The commit 8d0ef1eb85 attempted to fix
test case MESH/SR/HM/CFS/BV-02-C, however inadvertently ended up
introducing a hidden bug. This bug was unearthed thanks to commit
686f5c79cf. We have to keep always track
of the FastPeriodDivisor state whether we're using it (faults > 0) or
not (faults == 0). Introduce a boolean field to the model publication
that's used to indicate whether the FastPeriodDivisor should be
applied or not, instead of zeroing the divisor when there are no
faults (this would cause wrong behavior when faults appear again).
Additionally, the PTS seems to require that we wait until the end of
the existing period before sending the next Health Current Status,
rather than sending it immediately when the fault count changes.
Fixes#15365
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Sending a model publication message could fail e.g. if there are no
buffers available, however this doesn't mean that we should stop doing
periodic publishing indefinitely. When an error occurs, make sure to
call the publish_sent() function so that the periodic publishing timer
gets resubmitted if necessary.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Previously the code only checked if any of the models within an
element had the destination address in their subscription list. After
that check the specific model that the message was addressed to was
not verified to have that address in it's subscription list. This
patch fixes the problem.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
There were several things broken with the initialization order during
node reset:
- The model->flags should not be touched since the writing to
persistent storage happends through delayed work, and the flags
need to be kept until that.
- The unprovision() function should only be called at the very end of
the reset procedure, since it calls model-specific init functions
which may clear things which the earlier reset routines depend on.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Keeping the model struct same sized, change the element pointer to two
indexes, and add a flags member that will be used to track pending
storage actions.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
There are valid use cases where the model layer must know the true
destination address. So far only the fact that it was one of the
addresses that the model subscribes to (its element's unicast
included) has been knowable.
Solve the issue by moving the destination address from the internal
net_rx context to the public bt_mesh_msg_ctx struct.
Fixes#7453
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Convert the mesh code to use the new net_buf_siple APIs. This has the
benefit of saving 4 bytes off the stack due to the not needed pointer.
Also update the publication context helpers to map to the new
net_buf_simple API in an intuitive way.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The bt_mesh_model_publish() is supposed to return a "not supported"
error if the publish context doesn't exist. Fixing the premature
dereferencing also fixes coverity warnings.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The separate checking for "now < reftime" is unnecessary, since the
integer over/under-flow for unsigned 32-bit values resulting from
subtraction will give the right delta even if 'now' is less than the
reference.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
After the Publish Retransmit state was introduced the Publish Period
measurement would begin once the previous Publish message has finished
transmitting. This will however cause inaccurate periods, which is
particularly an issue with the PTS that expects accuracy of less than
0.5 seconds (apparently).
Since the publication timer is also used for the retransmissions we
can't simultaneously use if for the period as well. Therefore, we
introduce a new variable called period_start which makes a note of
when the period was supposed to start, and then once all
retransmissoins are done initializes the timer with the send duration
taken into account.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The only generally available model supporting publication that's
convenient to be used for testing is the Health Server Model.
Unfortunately since this model supports period publication, the
non-periodic side got less attention and had some bugs.
The first thing that needs to be done is to verify that the period
returned by bt_mesh_model_pub_period_get() is positive. If it's zero
then no periodic publication should take place.
Another thing that this patch cleans up is the naming of the callback
used for periodic publishing. There's no need do require the callback
to call bt_mesh_model_publish() since this must happen no matter what,
so instead rename the callback from 'func' to 'update' and have the
access layer call bt_mesh_model_publish() if the callback was
successful.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Model publication was broken in a couple of ways:
- The Publish Retransmit State was not taken into account at all
- Health Server used a single publish state for all elements
To implement Publish Retransmit properly, one has to use a callback to
track when the message has been sent. The problem with the transport
layer sending APIs was that giving a callback would cause the
transport layer to assume that segmentation (with acks) is desired,
which is not the case for Model Publication (unless the message itself
is too large, of course). Because of this, the message sending context
receives a new send_rel ("Send Reliable") boolean member that an app
can use to force reliable sending.
Another challenge with the Publish Retransmit state is that a buffer
is needed for storing the AppKey-encrypted SDU once it has been sent
out for the first time.To solve this, a new new net_buf_simple member
is added to the model publication context. The separate 'msg' input
parameter of the bt_mesh_model_publish() API is removed, since the
application is now expected to pre-fill pub->msg instead.
To help with the publishing API change, the Health Server model gets a
new helper macro for initializing the publishing context with a
right-sized publishing message.
The API for creating Health Server instances is also redesigned since
it was so far using a single model publishing state, which would
result in erratic behavior in case of multiple elements with the
Health Server Model. Now, the application needs to provide a unique
publishing context for each Health Server instance.
The changes are heavily intertwined, so it's not easily possible to
split them into multiple patches, hence the large(ish) patch.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
There's no need for callback exposed in the public API to be something
different than what's used internally. In fact this would just
complicate things. This patch exposes the internal callback under a
bt_mesh_adv_cb name and uses it throughout the mesh stack.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The common interpretation (among other implementations) seems to be
that Model Publication for a given AppKey Index implies a binding for
that AppKey. This isn't currently explicitly stated in the spec, but
in order to improve interoperability go with this interpretation as
well.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
This is in anticipation of soon adding health client support, which
could then cause confusion due to the ambiguous API names.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Now that there's support for configuration client as well, rename cfg
to cfg_srv to avoid any confusion.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The only messages that should be encrypted using the friendship
credentials are those coming through the Friend Queue on the Friend
node, most request-response pairs between LPN & Friend (exceptions are
Friend Request - Friend Offer, and Friend Clear - Friend Clear
Confirm), as well as Model Publication messages when the Friendship
Credentials Flag has been enabled in the model publication.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
This makes it possible (in a subsequent patch) to fine-tune some
special cases, like the LPN poll messages.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Encrypting and sending a message takes a considerable amount of time
which makes the publication period longer than expected.
With this patch it is possible to pass MESH/SR/HM/CFS/BV-02-C test.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The API name space for Bluetooth is bt_* and BT_* so it makes sense to
align the Kconfig name space with this. The additional benefit is that
this also makes the names shorter. It is also in line with what Linux
uses for Bluetooth Kconfig entries.
Some Bluetooth-related Networking Kconfig defines are renamed as well
in order to be consistent, such as NET_L2_BLUETOOTH.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Add an initial implementation for the Bluetooth Mesh Profile
Specification. The main code resides in subsys/bluetooth/host/mesh and
the public API can be found in include/bluetooth/mesh.h. There are a
couple of samples provided as well under samples/bluetooth and
tests/bluetooth.
The implementation covers all layers of the Bluetooth Mesh stack and
most optional features as well. The following is a list of some of
these features and the c-files where the implementation can be found:
- GATT & Advertising bearers (proxy.c & adv.c)
- Network Layer (net.c)
- Lower and Upper Transport Layers (transport.c)
- Access Layer (access.c)
- Foundation Models, Server role (health.c & cfg.c)
- Both PB-ADV and PB-GATT based provisioning (prov.c)
- Low Power Node support (lpn.c)
- Relay support (net.c)
- GATT Proxy (proxy.c)
Notable features that are *not* part of the implementation:
- Friend support (initial bits are in place in friend.c)
- Provisioner support (low-value for typical Zephyr devices)
- GATT Client (low-value for typical Zephyr devices)
Jira: ZEP-2360
Change-Id: Ic773113dbfd84878ff8cee7fe2bb948f0ace19ed
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
