This patch fixes a regression which the original patch introducing
this code (improving concurrent connection handling) had on
*sequential* connection handling. Without this patch, with the
default CONFIG_NET_TCP_BACKLOG_SIZE of 1, after each connection
request, there was 1s (ACK timeout) "dead time" during which new
connection wasn't ptocessed.
This is because k_delayed_work_remaining_get() was checked the
wrong way. But there's no need to use k_delayed_work_remaining_get()
at all, instead just call k_delayed_work_cancel() and dispatch on
its return code.
Note that there's still a problem of synchronizing access to
the global array tcp_backlog, as worker (which modifies it) may
preempt packet handling code (which also modifies it).
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
The commit 210c30805b ("net: context: Close connection fast
if TIME_WAIT support is off") was not a proper way of closing
the connection. So if Zephyr closes the connection (active close),
then send FIN and install a timer that makes sure that if the peer
FIN + ACK is lost, we close the connection properly after a timeout.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
The commit 00ac0487b0 ("net: context: Remove tcp struct SYN-ACK
timer handling") removed also the passive close ACK timer.
Adding that ACK timer back so that we can close the connection
properly even if the last ACK from peer is lost.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Remove NET_TCP_HDR() macro as we cannot safely access TCP header
via it if the network packet header spans over multiple net_buf
fragments.
Fixed also the TCP unit tests so that they pass correctly.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Remove NET_UDP_HDR() macro as we cannot safely access UDP header
via it if the network packet header spans over multiple net_buf
fragments.
Fixed also the UDP unit tests so that they pass correctly.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
net_context_unref() is protocol agnostic, after being un-referenced,
the same context might end up being used then for UDP,
if context->tcp is not reset to NULL, calling net_context_unref()
after this UDP usage will again try to release this TCP pointer
which might lead to random error.
Signed-off-by: june li <junelizh@foxmail.com>
The SYN-ACK timer is now handled by the TCP backlog functionality,
while the remaining ACKs for established connections use the tcp
struct ack timer. With this, code setting tcp struct SYN-ACK state
timers can now be removed.
Signed-off-by: Patrik Flykt <patrik.flykt@intel.com>
Add timer for sent TCP SYN-ACKs. If the timer is already
scheduled to run before canceling it is attempted, set
the cancelled flag and let the callback remove the
delayed work.
Signed-off-by: Patrik Flykt <patrik.flykt@intel.com>
Add an array of configurable size that holds TCP backlog entries.
The array is shared between all incoming TCP connections in order
to make it possible to get away with less memory consumed than
with a connection based approach.
The backlog entries are created when a SYN is received for a
listening TCP socket and removed once the corresponding ACK is
seen. With an incoming RST the corresponding backlog entry is
cleared, if any.
The size of the global backlog array is defined with the
CONFIG_NET_TCP_BACKLOG_SIZE Kconfig variable.
Signed-off-by: Patrik Flykt <patrik.flykt@intel.com>
In LISTEN state ignore a TCP RST. In SYN RCVD state, reset TCP
connection state to LISTEN when a valid RST segment is received. In all
other states close the connection - except that these other states will
not be handled in tcp_syn_rcvd() function.
Jira: ZEP-2279
Signed-off-by: Patrik Flykt <patrik.flykt@intel.com>
If the CONFIG_NET_TCP_TIME_WAIT support is disabled, then do not
start to wait for reply to sent FIN in active close, but unref
the corresponding net_context in order to close the connection
as soon as possible.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
From
http://pubs.opengroup.org/onlinepubs/7908799/xns/netinetin.h.html:
in_addr_t
An unsigned integral type of exactly 32 bits.
[] the in_addr structure [] includes at least the following member:
in_addr_t s_addr
In other words, POSIX requires s_addr to be a single integer value,
whereas Zephyr defines it as an array, and then access as s_addr[0]
everywhere. Fix that by following POSIX definition, which helps to
port existing apps to Zephyr.
Jira: ZEP-2264
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
In certain TCP states we should not try to send RESET segment
to peer. So check this and do not try to use NULL pkt to send
a message.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
If we cannot send network data, then print the error code when
printing debug information about the issue. This is needed when
debugging the issue.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Semantics of ENOENT error as used previously is "named entity not
found", whereas for "I/O handle is not valid", there's EBADF. For
example, POSIX/SUSV2 doesn't even list ENOENT as a possible error
for accept(), connect(), recv(), etc. whereas it lists EBADF, e.g.:
http://pubs.opengroup.org/onlinepubs/7908799/xns/connect.html
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
If we do not receive last ACK when the connection is tore down,
then do not wait forever as that would eat all the resources in
the network stack. So when we enter the LAST_ACK state, we setup
a timer that will unref the connection if the last ACK is not
received.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
The commit "net: tcp: Handle retransmitted packets from peer"
introduced over 80 character line that was missed in review.
Fixing it now.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
When we receive a packet with the sequence we already seen (and
processed), the most likely cause of it is that our ACK was lost,
and peer has to retransmit that packet. Then, we should just ACK
it, because otherwise peer will retransmit it again and again,
falling into exponential backoff and hosing the entire TCP
connection.
This makes changes to send_ack(), adding a flag to force sending
an ACK regardless of its cached status, and remove inline modifier,
as the function is big and called from many places.
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
We must check if we receive RST in any of the TCP states.
If we do not do this, then the net_context might leak as it
would never be released in some of the states. Receiving RST
in any TCP state is not described in TCP state diagram but is
described in RFC 793 which says in chapter "Reset Processing"
that system "...aborts the connection and advises the user and
goes to the CLOSED state."
We need to also validate the received RST and accept only those
TCP reset packets that contain valid sequence number.
The validate_state_transitions() function is also changed to
accept CLOSED state transition from various other states.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
When we send TCP data segment, we need to set the length
of the application data by calling net_pkt_set_appdatalen().
This is done so that sequence number can be properly
advanced when we receive ACK to that pending packet.
Signed-off-by: xiaorui hu <xiaorui.hu@linaro.org>
Convert code to use u{8,16,32,64}_t and s{8,16,32,64}_t instead of C99
integer types.
Jira: ZEP-2051
Change-Id: I4ec03eb2183d59ef86ea2c20d956e5d272656837
Signed-off-by: Kumar Gala <kumar.gala@linaro.org>
We need to set the connect_cb for the context before sending SYN packet.
This is required if we have a loopback connection in which case everything
is happening more or less synchronously and the connect_cb would not be
called in this case because its value would still be null.
For remote network connections this patch makes no difference.
Change-Id: Id7f837cd9e81cd79c4666c98cae84f6cb1a77af0
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
We must not let ACK timer to run if we have already cancelled it.
So keep track that the timer is cancelled and refuse to run it
if it was indeed cancelled. The reason why the timer might be run
in this case is because the timer might be scheduled to be triggered
after which one cannot cancel it.
Change-Id: I1c8b8cee72bc7a644e02db154d9d009b8d98ade2
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
The local port was set to 0 for IPv6 when registering
the connection handler. For IPv4, the code was not setting
the port in local_addr struct.
Change-Id: I82f0c08641a94d75d255ac306eca7bec6c332fba
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Most of these macros are not exactly exposing a buffer, but a specific
header pointer (ipv6, ivp4, ethernet and so on), so it relevant to
rename them accordingly.
Change-Id: I66e32f7c3f2bc75994befb28d823e24299a53f5c
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
- net_pkt becomes a stand-alone structure with network packet meta
information.
- network packet data is still managed through net_buf, mostly named
'frag'.
- net_pkt memory management is done through k_mem_slab
- function got introduced or relevantly renamed to target eithe net_pkt
or net_buf fragments.
- net_buf's sent_list ends up in net_pkt now, and thus helps to save
memory when TCP is enabled.
Change-Id: Ibd5c17df4f75891dec79db723a4c9fc704eb843d
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
There have been long lasting confusion between net_buf and net_nbuf.
While the first is actually a buffer, the second one is not. It's a
network buffer descriptor. More precisely it provides meta data about a
network packet, and holds the chain of buffer fragments made of net_buf.
Thus renaming net_nbuf to net_pkt and all names around it as well
(function, Kconfig option, ..).
Though net_pkt if the new name, it still inherit its logic from net_buf.
'
This patch is the first of a serie that will separate completely net_pkt
from net_buf.
Change-Id: Iecb32d2a0d8f4647692e5328e54b5c35454194cd
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Following flow does not work:
net_context_get
net_context_bind
..
..
net_context_put
net_context_get
net_context_bind
At instance of call to net_context_bind, conn_handler
is not NULL and returns with EISCONN.
This patch sets conn_handler to NULL in net_context_unref
Change-Id: I56a50839101b22161644b3cd7c5f510fa1abae3e
Signed-off-by: Mitul Shah <mitul.a.shah@intel.com>
We incorrectly changed the state from FIN_WAIT_1 to CLOSE_WAIT.
This caused ACK be sent in CLOSE_WAIT state when the connection
was closed by peer. Sending ACK in this state is not allowed
according to RFC. The connection was still closed but slightly
wrong way.
Jira: ZEP-1961
Change-Id: Ie4aa6818128d4190230b679e26ac9630c7d45d69
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
If the bind port is set to 0, then return the real bind port and
show it in the debug print.
Change-Id: If75b52bdacfc916329222d0d9e8aa4669e7a7160
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
This is part of a patch series to decouple TCP/IP offload, which
currently occurs at the transport layer (L4), from the data link
layer (L2).
- Rename struct net_l2_offload_ip to struct net_offload
- Rename struct field offload_ip to just offload
- Rename include/net/offload_ip.h -> include/net/net_offload.h
Change-Id: I3cd891c2b13e0e8f3ad1c66264f90b5031ae17c2
Signed-off-by: Gil Pitney <gil.pitney@linaro.org>
This is part of a patch series to decouple TCP/IP offload, which
currently occurs at the transport layer (L4), from the data link
layer (L2).
This patch simply renames the NET_L2_OFFLOAD_IP Kconfig variable
to NET_OFFLOAD.
Change-Id: Ic8b1d004cbac09b7c636475aaed75b0a31e4be1c
Signed-off-by: Gil Pitney <gil.pitney@linaro.org>
It's not enough to know just network address and port number to
disambiguate an endpoint, protocol (TCP vs UDP, etc.) should be
known too. Without this patch, there is a confusing output from
e.g. echo_server if both TCP and UDP is enabled.
Jira: ZEP-1086
Change-Id: I247a2dfc87df634ceac3b8733d7349b1d6aca80c
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
Current finalize api's takes buf as input parameter and returns
the finalized buf. But if there are any issue while finalizing,
it failed to throw an error.
Change-Id: I6db54b7453eec41a8051fab50d5c0dc937debd54
Signed-off-by: Ravi kumar Veeramally <ravikumar.veeramally@linux.intel.com>
When IPv6 header contains extra header (e.g. HBH), extra header length
was not part of app data length calculation. Result applications get
invalid total length. Also simplified set_appdata_values().
Change-Id: I07438d62dfd0f9abc1452484deb8b5eacdc37cf1
Signed-off-by: Ravi kumar Veeramally <ravikumar.veeramally@linux.intel.com>
If a UDP context had net_context_recv() called before
net_context_bind(), then it will have a stale connection handle
associated with initial (random) port number, while will be "bound"
to a new port as specified in net_context_bind(). So, it silently
won't behave as a user expects. net_context_bind() should really
update (or destroy/recreate) conn_handle in this case, but until
it's implemented, apply stopgap measure of at lease reporting error
back to user in this case.
Jira: ZEP-1644
Change-Id: I22ad55f94eaac487a4d5091ccbb24f973ec71553
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
Do not shortcut the sending side for TCP as we need to use
net_tcp_send_data() to send TCP data in order not to have
double unref in Bluetooth or IEEE 802.15.4 technologies which
has a special TCP packet sending logic.
Change-Id: I077db336c6335ccdbbafa4600f98388ecf33955f
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
User can configure its own pools for data that needs to
be transmitted out (TX). This helps to avoid deadlocking
the system if user space application uses all the buffers
in the system, and the core IP stack tries to get buffer
that needs to be sent out.
By default the net_buf pool support in net_context is turned
off as application developer needs to create the pools and
tie them to desired contexts.
Change-Id: Ida4a1771d34d6c250974e56fba4f0e0b2592cb29
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
This function is used only within net_context.c, make the symbol static
then.
Change-Id: Ib2c00a21c25e8c6a1404d6345d4b8ae05b779525
Signed-off-by: Jesus Sanchez-Palencia <jesus.sanchez-palencia@intel.com>
Since we try to match with POSIX behavior as much as possible, let's
not bother the user unless they need to be bothered. The recv(2) POSIX
syscall won't return 0 on stream sockets unless the connection has been
closed by the peer; however, that was happening with the callback set
by net_context_recv().
Change it so that the callback is never called if operating on a stream
socket and appdatalen is set to 0.
This is similar to a previous patch sent by Michael Scott in [1], but
not relying on the actual TCP flags: only on the appdatalen and the
socket type.
[1] https://gerrit.zephyrproject.org/r/#/c/9949/
Jira: ZEP-1632
Change-Id: Ib0c214fc9269d305a03e8d85eb606f106c45b038
Signed-off-by: Leandro Pereira <leandro.pereira@intel.com>
Due to commit fece856959 ("net: tcp: Clean up FIN handling") the
tcp_established() callback now handles TCP connections which are
in various ending/closing states other than TCP_ESTABLISHED.
Currently, these states are generating the following error and not
being processed:
Context 0x123456778 in wrong state 6.
(Shown when TCP is in LAST_ACK state).
This commit also fixes a memory leak issue discribed in
Jira: ZEP-1658
Analysis of the memory leak issue is here:
When TCP connection is established, tcp context is in
NET_TCP_ESTABLISHED state. Once it receives FIN message from client
it goes to NET_TCP_CLOSE_WAIT and then it turns to NET_TCP_LAST_ACK
after connection closing request from server. Now server gets final
ack from client, but tcp_established() will reject it because current
state is not in NET_TCP_ESTABLISHED. Even if server receives proper
ack, it is not handled by server. Hence 'sent_list' is not freed.
Change-Id: I41c8af2e6851809f87a02c271a4290cf3d823ebb
Signed-off-by: Michael Scott <michael.scott@linaro.org>
NET_ASSERT(net_nbuf_iface(buf)) should be called before setting
it on context [net_context_set_iface(context, net_nbuf_iface(buf))].
Change-Id: I9a1da1214857e96e03784bc98a9aae5cf59ef0fc
Signed-off-by: Ravi kumar Veeramally <ravikumar.veeramally@linux.intel.com>
Because of the change in next commit "net: if: Change
the iface param in net_if_ipv6_maddr_lookup",
we must initialize the network interface to NULL.
Without this the multicast address lookup will fail.
Change-Id: I113b44ce23c5f2ecbbf1698972078f102995e891
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
