The coap_client.h header won't build if included from c++ file:
error: expected primary-expression before ‘.’ token
.value[0] = coap_bytes_to_block_size(CONFIG_COAP_CLIENT_BLOCK_SIZE),
Therefore move the actual function implementation to the library C file
to prevent this.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
option struct was uninitialized in insert_option() function
during encode_options() call when the while loop wasn't
executed before.
Signed-off-by: Jakub Witowski <jakub.witowski9302@gmail.com>
The hdr_len was uninitialized when none of if-else
conditions were met in decode_delta() function.
Signed-off-by: Jakub Witowski <jakub.witowski9302@gmail.com>
The variable block_len in function coap_next_block_for_option()
can be used uninitialized in case of empty packet in
coap_packet_get_payload(). Mute the compiler warning with
default value of 0.
Signed-off-by: Dave Lacerte <lacerte.dave@hydroquebec.com>
Empty code was incorrectly matched as a request, fix that.
Align coap_handle_request_len() function to behave as documented in the
API documentation - in case of invalid request code (which is also the
case for empty code) -ENOTSUP Should be returned.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
The algorithm for matching request with response was incorrect, which
could lead to false matches (for example if request had a token, and
piggybacked reply had no token but matching message ID only, that would
still be counted as a match).
This commit fixes it. The request/reply matching is implemented based on
RFC now, with separate conditions for piggybacked/separate responses.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
If the CoAP server receives a message that doesn't fit into the receive
buffer, we should stop processing the message and respond to the client
with 4.13 "Request Entity too large".
Signed-off-by: Pieter De Gendt <pieter.degendt@basalte.be>
Non-confirmable CoAP requests need lifetime tracking as well
so we can free the structure after a timeout.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
When waiting for response after receiving the empty Ack, client
actually used way too timeout.
CoAP timeout only holds the timeout value in ms. t0 is the starting time.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
Add a new API to cancel just one, or mathing requests,
instead of cancelling all ongoing requests.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
poll() only for sockets that have traffic ongoing or have some lifetime
left.
On socket failures during a poll(), stop listening for the socket.
Application can recover by reconnecting the socket.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
When the client fails when parsing the response and we stop proceeding,
we should report that to the application.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
Even if we receive duplicate confirmable message, we should still
respond with the Ack. Just don't deliver the second callback.
This is achieved by moving the MID deduplication to after Ack handling.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
Return the -errno when zsock_sendto() or zsock_recvfrom() fails, so
rest of the code can deal with return values, instead of separately
comparing errno and return value.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
When response is received and handled, don't just clear the structure
but instead mark it as ongoing=false.
So if we later on receive a duplicate response for it, we can still
respond with Ack or Rst.
This is achieved by using release_internal_request() when we don't
expect any response for it and reset_internal_request() when we really
fill up a new request.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
In receiving thread, continuing the loops is based on
has_ongoing_exchanges() so it does not need atomic
coap_client_recv_active variable.
When idling, it wakes from semaphore. But there was potential
deadlock when coap_client_schedule_poll() would not signal the
semaphore, if atomic variable was already showing that it runs.
Removing the atomic variable removes this deadlock.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
It is error prone to flag separate booleans, so try to use
reset_internal_request() every time we release the internal request
structure.
Also refactor the reset_internal_request() so that we reset the
timeout value so it does not trigger again.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
If send() fails, we have not technically send the CoAP retry yet, so
restore the same pending structure, so our timeouts and retry counters
stay the same.
This will trigger a retry next time the poll() return POLLOUT, so we
know that we can send.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
Refactor the CoAP retry handling into the handle_poll() function,
so that we only try to send retries if the socket reports POLLOUT.
Also move the receiving into same loop, so when poll() reports POLLIN
we recv() the message and handle it before proceeding to other sockets.
Also fix tests to handle POLLOUT flag and add support for testing
multiple clients.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
When transmission of first request fails, reset the internal request
buffer as there is no ongoing CoAP transaction.
Application can deal with the failure.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
Before this patch, any unexpected socket error during poll (caused by
LTE disconnects for instance) would lead to a infinite loop because the
error state was never cleared, handled or even signaled to the user.
Signed-off-by: Benjamin Lindqvist <benjamin@eub.se>
CONFIG_ZVFS_POLL_MAX is now used to control the maximum number of poll()
entires. Thereby, CONFIG_NET_SOCKETS_POLL_MAX is redundant and shall
be deprecated.
Modify the defaults for NET_SOCKETS_POLL_MAX and ZVS_POLL_MAX so that
the deprecation actually makes sense instead of symbol removal. In case
the application still sets the old config, it will modify the
ZVS_POLL_MAX default.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
Signed-off-by: Chris Friedt <cfriedt@tenstorrent.com>
Extend the `coap_transmission_parameters` struct with the field
`ack_random_percent`. This was the last remaining CoAP transmission
parameter that was not configurable at runtime.
Signed-off-by: Adrian Friedli <adrian.friedli@husqvarnagroup.com>
Fix the following compilation warning given when using newlibc:
warning: 'response_truncated' may be used
uninitialized [-Wmaybe-uninitialized]
Issue is not seen with picolibc.
The variable was introduced as part of PR #76257
Signed-off-by: Tommi Rantanen <tommi.rantanen@nordicsemi.no>
Not all offloaded network stacks support this socket option so
control it using a Kconfig CONFIG_COAP_CLIENT_TRUNCATE_MSGS,
and enable it by default.
Signed-off-by: Pete Skeggs <peter.skeggs@nordicsemi.no>
This commit makes sure we continue to wait for extra confirmations even
after the request is done so we can handle duplicate confirmations if any.
Detailed description:
rfc7252#section-4.5 specifies that:
"The recipient SHOULD acknowledge each duplicate copy of a
Confirmable message".
So if, for example, the client sends to a multicast destination address,
the server will get multiple requests and will confirm all of them.
Without this commit, the client will set the request to done after
receiving the first answer.
From here the request object will be marked as free and the duplicate
acknowledgements will stay buffered in the network stack.
Once the client tries to send a new request, it will unbuffer those
duplicate acknowledgements but now the request object is unallocated
so the client won't be able to handle those acknowledgements as duplicates.
It will instead treat it as an unexpected ACK.
To work around this issue, rfc7252#section-4.8.2 states that:
"EXCHANGE_LIFETIME is the time from starting to send a Confirmable
message to the time when an acknowledgement is no longer expected,
i.e., message-layer information about the message exchange can be
purged."
Keeping the request object allocated for EXCHANGE_LIFETIME ensures that
duplicate acknowledgements can be handled accordingly.
This commit adds a basic implementation of what is stated in the RFC.
EXCHANGE_LIFETIME has been arbitrarily set to 3 * ACK_TIMEOUT which
seems more reasonable than the 247 seconds stated in the RFC.
Signed-off-by: Francois Gervais <francoisgervais@gmail.com>
Parse the more flag in coap_get_block2_option(), so that the function
can be used not only with requests but also with replies (where the more
flag should not be ignored).
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
The block number in block1/2 options can be encoded on up to 20 bits
according to RFC 7959, therefore the underlying type used in helper
functions to retrieve the block number should be large enough to hold
the result. Therefore, replace the container for block number with
uint32_t instead of uint8_t.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
Allow an application to add a Block2 option to an initial request for a
resource. For any subsequent requests as part of a blockwise transfer,
drop the application-added Block2 option since the coap_client must
append a Block2 option with updated NUM and SZX fields based on the
server response.
Signed-off-by: Matt Rodgers <mrodgers@witekio.com>
Limit the coap payload size passed up to the application callback to the
block size, when a block transfer is in progress and the current payload
is not the final block.
If the current payload is not part of a block transfer, or is the final
block of a transfer, then the full payload can be passed to the
application to avoid having to make another request over the network for
data that has already been received.
This avoids a problem raised in issue #76089, where a payload longer
than CONFIG_COAP_CLIENT_MESSAGE_SIZE causes the same data to be passed
to the application callback twice (once in the large packet, and once in
the next block which must have an offset that is a multiple of the block
size).
Signed-off-by: Matt Rodgers <mrodgers@witekio.com>
Use the MSG_TRUNC flag to check the total length of UDP packets
receieved by the coap_client, and hence check if the receive buffer
contains the whole message, or if it is truncated.
If the message is truncated, then use a blockwise transfer to fetch the
rest of the data.
This is related to issue #76089.
Signed-off-by: Matt Rodgers <mrodgers@witekio.com>
Introduce a more generic mutex for protecting coap_client structure.
This allows to avoid a certain race condition when sending consecutive
CoAP requests. The case was, that when a CoAP receive thread notified
the application that a complete response was received, and the
application wanted to send another request from the application thread,
the consecutive call to coap_client_req() might've failed if the
application thread has higher priority than the CoAP receive thread
because of the request context cleanup is done after calling the
application callback.
Having a mutex, which is locked while processing the response, and when
attempting to send a new request allows to synchronize threads as
expected.
As this new mutex seemed redundant with the more specialized send_mutex
already present in the coap_client, the latter was removed (i. e.
transmission is also protected with the new mutex).
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
An earlier pull request implementing observe support was merged too
hastily. It had a few issues:
1. The predicate for whether a request should be marked not ongoing was
wrong (it checked ret != 0 instead of ret < 0)
2. Without observes in mind, MID-based deduplication is not a required
feature. Deduplication was handled implicitly - the exchange would get
dropped after the first response anyway, so duplicate responses would
not get matched to anything. But with observes, there are several
responses in an exchange. This commit adds this.
3. Using coap_request_is_observe(&internal_req->request) in the response
handler requires the whole request to stay in scope for the lifetime of
the observation, which I observed was not always the case. Adding an
is_observe bool to the internal struct improved stability significantly.
With these fixes, GETs with observe option works very well.
Signed-off-by: Benjamin Lindqvist <benjamin@eub.se>
Improve coap_client_cancel_requests(). Ensure it can be
called from a callback. Report error to waiting callbacks.
Clear active flag.
This is useful when the network becomes unavailable
or prior to disconnecting in order to save power.
Signed-off-by: Pete Skeggs <peter.skeggs@nordicsemi.no>
The coap_client lib only handled "one-shot" requests properly. This
patch allows it to keep listening for additional responses to a request,
if the request was made with the CoAP OBSERVE option appended.
An API for canceling such requests is also added.
Signed-off-by: Benjamin Lindqvist <benjamin@eub.se>
Outgoing block-transfers now set the socket hint
to ONGOING as long as the BLOCK1/BLOCK2 header has
MORE flag set to true.
This means as only the last packet in the block-transfer
set the socket hint to LAST or ONE_RESPONSE.
Signed-off-by: Seppo Takalo <seppo.takalo@nordicsemi.no>
Changing remaining users of fcntl.h to use the include from our own
POSIX file so that the values in there are consistent in all parts
of the sources.
Signed-off-by: Jukka Rissanen <jukka.rissanen@nordicsemi.no>
