This function is only used in sockets, thus making it a private function
of socket library and renaming it relevantly.
Note that sockets should be reviewed at some point to avoid using such
function: zsock_received_cb() already get the ip header and the protocol
header, so it could grab the src addr/port from there. It would be way
more optimized to do so, since net_pkt_get_src_addr is costly as it
parses all over again the ip/protocol headers.
utils unit test is updated and the test of the former
net_pkt_get_src_addr/net_pkt_get_dst_addr are removed.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
For networking code, it does not make sense to have zero net_buf
or net_pkt instances. Make sure this is enforced by code.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
pkt->buffer is represented by 1+ net_buf. If some are unused, this will
deallocates them.
This situation can happen on TCP where net_pkt allocator evaluates the
header size to its maximum size. Which space might not be (fully) used
in the end. On fixed data size buffer, this might end up by having last
buffer(s) not bein used. So better removing those.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
This will take into account the family and the protocol, as well as
existing buffer occupation, to return the available buffer space that
can be used for payload.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
No need to enable IPv4 any more as that is now optional.
This saves some memory as the application can work without
IPv4, IPv6, UDP or TCP.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
This commit fixes compilation warnings if user disables
CONFIG_NET_IPV4, CONFIG_NET_IPV6, CONFIG_NET_TCP and
CONFIG_NET_UDP.
E.g Samples like packet-socket doesn't need above configuration.
Signed-off-by: Ravi kumar Veeramally <ravikumar.veeramally@linux.intel.com>
When creating net_pkt, the default value of net_pkt data_len was
set to zero. In case of RAW sockets, when family is AF_PACKET,
the data_len will be zero as there is no higher level protocol
information. So in this case, the default value of data_len
must be set to interface MTU
Signed-off-by: Ravi kumar Veeramally <ravikumar.veeramally@linux.intel.com>
In case of Ethernet, if the requested size is larger than MTU and if
AF_UNSPEC is provided, the allocator will need to take into account
the ethernet header size which is not accounted in the MTU.
Other current L2 do not follow that rule as their MTU is based on IP
one (IPv6 most of the time). What they declare as MTU is the full frame
size they handle (minus the FCS for instance in 15.4). So with
AF_UNSPEC, such assumption on L2 header size is unrelevant.
(On 15.4 the header size is variable anyway and cannot be known until
the frame is parsed).
Fixes#12982
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Now that net_pkt are accessed through a common r/w API, using below a
net_pkt_cursor, let's have an option that will reset this cursor once
the net_pkt is freed.
Result is instead of segfaulting on r/w access, these operations will
bail out properly. Subsequent, and logical (unless you have a leak
which is another issue) net_pkt_unref will tell you who/where the pkt
was freed. Without it, you will get a segfault for instance, but that
won't tell you the exact reason. This options can help you then.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Extension headers are specific and use nexthdr attribute from net_pkt.
next_header_proto is the user protocol (if any) that might require
finalization too (calculating the checksum).
However, if there are extension header, we need to skip those and jump
to the position where the user protocol starts.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Also, store the actual next_hdr value and not it's position.
This permits to reduce net_pkt from some bytes.
Such field was unused until now, but it will be soon.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
These ones would support linearizing non-contiguous area, however
requiring a bit more complex type as an "accessor".
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Adding a cursor into net_pkt. This is used to read/write data in a much
simpler way, for pre-allocated buffers in net_pkt. This avoids API users
to deal with net_buf below directly.
However, to be used - as for the new allocators - it will require deep
net stack core and API changes.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
These struct net_pkt allocators will give the possibility to allocate at
once the net_pkt and the buffer associated with, taking care of the
header space and MTU relevantly.
This enables to use the variable length allocator from net_buf. However,
it is not yet the default and is set as experimental.
As it is provided in parallel to existing allocators, it has to keep a
slab per-direction and thus a pointer in net_pkt, as well as appdata,
appdatalen etc... Resulting in "bloating" net_pkt. This will be solved
when, finally, former allocators will be removed.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
The previous commit replaced the net_pkt element ref with an element
atomic_ref. CI tests turned up more places where ref was used directly.
This commit converts them to use the new element.
Signed-off-by: Daniel Glöckner <dg@emlix.com>
It has been observed that some network drivers, f.ex. the SAM E70 GMAC,
call net_pkt_unref from inside the interrupt that signals the successful
transmission of a packet. This conflicts with the net_pkt_unref call
made by ethernet_send after the packet has been given to the driver.
We fix this by using an atomic_t to hold the reference count as there
might be other, difficult to find cases of net_pkt_(un)ref being used
across threads and interrupts.
The name of the element has been changed from "ref" to "atomic_ref" to
cause a compile error when code still has not been converted to use the
atomic_* functions.
Fixes#12708
Signed-off-by: Daniel Glöckner <dg@emlix.com>
The NET_LOG_LEVEL is enabled also when allocation debugging is
enabled but when net_pkt debugging is not. Thus we need to use
the CONFIG_NET_PKT_LOG_LEVEL when printing normal debug log
prints.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
net_frag_linearize() is just a wrapper for net_buf_linearize(). As
the latter was refactored to never return error, and instead just
return actual copied length, update the former and its usages too.
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
The slab2str function which is used in the debug log call is not defined
unless NET_LOG_PKT_LOG_LEVEL >= LOG_LEVEL_DEBUG.
It looks like this change was accidentally introduced in #11374.
Signed-off-by: Tomasz Gorochowik <tgorochowik@antmicro.com>
ll reserve is of no use as the l2 is allocating what it needs for
filling in its header.
This is another step forward to removing ll reserve concept.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
The net-shell "net allocs" command should print network buffer
allocations even if network packet debugging is not enabled.
This is how it used to work earlier but the behaviour got lost
at some point. So user needs to set CONFIG_NET_DEBUG_NET_PKT_ALLOC
in order to see the buffer allocations. The option will be enabled
by default if network packet log level is set to DBG.
The reason for a separate option is that the network packet debug
logging prints just too much data and it is very difficult to
track allocations when that happens.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Remove network specific default and max log level setting
and start to use the zephyr logging values for those.
Remove LOG_MODULE_REGISTER() from net_core.h and place the
calls into .c files. This is done in order to avoid weird
compiler errors in some cases and to make the code look similar
as other subsystems.
Fixes#11343Fixes#11659
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
In earlier commit 15e7e3ea4 ("net: ip: Split debug prints into
smaller pieces"), the net_pkt debug prints were split to two
lines because of the argument count limitation in logging system.
As the logging subsystem increased the limit count in
commit 62d011549a ("logging: Support for up to 15 arguments in log
message") we can restore the original version as it is easier
to read.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Cache the used transport protocol in net_pkt. This way we can
avoid traversing IP header to get the last protocol in network
packet. This is mostly an issue in IPv6 which can have a long
list of extension headers after IPv6 header and before the
transport protocol header.
Fixes#10853
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Currently logging subsystem supports quite small number of function
parameters. So split some long functions into smaller pieces.
Hopefully this is just a temporary patch and we can support more
parameters to logging macros.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Instead of one global log level option and one on/off boolean
config option / module, this commit creates one log level option
for each module. This simplifies the logging as it is now possible
to enable different level of debugging output for each network
module individually.
The commit also converts the code to use the new logger
instead of the old sys_log.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
The return of memset is never checked. This patch explicitly ignore
the return to avoid MISRA-C violations.
The only directory excluded directory was ext/* since it contains
only imported code.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
1. Where we calculate max size, name variable (preprocessor define)
correspondingly.
2. Calling TCP/UDP an "app protocol" is original, use "next protocol"
terminology of IPv6.
3. As headers go as IP, then "next", order calculations that way too.
4. Add more comments.
Addresses: #8723
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
Some locations like DHCPv4 client create a prefilled packet by appending
new fragments in a loop with one byte each via net_pkt_append_u8() which
is wasteful and noisy. This patch adds the new functions
net_pkt_append_memset() which creates fragments as needed in the desired
size and initialises it to the specified value.
This change also adds a unittest for the new function.
Prerequisite for #9287
Signed-off-by: Daniel Egger <daniel@eggers-club.de>
Current implementation only considers IP header length while setting
appdata value on a cloned packet. It will give bogus value if original
packet contains extension headers and if extension headers are large
(i.e. more than one fragment). Only consider appdata length from the
original packet.
Signed-off-by: Ravi kumar Veeramally <ravikumar.veeramally@linux.intel.com>
This change moves the logic for linearize and append_bytes from
the net_pkt sources into the net_buf sources where it can be
made available to layers which to not depend on net_pkt. It also,
adds a new net_buf_skip() function which can be used to iterated
through a list of net_buf (freeing the buffers as it goes).
For the append_bytes function to be generic in nature, a net_buf
allocator callback was created. Callers of append_bytes pass in
the callback which determines where the resulting net_buf is
allocated from.
Also, the dst buffer in linearize is now cleared prior to copy
(this was an addition from the code moved from net_pkt).
In order to preserve existing callers, the original functions are
left in the net_pkt layer, but now merely act as wrappers.
Signed-off-by: Michael Scott <mike@foundries.io>
Current implementation does not handle large extension headers
(e.g HBHO). Which resulted network stack crashes or due to
misinterpretation of lengths network packets are dropped. Also
caused issues while preparing IPv6 packet (e.g. large HBHO header
with IPv6 fragmentation support).
Issues fixed and provided more unit tests.
Signed-off-by: Ravi kumar Veeramally <ravikumar.veeramally@linux.intel.com>
Let's set it by default when allocating net_pkt. A macro will avoid
ifdefs as well
CONFIG_NET_TX_DEFAULT_PRIORITY is always defined.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Similar to UDP, some drivers can make use of the following functions:
net_tcp_get_hdr()
net_tcp_set_hdr()
Let's expose them as <net/tcp.h> and change all internal references
to "tcp_internal.h".
Signed-off-by: Michael Scott <michael@opensourcefoundries.com>
This allows creation of virtual lan (VLAN) networks. VLAN support is
only available for ethernet network technology.
Fixes#3234
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
SYN flood causes crash in RX thread due to NULL pointer access. After
the crash available RX memory is zero, hence echo server does not
respond to echo request.
Signed-off-by: Ruslan Mstoi <ruslan.mstoi@intel.com>
With this commit it is possible to add priority to sent or received
network packets. So user is able to send or receive higher priority
packets faster than lower level packets.
The traffic class support is activated by CONFIG_NET_TC_COUNT option.
The TC support uses work queues to separate the traffic. The
priority of the work queue thread specifies the ordering of the
network traffic. Each work queue thread handles traffic to one specific
work queue. Note that you should not enable traffic classes unless
you really need them by your application. Each TC thread needs
stack so this feature requires more memory.
It is possible to disable transmit traffic class support and keep the
receive traffic class support, or vice versa. If both RX and TX traffic
classes are enabled, then both will use the same number of queues
defined by CONFIG_NET_TC_COUNT option.
Fixes#6588
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Redesign of the net_buf_simple and net_buf structs, where the data
payload portion is split to a separately allocated chunk of memory. In
practice this means that buf->__buf becomes a pointer from having just
been a marker (empty array) for where the payload begins right after
the meta-data.
Fixes#3283
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
We shouldn't limit the amount of data appended to RX packets based on
the max send size of TX packets. Skip this check for packets in the
RX slab.
Signed-off-by: Michael Scott <michael@opensourcefoundries.com>
Commit 753daa6 ("net: pkt: Compute TX payload data length")
removed the default packet setup on incoming packets when they
belong to the rx_pkt pool.
Let's restore this behavior, as MBEDTLS processing in net_app library
needs to use packet family to determine IP header length on
incoming packets.
NOTE: A future cleanup patch could set the IP header length based
on the context IP family. However, there are many places in the code
where this is being set, so care should be taken.
Signed-off-by: Michael Scott <michael@opensourcefoundries.com>
The maximum data length that can be appended using net_pkt_append()
should be set to TCP send_mss only if it is smaller than allowed
payload length in net_pkt.
Signed-off-by: Vakul Garg <vakul.garg@nxp.com>
For calculating amount of payload data that can be added in a packet,
we need to subtract IPv6 or IPv4 header lengths from MTU.
Signed-off-by: Vakul Garg <vakul.garg@nxp.com>
When net debugging is enabled, the count variable is initialized to -1.
This may cause division by zero if there is only one fragment in pkt.
Solve this by setting the count to 0 and checking the value before the
print at the end of the function.
Successfully tested on STM32F407 SoC.
Signed-off-by: Ding Tao <miyatsu@qq.com>
Compute the length of the TX payload that is transported in one
IPv4 or IPv6 datagram taking into account UDP, ICMP or TCP
headers in addition to any IPv6 extension headers added by RPL.
The TCP implementation in Zephyr is known to currently carry at
maximum 8 bytes of options. If the protocol is not known to the
stack, assume that the application handles any protocol headers
as well as the data. Also, if the net_pkt does not have a
context associated, length check on the data is omitted when
appending.
Although payload length is calculated also for TCP, the TCP MSS
value is used as before.
Define IPv4 minimum MTU as 576 octets, See RFC 791, Sections 3.1.
and 3.2.
Signed-off-by: Patrik Flykt <patrik.flykt@intel.com>
- Renaming NET_L2_RAW_CHANNEL to NET_RAW_MODE
- Create a generic IEEE 802.15.4 raw mode for drivers
- Modify the IEEE 802.15.4 drivers so it passes the packet unmodified,
up to code using that mode to apply the necessary changes on the
received net_pkt according to their needs
- Modify wpanusb/wpan_serial relevantly
Fixes#5004
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
If we were asked to add 10KB to a packet, adding it won't help -
such packet won't be even sent by hardware on our side, and if
it is, it will be dropped by receiving side. So, make sure we
never add more data than MTU as set for the owning interface.
This actually gets a bit tricky, because we need also to account
for protocol header space. Typically, when net_pkt_append() is
called, protocol headers aren't even added to packet yet (they
are added in net_context_send() currently), so we have little
choice than to assume the standard header length, without any
extensions.
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
The commit 971da9d0 ("net: pkt: adjust_offset: Simplify and optimize
code") changed the adjust_offset() function but left the error print
intact. This print is now invoked even if there is no error which
looks bad in debug prints.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
An edge condition was handled in a special way, even though the main
condition covered it well. More code, more jumps == slower code,
bigger binaries.
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
Rename net_pkt_get_src_addr() to net_pkt_get_addr() and make it able to
handle source or destination address.
Signed-off-by: Aska Wu <aska.wu@linaro.org>
net_pkt_tcp_data() and net_pkt_udp_data() simply returns the start
address of the header. However the header may span over multiple
fragments, unexpected data or memory corruption might happen when
reading or writing to the pointer directly.
Use net_tcp_get_hdr() and net_udp_get_hdr() instead.
Signed-off-by: Aska Wu <aska.wu@linaro.org>
Previously net_pkt.h, defined macros NET_PKT_TX_SLAB_DEFINE,
NET_PKT_DATA_POOL_DEFINE, but advertised them as intended for
"user specified data". However, net_pkt.c effectively used the
same parameters for slabs/pools, but this wasn't obvious due
to extra config param redirection. So, make following changes:
1. Rename NET_PKT_TX_SLAB_DEFINE() to NET_PKT_SLAB_DEFINE()
as nothing in its definition is TX-specific.
2. Remove extra indirection for config params, and use
NET_PKT_SLAB_DEFINE and NET_PKT_DATA_POOL_DEFINE to define
system pools.
3. Update docstrings for NET_PKT_SLAB_DEFINE and
NET_PKT_DATA_POOL_DEFINE.
Overall, this change removes vail of magic in the definition of
system pkt slabs/pools, making obvious the fact that any packet
slabs/pools - whether default system or additional, custom - are
defined in exactly the same manner (and thus work in the same manner
too).
Fixes#4327
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
Introduce net_pkt_get_src_addr() as a helper function to get the source
address and port from the packet header.
Signed-off-by: Aska Wu <aska.wu@linaro.org>
If the data parameter in net_pkt_insert() is NULL, then just
insert amount of data but clear the area instead of copying.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
It is useful to clone just the net_pkt which does not have any
data fragments linked to it.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Fixes a tx_pkts slab leak since the cloned pkt was referencing the
original pkt slab but was not originated from it (net_pkt_unref uses
pkt->slab when releasing the pkt).
Signed-off-by: Ricardo Salveti <ricardo.salveti@linaro.org>
When trying to figure out where ICMP / TCP / UDP header is located
in net_buf, print more information about what we were doing if the
header cannot fit the first fragment. This is much needed infomation
in debugging weird issues.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
This is needed when one wants to copy the whole fragment chain
and its head pointer net_pkt.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
This option enables full debugging output for memory allocations.
As that can produce lot of output and slow down the device under test,
it is disabled by default.
The previous CONFIG_NET_DEBUG_NET_PKT will collect information about
memory allocations but will not print any output. Use "net mem" or
"net allocs" commands in net-shell to see the memory allocation status.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Instead of printing [net/net_pkt], print [net/pkt] if debug log
is enabled for network packet allocator. The double net in earlier
print is redundant information.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
We had various asserts when checking network packet length but
printed also error when there was none. Fix this by checking
do we really have a too short message.
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>
Remove NET_ICMP_HDR() macro as we cannot safely access ICMP header
via it if the network packet header spans over multiple net_buf
fragments.
Jira: ZEP-2306
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Moving the net_buf_pool objects to a dedicated area lets us access
them by array offset into this area instead of directly by pointer.
This helps reduce the size of net_buf objects by 4 bytes.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The net_pkt_split() was incorrectly checking fragA pointer
even before it was allocated.
The unit test is fixed and converted to ztest.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
If we could not split the packet properly, make sure that the
fragments that we managed to allocate are unreffed and marked
as NULL.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
This helper copies desired amount of data from network packet
buffer info a user provided linear buffer.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
For stream-based protocols (TCP), adding less data than requested
("short write") is generally not a problem - the rest of data can
be sent in the next packet. So, make net_pkt_append() return length
of written data instead of just bool flag, which makes it closer
to the behavior of POSIX send()/write() calls.
There're many users of older net_pkt_append() in the codebase
however, so net_pkt_append_all() convenience function is added which
keeps returning a boolean flag. All current users were converted to
this function, except for two:
samples/net/http_server/src/ssl_utils.c
samples/net/mbedtls_sslclient/src/tcp.c
Both are related to TLS and implement mbedTLS "tx callback", which
follows POSIX short-write semantics. Both cases also had a code to
workaround previous boolean-only behavior of net_pkt_append() - after
calling it, they measured length of the actual data added (but only
in case of successful return of net_pkt_append(), so that didn't
really help). So, these 2 cases are already improved.
Jira: ZEP-1984
Change-Id: Ibaf7c029b15e91b516d73dab3612eed190ee982b
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@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>
This is actually useless as there is only 1 RX memory slab, and thus can
be removed to reduce net_pkt structure size.
Change-Id: I62d716515120e7356ee1e2d75bbe1ec32e22c35d
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>
