| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
| |
* Fix macro code to use boolean value in while instruction (false)
Signed-off-by: Michael Methner <mmethner@de.adit-jv.com>
|
|
|
|
|
|
| |
make sure that a string with exactly UINT16_MAX bytes does not
overflow the dlt buffer
Signed-off-by: Alexander Mohr <alexander.m.mohr@mercedes-benz.com>
|
|
|
|
|
|
|
| |
in case the house keeper thread does not start
we should not wait indefinitely but abort
the wait after a 10s
Signed-off-by: Alexander Mohr <alexander.m.mohr@mercedes-benz.com>
|
|
|
|
|
|
|
|
|
| |
wait until the house keeper is up and running
this makes sure dlt_init / dlt_free in close succession
closes this thread properly.
Otherwise we might keep the dlt socket open
Check if the thread is running every 10ms
Signed-off-by: Alexander Mohr <alexander.m.mohr@mercedes-benz.com>
|
|
|
|
|
|
|
|
|
| |
If dlt_free is called while dlt_init is running, the dlt-user library
crashes because the init and free where not thread safe.
This commit introduces a new field dlt_user_initialising
which will prevent entering dlt_free while dlt_init is running.
Correctness is ensured by a unit test
Signed-off-by: Alexander Mohr <alexander.m.mohr@mercedes-benz.com>
|
|
|
|
|
|
|
|
|
|
|
| |
dlt_daemon_client.c Adding NULL check for tok
dlt_daemon_offline_logstorage.c : Adding NULL check for application
dlt_user.c : Fix for Memory Leak
dlt-daemon.c : Fix for Resource Leak
dlt_config_file_parser.c : Add termination character at the end of string
dlt_offline_trace.c : Fix for Resource Leak
Signed-off-by: Mvaradaraj2 manoj.varadaraj2@harman.com
Co-authored-by: Le Tin <tin.le@vn.bosch.com>
|
|
|
|
|
| |
Closes #352
Signed-off-by: Saya Sugiura <ssugiura@jp.adit-jv.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* some previous dlt_vlog was using %lu specifier to display an uint64_t
variable. However, this isn't portable as depending on the system you're
targeting (32bits, 64bits, Linux, android, etc.), the uint64_t can be an
unsigned long long. This prevents compilation on system where -Werror
and -Wformat is activated coupled with a less tolerant compiler
(e.g clang). Instead, PRIxN macros (e.g PRIu64 for uint64_t) are now
used as specifier to ensure compatibility for each platforms
* for bytes_read which is of type ssize_t, zd specifier is used as
indicated in C99 standard
* for st_size (from struct stat) which is of type off_t, POSIX says that
it should be a signed integer. To be safe, we now use %jd specifier to
ensure that it will fit correctly
|
|
|
|
|
|
|
| |
* function atomic_compare_exchange_strong expect a pointer to an _Atomic type. dlt_user_initialised is now an atomic_bool
* function dlt_set_filesize_max returned a DLT_LOG_ERROR which isn't a DltReturnValue. This commit replaces DLT_LOG_ERROR by DLT_RETURN_ERROR which is the expected return type
Signed-off-by: Sebastien RAILLET <sebastien.raillet@marelli.com>
|
|
|
|
| |
Signed-off-by: Saya Sugiura <ssugiura@jp.adit-jv.com>
|
|
|
|
|
|
|
| |
Extern logging_mode and logging_handle variables from commom.c
Reusing dlt_log_init() in cases dlt library internal log file is not opened and logging_mode is DLT_LOG_TO_FILE
Signed-off-by: Le-Tin <Tin.Le@vn.bosch.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit introduces the use of atomic_compare_exchange at beginning
of both dlt_init functions. This function checks in an atomic way, if
'dlt_user_initialised' is 'false' (first run through dlt_init) and directly
sets it 'true'. If a second thread also entered dlt-init at this point of time,
its call of atomic_compare_exchange will return false and dlt_init will be
immediately left due to that. This is only interesting for startup, because
before each call of dlt_init, there is a check on 'dlt_user_initialised' that
will fail for all later executions.
Signed-off-by: Darian Biastoch <dbiastoch@de.adit-jv.com>
|
|
|
|
| |
Signed-off-by: Saya Sugiura <ssugiura@jp.adit-jv.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit adds new interfaces:
- dlt_user_log_write_start_w_given_buffer(): accepts a log buffer as an input.
- dlt_user_log_write_finish_w_given_buffer(): finalizes logging and send the message to corresponding output
This is to reduce DLT API calls in DLT application as much as possible
and avoid allocating dynamic memory within DLT library. The log buffer
has to be prepared in DLT application in order to use these interfaces.
Signed-off-by: Saya Sugiura <ssugiura@jp.adit-jv.com>
|
|
|
|
| |
Signed-Off By: Saya Sugiura <ssugiura@jp.adit-jv.com>
|
|
|
| |
Co-authored-by: Saya Sugiura <39760799+ssugiura@users.noreply.github.com>
|
|
|
|
|
|
|
| |
Use POSIX thread API to rename the housekeeper and segmented threads.
The currently used prctl() function is Linux specific and in general not
available on other platforms (e.g. QNX).
Signed-off-by: Sebastian Lipponer <mail@sebastianlipponer.de>
|
|
|
| |
According to AUTOSAR standard the TYLE for the BOOL type shall be set to 1. So I propose to initialize type_info variable for the BOOL type the same way as for uint8 type.
|
|
|
|
|
|
| |
Fixing casting incorrect type
Signed-off-by: Dinh Cong Toan <toan.dinhcong@vn.bosch.com>
|
|
|
|
|
|
|
|
| |
When processing dlt_user_log_send_log() function, message logs will be stored in
internal buffer in error case. Then, the error value is needed to handle correctly
before exiting.
Signed-off-by: Dinh Cong Toan <toan.dinhcong@vn.bosch.com>
|
|
|
| |
Signed-off-by: Martin Willers <M.Willers@gmx.net>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* Make Non-Verbose mode non-exclusive
Switching to global Non-Verbose mode now does not force Verbose messages to
also be sent as Non-Verbose ones anymore. That would not make any sense,
because Verbose messages don't have a MessageId and thus are all getting
the same MessageId of 65535.
Instead, setting global "Non-Verbose" mode will allow both Verbose and
Non-Verbose messages to be sent in a single session.
The "Verbose-APIs" (e.g. dlt_user_log_write_start()) will then only write
Verbose messages, whereas the "Non-Verbose APIs" (e.g. dlt_user_log_write_start_id())
will then only write Non-Verbose messages.
Signed-off-by: Martin Willers <M.Willers@gmx.net>
|
|
|
| |
Co-authored-by: FIXED-TERM Kaczorowski Radoslaw (BSOT/PJ-ES21) <Fixed-Term.Radoslaw.Kaczorowski@bosch-softtec.com>
|
|
|
|
|
|
| |
dlt_user_log_write_*_attr() enables to writing these types, but also support
adding "attributes" for them, i.e. a "name" and a "unit".
Signed-off-by: Martin Willers <M.Willers@gmx.net>
|
|
|
|
|
|
|
|
|
|
|
| |
- Converting datatype to the correct one.
- As macro do not perform type-checking before, so these functions used
it with different datatype could create a lot of conversion warnings.
For this reason, these warnings could be consider as normal
information and will not be fixed.
Signed-off-by: Dinh Cong Toan <toan.dinhcong@vn.bosch.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Change the data type of variable 'size' in DltContextData struct
(dlt_user.h.in). Variable 'size' was used repeated with function
'sizeof()' many time, so change type to 'size_t' can avoid a lot of
warnings.
- Adding temporary variable to calculate the datasize header (define in
dlt_common.h) in 'dlt_common.c' file.
Signed-off-by: Dinh Cong Toan(RBVH/ECM12) <Toan.DinhCong@vn.bosch.com>
|
|
|
|
|
|
|
|
|
| |
- Correct the data type of 'headersize' and 'datasize' variables in
header 'dlt_common.h', then config those functions using this library.
- Explicit the data type.
Signed-off-by: Dinh Cong Toan(RBVH/ECM12) <Toan.DinhCong@vn.bosch.com>
|
|
|
|
|
|
|
| |
Since bionic does not support pthread_cancel(),
try to terminate housekeeper thread via SIGUSR1.
Signed-off-by: Bui Nguyen Quoc Thanh <thanh.buinguyenquoc@vn.bosch.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Use timeout of polling as time for retry.
In case of timeout or failure of polling,
continue to close the socket.
It could take some times to get the socket is shutdown
So it means there could be some data available to read.
Try to consume the data and poll the socket again.
If read fails, time to close the socket then.
Signed-off-by: Bui Nguyen Quoc Thanh <thanh.buinguyenquoc@vn.bosch.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Current implementation had two issues:
- Remaining data in startup_buffer wasn't able to be sent to daemon
even dlt_user.timeout_at_exit_handler was set to more than 0
- Socket could be closed before all the data was sent to daemon
To solve this:
- dlt_unregister_app_flush_buffered_logs() won't clean appIDs if
any data is remaining in startup_buffer. atexit_handler will be
in charge of cleaning them
- Socket won't be closed if data wasn't sent due to errno EAGAIN or
EWOULDBLOCK. It retries until execution count reaches to maximum
DLT_USER_MAX_RETRY_COUNT (1000 as default)
Signed-off-by: Saya Sugiura <ssugiura@jp.adit-jv.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In order to disable injection message at dlt user,
set DLT_DISABLE_INJECTION_MSG_AT_USER to any value.
When injection message is disabled, libdlt won't poll
any data from receiver. It means libdlt will ignore all
data/messages from dlt-daemon.
By default, variable is unset so that dlt users are able to
handle data/messages from dlt-daemon.
Signed-off-by: Bui Nguyen Quoc Thanh <thanh.buinguyenquoc@vn.bosch.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The poll timeout was only set for fifo.
TCP connections require this timeout as well.
dlt_user_log_check_user_message also validated the file descriptor
to be greater 0.
Because 0 is a valid file descriptor this check has been changed
to greater or equal 0.
poll receives a timeout in milliseconds.
The given paramter was in nanoseconds.
An additional define takes adds the delay in miliseconds.
Signed-off-by: Alexander Mohr <alexander.m.mohr@daimler.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
For more information about VSOCK, see "man vsock"
( https://man7.org/linux/man-pages/man7/vsock.7.html ).
Makes it possible for processes in a virtual machine to log directly in
DLT running on host without setting up a network connection between
guest and host. It is also probably more efficient. Have not done any
performance measurements (not main reason for patch), but no forwarding
is required as when running DLT in a multi-node setup.
When building dlt-daemon for host, WITH_DLT_DAEMON_VSOCK_IPC should be
enabled for daemon to listen on incoming VSOCK requests. Local
communication method between applications and daemon is still determined
with DLT_IPC. When building for guest, WITH_DLT_LIB_VSOCK_IPC should be
enabled and DLT_IPC will be ignored, which will make libdlt open a VSOCK
socket to the deamon for logging messages.
VSOCK can be tested without a virtual machine. Since VMADDR_CID_HOST is
used by libdlt when connecting, see vsock man page, clients can be run
on host to test VSOCK communication.
Some modifications has been done to be able to handle logging through
FIFO pipe and socket in the same build of dlt-daemon:
- dlt_receiver_init/free_unix_socket() is renamed to
dlt_receiver_init/free_global_buffer() and used for FIFO as well.
Also fixes memory leak since dlt_receiver_free_unix_socket() was used
regardless of whether DLT_USE_UNIX_SOCKET was defined or not.
- Pass type to dlt_receiver_init() instead of dlt_receiver_receive().
And remove preprocessor conditionals for handling
DLT_CONNECTION_APP_MSG in dlt_daemon_process_user_messages(). Also
fixes wrong enum type being passed to dlt_receiver_receive() in
dlt_client.c (DltClient::mode was used as a DltReceiverType enum but
it is a DltClientMode enum).
- Add a flag to DltDaemonApplication to indicate whether file descriptor
is "owned" by the DltDaemonApplication or not. When
dlt_daemon_application_add() is called due to message received on a
socket, fd is passed as an argument (app does not own fd). For FIFO,
a per application FIFO is opened (app owns the fd). Also fixes so that
user handle is reset for both application and all its contexts when
resetting any. Prevents fd from being used by accident after it has
been closed.
dlt_mkdir_recursive() is moved to src/daemon since it is only used in
the daemon. Minimizes use of DLT_USE_UNIX_SOCKET_IPC.
Other bugfixes:
- Call DLT_SEM_FREE() if setting socket to O_NONBLOCK fails in
src/lib/dlt_user.c:dlt_initialize_socket_connection().
- Close socket if dlt_receiver_init() fails in
src/lib/dlt_user.c:dlt_initialize_socket_connection().
Signed-off-by: Martin Ejdestig <martin.ejdestig@volvocars.com>
|
|
|
|
| |
Signed-off-by: KHANH LUONG HONG DUY <khanh.luonghongduy@vn.bosch.com>
|
|
|
|
| |
Signed-off-by: KHANH LUONG HONG DUY <khanh.luonghongduy@vn.bosch.com>
|
|
|
|
|
|
|
| |
Relocate the setting of dltFifoBaseDir variable from dlt_init() to
dlt_check_envvar()
Signed-off-by: LUONG HONG DUY KHANH(RBVH/ENG42) <KHANH.LUONGHONGDUY@vn.bosch.com>
|
|
|
|
|
|
|
|
|
|
| |
There is some OS (e.g. Android) which doesn't support message queue.
Since network trace uses it as IPC, we need to disable it if following
calls are not available:
mq_open, mq_close, mq_unlink, mq_send, mq_receive
Signed-off-by: Saya Sugiura <ssugiura@jp.adit-jv.com>
Signed-off-by: KHANH LUONG HONG DUY <khanh.luonghongduy@vn.bosch.com>
|
|
|
|
| |
Signed-off-by: LUONG HONG DUY KHANH(RBVH/ENG42) <KHANH.LUONGHONGDUY@vn.bosch.com>
|
|
|
|
|
|
|
|
|
| |
Many stringop-truncation and stringop-overflow warnings are still
there (so many).
https://developers.redhat.com/blog/2018/05/24/detecting-string-truncation-with-gcc-8/
Signed-off-by: Felix Herrmann <fherrmann@de.adit-jv.com>
Signed-off-by: KHANH LUONG HONG DUY <khanh.luonghongduy@vn.bosch.com>
|
|
|
|
|
| |
Signed-off-by: Felix Herrmann <fherrmann@de.adit-jv.com>
Signed-off-by: KHANH LUONG HONG DUY <khanh.luonghongduy@vn.bosch.com>
|
|
|
|
|
|
|
|
|
|
|
| |
Most pthreads functions return 0 on success, and an error number on
failure. Note that the pthreads functions do not set errno. Refer to
man page of pthreads for detail.
So we have to use the return value as the error number instead of errno
variable.
Signed-off-by: Vo Trung Chi <chi.votrung@vn.bosch.com>
|
|
|
|
|
|
|
|
|
| |
Problem occurs always if application exits and
dlt_user.dlt_segmented_queue_read_handle is not initialized.
Warning "NWTSegmented: Error while reading queue" is printed
on terminal on embedded devices.
Signed-off-by: Rudolf Svanda <rudolf.svanda@globallogic.com>
|
|
|
|
|
|
|
| |
There is no need to have both an array and a pointer to it;
just use the array directly.
Signed-off-by: Martin Willers <M.Willers@gmx.net>
|
|
|
|
| |
Signed-off-by: Martin Willers <M.Willers@gmx.net>
|
|
|
|
|
|
|
|
|
|
|
| |
This use case is common in C++ code nowadays, due to the usage of e.g.
std::string_view, which does not provide a function for retrieving a
null-terminated C-string, but which provides quick access to the string
length via std::string_view::size().
Such use cases can now avoid an entirely unnecessary strlen() call.
Signed-off-by: Martin Willers <M.Willers@gmx.net>
|
|
|
|
|
|
|
|
|
| |
poll.h being included directly.
This fixes a build failure on musl systems with strict c hardening flags
Signed-off-by: Gianfranco Costamagna <costamagnagianfranco@yahoo.it>
Signed-off-by: Gianfranco Costamagna <locutusofborg@debian.org>
|
|
|
|
|
|
|
|
| |
for fixing the coverity warnings type Uninitialized scalar variable (UNINIT).
Using uninitialized value log.use_timestamp when
calling dlt_user_log_send_log.
Signed-off-by: Phong Tran <tranmanphong@gmail.com>
|
|
|
| |
Signed-off-by: Christoph Lipka <clipka@jp.adit-jv.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Improvement: housekeeper thread with poll
- Change name receiver thread to housekeeper thread
- Use poll() in housekeeper thread to get new received message
from DLT Daemon.
- Remove resend buffer inside of dlt_user_log_reattach_to_daemon()
- Only try to resend buffer if fd is valid.
This commit removes the event driven dlt_user_log_reattach_to_daemon()
that is called e.g. by DLT_LOG. Now, only the DLT housekeeper thread
handles the retrial to attach to daemon.
Signed-off-by: Vo Trung Chi <Chi.VoTrung@vn.bosch.com>
Signed-off-by: Saya Sugiura <ssugiura@jp.adit-jv.com>
Signed-off-by: Simon Herkenhoff <sherkenhoff@jp.adit-jv.com>
Signed-off-by: Bui Nguyen Quoc Thanh <Thanh.BuiNguyenQuoc@vn.bosch.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The fix from commit 376bf6c is considered as a workaround.
From now, the resend buffer size is calculated based on
(dlt_user.log_buf_len + header_size)
- Remove the macro DLT_USER_EXTRA_BUFF_SIZE
- Use env_disable_extended_header_for_nonverbose instead of
env_nonverbose_extended_header
Related commit: 376bf6c Size of Resend buffer less than or equal to DLT_USER_BUF_MAX_SIZE res… (#116)
Signed-off-by: Bui Nguyen Quoc Thanh <Thanh.BuiNguyenQuoc@vn.bosch.com>
|