addr ffff888116aa21d0 by task syz-executor.0/5478
CPU: 1 PID: 5478 Comm: syz-executor.0 Not tainted 5.5.0-rc6 #3
Call Trace:
tcp_mstamp_refresh+0x80/0xa0 net/ipv4/tcp_output.c:57
tcp_rcv_space_adjust+0x72/0x7f0 net/ipv4/tcp_input.c:612
tcp_read_sock+0x622/0x990 net/ipv4/tcp.c:1674
tcp_splice_read+0x20b/0xb40 net/ipv4/tcp.c:791
do_splice+0x1259/0x1560 fs/splice.c:1205
To prevent build error with ipv6, add the recv/sendmsg function
declaration to ipv6.h. The functions are already accessible "thanks"
to retpoline related work, but they are currently only made visible
by socket.c specific INDIRECT_CALLABLE macros.
Reported-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With MPTCP v1, passive connections can fallback to TCP after the
subflow becomes established:
syn + MP_CAPABLE ->
<- syn, ack + MP_CAPABLE
ack, seq = 3 ->
// OoO packet is accepted because in-sequence
// passive socket is created, is in ESTABLISHED
// status and tentatively as MP_CAPABLE
ack, seq = 2 ->
// no MP_CAPABLE opt, subflow should fallback to TCP
We can't use the 'subflow' socket fallback, as we don't have
it available for passive connection.
Instead, when the fallback is detected, replace the mptcp
socket with the underlying TCP subflow. Beyond covering
the above scenario, it makes a TCP fallback socket as efficient
as plain TCP ones.
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch implements the handling of MP_CAPABLE + data option, as per
RFC 6824 bis / RFC 8684: MPTCP v1.
On the server side we can receive the remote key after that the connection
is established. We need to explicitly track the 'missing remote key'
status and avoid emitting a mptcp ack until we get such info.
When a late/retransmitted/OoO pkt carrying MP_CAPABLE[+data] option
is received, we have to propagate the mptcp seq number info to
the msk socket. To avoid ABBA locking issue, explicitly check for
that in recvmsg(), where we own msk and subflow sock locks.
The above also means that an established mp_capable subflow - still
waiting for the remote key - can be 'downgraded' to plain TCP.
Such change could potentially block a reader waiting for new data
forever - as they hook to msk, while later wake-up after the downgrade
will be on subflow only.
The above issue is not handled here, we likely have to get rid of
msk->fallback to handle that cleanly.
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
New MPTCP sockets will return -ENOPROTOOPT if MPTCP support is disabled
for the current net namespace.
We are providing here a way to control access to the feature for those
that need to turn it on or off.
The value of this new sysctl can be different per namespace. We can then
restrict the usage of MPTCP to the selected NS. In case of serious
issues with MPTCP, administrators can now easily turn MPTCP off.
Co-developed-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the current sendmsg() lands on the same subflow we used last, we
can try to collapse the data.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With the previous patch in place, the msk can detect which subflow
has the current map with a simple walk, let's update the main
loop to always select the 'current' subflow. The exit conditions now
closely mirror tcp_recvmsg() to get expected timeout and signal
behavior.
Co-developed-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Co-developed-by: Davide Caratti <dcaratti@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Co-developed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add new SEND_SPACE flag to indicate that a subflow has enough space to
accept more data for transmission.
It gets cleared at the end of mptcp_sendmsg() in case ssk has run
below the free watermark.
It is (re-set) from the wspace callback.
This allows us to use msk->flags to determine the poll mask.
Co-developed-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Parses incoming DSS options and populates outgoing MPTCP ACK
fields. MPTCP fields are parsed from the TCP option header and placed in
an skb extension, allowing the upper MPTCP layer to access MPTCP
options after the skb has gone through the TCP stack.
The subflow implements its own data_ready() ops, which ensures that the
pending data is in sequence - according to MPTCP seq number - dropping
out-of-seq skbs. The DATA_READY bit flag is set if this is the case.
This allows the MPTCP socket layer to determine if more data is
available without having to consult the individual subflows.
It additionally validates the current mapping and propagates EoF events
to the connection socket.
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Co-developed-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Co-developed-by: Davide Caratti <dcaratti@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Per-packet metadata required to write the MPTCP DSS option is written to
the skb_ext area. One write to the socket may contain more than one
packet of data, which is copied to page fragments and mapped in to MPTCP
DSS segments with size determined by the available page fragments and
the maximum mapping length allowed by the MPTCP specification. If
do_tcp_sendpages() splits a DSS segment in to multiple skbs, that's ok -
the later skbs can either have duplicated DSS mapping information or
none at all, and the receiver can handle that.
The current implementation uses the subflow frag cache and tcp
sendpages to avoid excessive code duplication. More work is required to
ensure that it works correctly under memory pressure and to support
MPTCP-level retransmissions.
The MPTCP DSS checksum is not yet implemented.
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Co-developed-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
set/getsockopt behaviour with multiple subflows is undefined.
Therefore, for now, we return -EOPNOTSUPP unless we're in fallback mode.
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Call shutdown on all subflows in use on the given socket, or on the
fallback socket.
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Generate the local keys, IDSN, and token when creating a new socket.
Introduce the token tree to track all tokens in use using a radix tree
with the MPTCP token itself as the index.
Override the rebuild_header callback in inet_connection_sock_af_ops for
creating the local key on a new outgoing connection.
Override the init_req callback of tcp_request_sock_ops for creating the
local key on a new incoming connection.
Will be used to obtain the MPTCP parent socket to handle incoming joins.
Co-developed-by: Davide Caratti <dcaratti@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add subflow_request_sock type that extends tcp_request_sock
and add an is_mptcp flag to tcp_request_sock distinguish them.
Override the listen() and accept() methods of the MPTCP
socket proto_ops so they may act on the subflow socket.
Override the conn_request() and syn_recv_sock() handlers
in the inet_connection_sock to handle incoming MPTCP
SYNs and the ACK to the response SYN.
Add handling in tcp_output.c to add MP_CAPABLE to an outgoing
SYN-ACK response for a subflow_request_sock.
Co-developed-by: Davide Caratti <dcaratti@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add hooks to tcp_output.c to add MP_CAPABLE to an outgoing SYN request,
to capture the MP_CAPABLE in the received SYN-ACK, to add MP_CAPABLE to
the final ACK of the three-way handshake.
Use the .sk_rx_dst_set() handler in the subflow proto to capture when the
responding SYN-ACK is received and notify the MPTCP connection layer.
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use ULP to associate a subflow_context structure with each TCP subflow
socket. Creating these sockets requires new bind and connect functions
to make sure ULP is set up immediately when the subflow sockets are
created.
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Co-developed-by: Davide Caratti <dcaratti@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Implements the infrastructure for MPTCP sockets.
MPTCP sockets open one in-kernel TCP socket per subflow. These subflow
sockets are only managed by the MPTCP socket that owns them and are not
visible from userspace. This commit allows a userspace program to open
an MPTCP socket with:
sock = socket(AF_INET, SOCK_STREAM, IPPROTO_MPTCP);
The resulting socket is simply a wrapper around a single regular TCP
socket, without any of the MPTCP protocol implemented over the wire.
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Co-developed-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
