Removing entries from an open addressed hash table creates holes in the
collision chains, preventing previous colliding entries to be found. By
inserting tombstones, rather than clearing deleted entries makes it
possible to distinguish the end of a chain from a hole.
Reviewed-by: Arun Kumar Neelakantam <aneela@codeaurora.org>
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
The dprintf macro shadows a real libc symbol which is called
differently, so we should change it to something else (in this
case LOGD) and change it to use the libqrtr logging facilities.
This also means we should add a minimum priority to the logging
library so we don't log DEBUG-level messages by default.
Additionally there were some warn/warnx calls left over that
have been changed to PLOGW/LOGW respectively.
The DIAG service has its own encoding for the instance number which
is unrelated to the version. This means that the information we
show to the user in the version and instance columns of the table is
parsed incorrectly and often useless. Instead, treat the instance
number as special if the service is a DIAG service and print out a
more meaningful interpretation.
Signed-off-by: Eric Caruso <ejcaruso@chromium.org>
The "instance" id of a service is a composition of the instance id and
the version. In some plaeces in the library I made this two half words
instead of the actually expected 8 bits version + 24 bit instance. Fix
this in the library.
This was not spotted previously because all prior work has dealt with
instance id 0, but with the QRTR support in Diag this is now an issue.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Similar to qmi_decode_message(), qmi_decode_header() doesn't modify the
`struct qrtr_packet' argument. This patch makes qmi_decode_header() to
take a `const struct qrtr_packet' argument like qmi_decode_message()
does.
Contributed by Jacob Rutherford <jruthe@chromium.org>
waiter_wait_timeout() doesn't initialize the 'callback' and 'updated'
fields of the waiter_ticket struct. The 'updated' field may contain some
garbage value, which is later read in waiter_ticket_check().
Signed-off-by: Ben Chan <benchan@chromium.org>
This patch adds an extern "C" guard in libqrtr.h, so that it can be
included in both C and C++ projects.
Signed-off-by: Ben Chan <benchan@chromium.org>
This function in qrtr-ns was largely copied from the main
function of qrtr-cfg. We should just factor it out so there is
one implementation instead.
Signed-off-by: Eric Caruso <ejcaruso@chromium.org>
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
The __progname symbol is a GNU extension and thus not portable,
but basename (from libgen.h) and argv[0] can be used portably
instead.
(qrtr-lookup already used argv[0] but did not apply basename, so
do that for consisntency.)
Signed-off-by: Eric Caruso <ejcaruso@chromium.org>
This include provides err, errx, etc. but it looks like all
error logging in qmi.c is done with fprintf, so there's no
reason to include it.
Change-Id: I4f4a71323a55fca981ef8df994fe624a590e3802
When running in Android we want to keep qrtr-ns in the foreground, so
that init's service handler can keep track of it. Add the -f flag to
skip the fork().
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Instead of calling qrtr-cfg separately to configure the local address
just pass this as an argument to qrtr-ns, allowing us to run a single
command to instantiate qrtr.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
By opening a socket before attempting to configure the address we
trigger kernel module loading, if qrtr is compiled as such.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Definition of AF_QIPCRTR was introduced in glibc in the following
commit and is only available since glibc 2.25 or later.
https://sourceware.org/git/?p=glibc.git;a=commit;h=acaff9b658720e4c887f4e44e6f28962d6f372d3
This patch modifies libqrtr.h to define AF_QIPCRTR if it isn't defined
in libc header, which allows compilation of libqrtr on a system with an
older version of glibc. The compiled libqrtr can still be used on a
system with kernel 4.7 or later with the QIPCRTR support.
Signed-off-by: Ben Chan <benchan@chromium.org>
The QMI functions can be used to deal with QMI-encoded data in
qrtr_packet objects. The structure-description is the same element-info
arrays as is used in the upstream Linux kernel.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
The qrtr_decode() takes a buffer of data and returns a decoded qrtr
packet object, this will either contain control data or a reference to
the message data.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
The uapi qrtr.h is extended with additional types and constants. Drop
the local qrtr.h to use the one from linux-headers and extend it
conditionally while we're waiting for the updated file.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
When sending a new_server message to the qrtr-ns the node and port in
the packet is ignored (to not allow clients spoofing the information),
so there's no reason to have this artificial check in the
qrtr_new_server() implementation either.
The socket will be bound to a port automatically when the new_server
message is sent out.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Restrict the origin of certain messages to limit the ability of clients
to inject invalid data in the service registry.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
When a server is disappearing, either from a DEL_SERVER, BYE or
DEL_CLIENT message the same messages needs to be sent; to local
observers and through broadcasts to remote nodes.
So move the logic from the DEL_SERVER command handler to server_del, in
order to invoke this regardless of reason for the server's
disappearance.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
There's no reason to require that the socket to be bound to a specific
port in order to register a lookup, the transmission of the lookup
request will autobind the socket to a port for us.
Also drop the node and port from the payload, as qrtr-ns will register
the lookup on the source sq, rather than the passed port.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
As we propagate the BYE to local listeners we have to also pass the
information on which node is down, so use the typical "node" field for
this.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
A number of services are mentioned in the Qualcomm downstream kernels,
add these to the list of known names to improve the output.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Instead of reporting "service" as id:instance split the "instance" in version
and instance and present the three entities separately.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Fix up the handling of endian translation and add support for decoding
new and del server messages.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Instead of having a custom lookup result, just pass new and delete
server messages - allowing clients to receive notifications about dying
services as well, without having to extend the lookup result message.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
The two added functions can be used by an application to check if a
sockaddr is from the control port and for parsing the control messages
that is passed to registered clients.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Request the return value of the operation from the kernel and check if
it's an error, to inform the user about e.g. not having sufficient
permission.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Now that we only have one socket for our context we can simplify the
name of the struct member.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Now that we have moved all operations to the control socket we can
remove the ns handler. The query function in the library is racy, as
there might be non-response messages in our incoming buffer when we
start reading, so this is dropped too - rather than just being
translated to the new interface.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Convert qrtr-lookup to use the control port for queries, allowing us to
drop the ns service altogether.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Introduce a subscription based lookup model, in order for a client to be
able to listen to a future spawning service. The client can register any
number of queries with the ns and the ns will post lookup-results as
services are registred.
As a special case the list of lookup-results that is sent out as a
result of the registration of a new lookup will be finished by an empty
lookup response, allowing tools like qrtr-lookup to terminate as the
entire list of currently registed services is received.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Replace the implementation of the two library functions with calls to
the new APIs, so that we can remove the server-side implementation of
the previous messages.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
The BYE control message signals that a node is gone, drop all services
from our register and propagate this message to all local servies - to
allow them to clean up any resources associated with potential remote
clients.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Following the downstream kernel we should propagate a del_client to all
local services as the dying socket might have acted - or once acted - as
a client as well.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Upon receiving a DEL_CLIENT some local or remote client has been
disconnected. If this is a local server we remove the server from the
lists and announce the removal to the remotes. Otherwise we need to
inform local servers that one of their clients might be gone, so we
propagate a DEL_SERVER message to all locally registered services.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
