The network packet ref count was not properly increased when
the TCP was retried. This meant that the second time the packet
was sent, the device driver managed to release the TCP frame even
if we had not got ACK to it.
Somewhat long debug log follows:
The net_pkt 0x08072d5c is created, we write 1K data into it, initial ref
count is 1.
net_pkt_write: pkt 0x08072d5c data 0x08075d40 length 1024
net_tcp_queue_data: Queue 0x08072d5c len 1024
net_tcp_trace: pkt 0x08072d5c src 5001 dst 5001
net_tcp_trace: seq 0x15d2aa09 (366127625) ack 0x7f67d918
net_tcp_trace: flags uAPrsf
net_tcp_trace: win 1280 chk 0x0bea
net_tcp_queue_pkt: pkt 0x08072d5c new ref 2 (net_tcp_queue_pkt:850)
At this point, the ref is 2. Then the packet is sent as you see below.
net_pkt_ref_debug: TX [13] pkt 0x08072d5c ref 2 net_tcp_queue_pkt():850
net_tcp_send_data: Sending pkt 0x08072d5c (1084 bytes)
net_pkt_unref_debug: TX [13] pkt 0x08072d5c ref 1 (ethernet_send():597)
Ref is still correct, packet is still alive. We have not received ACK,
so the packet is resent.
tcp_retry_expired: ref pkt 0x08072d5c new ref 2 (tcp_retry_expired:233)
net_pkt_ref_debug: TX [10] pkt 0x08072d5c ref 2 tcp_retry_expired():233
net_pkt_unref_debug: TX [10] pkt 0x08072d5c ref 1 ... (net_if_tx():173)
net_pkt_unref_debug: TX [10] pkt 0x08072d5c ref 0 ... (net_if_tx():173)
Reference count is now wrong, it should have been 1. This is because we
did not increase the ref count when packet was placed first time into
sent list in tcp.c:tcp_retry_expired().
The fix is quite simple as you can see from this commit.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
This patch fixes following issues.
* If IPv6 neighbor table is full, stack can not add any new
neighbors. So stale counter is introduced. Whenever neighbor
enters into STALE state, stale counter will be incremented
by one. When table is full and if stack wants to add new
neighbor, oldest neighbor in STALE state will be removed
and new neighbor will be added.
* When neighbor is in PROBE state and when it exceeds max
number of PROBEs, only neighbor with router is removed.
As per RFC 4861 Appendix C, entry can be discarded. Now
neighbor will be removed from the table.
* Reachability timer has an issue. e.g. if a first entry timer
is 10 seconds, after 3 seconds, a new entry added with
only 3 seconds. But current implementation does not check
whether remaining time of current left over timeout is more
than new entry timeout or not. In this example, when new entry
timeout is 3 seconds, left over timeout from first etnry is
still 7 seconds. If k_delayed_work_remaining_get() returns
some value then new entry time out was not considered.
Which is bad. It fixed now.
* nbr_free is used sometimes to remove the neighbor. Which does
not remove route if that particulat neighbor is route to some
other neighbor. net_ipv6_nbr_rm() should be used in such places.
* Trivial changes which does not affect functionality.
Fixes#14063
Signed-off-by: Ravi kumar Veeramally <ravikumar.veeramally@linux.intel.com>
Fix regression and allow incoming packet when source and
destination port numbers are the same.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
In case of Ethernet for instance, the MTU is larger than the minimal
IPv6 MTU, so it is not required to fragment a packet that fits in
Ethernet MTU.
Fixes#14659
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
The "is this packet for us?" filter in net_ipv4_input() has a minor
logic error which fails to discard many packets which are.. not for us.
Fixes: #14647
Signed-off-by: Charles E. Youse <charles.youse@intel.com>
The context_alloc_pkt() might run out of memory, and if that
happens we must not try to set the context pointer in it.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Since the new packet flow came in, payload comes at the end so udp
length for instance is known only when we "finalize" the packet.
However such finalization was still under the condition of chksum
offload, like it used to be in the former flow (udp headers were
inserted). This is obviously wrong but that was not caught with
existing driver in master as none of these drivers offloading
chksum calculation.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Commit fc27a81ed2 ("net: context: Select proper network interface
when binding") moved the bind_default() call to after the remote
address was set for the net_context.
This and a later net_pkt API refactor broke net_offload() handling
so that context->iface wasn't set and context->flags didn't have
NET_CONTEXT_REMOTE_ADDR_SET correctly.
Let's fix this by relocating the net_offload handling to after
these have happened.
Signed-off-by: Michael Scott <mike@foundries.io>
net_offload API is left untouched, so it still takes a net_pkt as input
for the buffer. This is under-optimized since offload drivers will copy
the data from that net_pkt back into contiguous buffer again.
Let's tackle this issue another time.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
This attribute, in case CONFIG_NET_STATISTICS is enabled, made sense
when L2's send() function did not return the length of the sent packet.
But now, it's a superflous optimization as is it used only to set the
stats on recv or send, where net_pkt_get_len() can be used directly.
This helps to save 2 bytes from struct net_pkt.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
And also to the relevant callbacks.
That parameter is not used anywhere so it is useless.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Seems like a useless attribute. Since net_context is not being used by
the user directly (socket is the unique interface now) and since no core
parts uses the token parameter of net_context API: let's remove the
attribute.
This helps to save 4 bytes from struct net_pkt.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Now that legacy - and unrelated - function named net_pkt_get_data has
been removed, we can rename net_pkt_get_data_new relevantly.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
That function was responsible for allocating new buffer element, but it
is now unused and can be removed safely. Buffer allocation is now done
via net_pkt_alloc_buffer().
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Now that legacy functions are removew, let's rename the new functions by
removing the _new suffix.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
There is no need for these anymore: all is dictated by the position of
the net_pkt's cursor now
- actual cursor position is like the former appdata attribute
- net_pkt_remaining_data() is like the former appdatalen attribute
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Former net_context_send/sendto and net_context_create_ipv4/ipv6 are now
unused and can be removed.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
IPv6 next header might be something else (here NET_IPV6_NEXTHDR_HBHO)
but when finalizing it is mandatory to give the actual last header
protocol type. In this case IPPROTO_ICMPV6, so the checksum can be
computed properly then by net_icmpv6_finalize() called from
net_ipv6_finalize().
Fixes#14663
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
As the L2 layer might have modified the cursor, reset it here
before giving the packet to promiscuous mode API. This way
the application will get a fresh copy of the buffer.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
TCP context is now created with refcount of 2, signifying that it's
jointly owned by an app and stack. Thus, net_context_put()
unconditionally calls net_context_unref() to decrement refcount on
app's behalf, and leaves stack's refcount to internal routines
which handle sending/receiving/timing out FINs, etc.
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
Now that IPv4 options are handled, ICMPv4 echo reply must be created
taking into account that IPv4 header length can be variable. So instead
of cloning and rewriting (that would copy the useless options), let's
allocate and copy only the payload.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
IPv4 header length might be bigger than struct net_ipv4_hdr if there are
options appended to it.
Fixes#11618
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
IPv4 header might come with options, unlike IPv6, these are not
encapsulated in option header but are fully part of the IPv4 header.
Zephyr must handles these. Now silently ignoring their content and
setting the cursor to the payload properly.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
This is probably the only place where net_pkt_alloc_from_slab() is going
to be used.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
This is meant for very particular use case as only logging uses that.
Where it makes entirely sense for it to send the logs through its own
slab/pool in order to not drain the core slabs/pools.
So enabling the new API to manage that. That has to be used with
net_context for the buffer pool. So one has to first allocate the
net_pkt from external slab, set the context and then (and only then)
allocate buffer. Basically, only net_context will uses that scheme
anyway.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Though core system is able to manage packet timestamping internaly (gptp
requires it for instance), it might be necessary to enable/disable
packet timestamping from net context directly.
Currently this will be only used by the tx timestamp test. So this
support is disabled by default. (And gptp does not require it anyway).
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
This is required if traffic class is enabled, so allocated packets from
net_context do get the right priority set.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
That can be useful on some tests which will not have any interface but
still allocate net_pkt. Also, one may allocate a packet with buffer not
knowing yet the interface it will be send through.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
subsys/net/lib/lwm2m/Kconfig.ipso is 'source'd within an 'if LWM2M', in
subsys/net/lib/lwm2m/Kconfig, so the 'depends on LWM2M' is redundant.
The 'depends on NET_IPV4' and 'depends on NET_L2_OPENTHREAD' are within
corresponding 'if's in the same file.
'if FOO' is just shorthand for adding 'depends on FOO' to each item
within the 'if'. Dependencies on menus work similarly. There are no
"conditional includes" in Kconfig, so 'if FOO' has no special meaning
around a source. Conditional includes wouldn't be possible, because an
if condition could include (directly or indirectly) forward references
to symbols not defined yet.
Tip: When adding a symbol, check its dependencies in the menuconfig
('ninja menuconfig', then / to jump to the symbol). The menuconfig also
shows how the file with the symbol got included, so if you see
duplicated dependencies, it's easy to hunt down where they come from.
Signed-off-by: Ulf Magnusson <Ulf.Magnusson@nordicsemi.no>
As per RFC2131 4.1.1 requires we wait a random period
between 1 and 10 seconds before sending the initial
discover. But tests can not wait that longer. So this
option helps test to configure the value to minimum.
Signed-off-by: Ravi kumar Veeramally <ravikumar.veeramally@linux.intel.com>
Multiple flag bits were set so the ACK flag set was not checked
properly which meant that connection establishment was not
successfull.
Fixes#13943
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
No need to hide the symbols in the header file if CONFIG_NET_LLDP
is not enabled. This also allows the documentation to be generated
properly.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
The ipv6_handle_ext_hdr_options() can return negative value
but we stored it into unsigned variable and then checked < 0.
Coverity-CID: 190995
Fixes#13830
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
