This commit fixes the following warning seen with CC=clang:
utils.c:376:6: warning: variable 'len' set but not used [-Wunused-but-set-variable]
int len = 0, offset = 0, ret, rem;
... and finally allows to build libosmocore with --enable-werror.
Change-Id: I0040ef20ba3fc53ee7ccefc4885170f333f80566
This code predates 2cbe25f4, adding osmo_strbuf API and so using its
own append-to-strbuf implementation. Let's use the new generic API.
Change-Id: Ifdfd18eeef6a0932995063259f9f179b22e781de
The member data[0] in struct osmo_ecu_state is used as an anchor to
attach private structs for a concrete ECU implementation. This works by
allocating more memory then struct osmo_ecu_state actually needs and
then using the excess memory to store the private struct of the concrete
ECU implementation.
However, this poses a problem since data[0] is at the end of the struct
it may land in an unaligned position. This also means that the struct we
store there is also unaligned.
We should fix this enclosing the public struct osmo_ecu_state into our
private struct fr_ecu_state. Then we can use container_of to cast from
osmo_ecu_state to fr_ecu_state and correct alignment is ensured as well.
Related: OS#6286
Change-Id: I28672856e8e8f47e04ffe09ee3e07b577108cdc7
bitvec.c:543:14: warning: variable 'pos' set but not used [-Wunused-but-set-variable]
unsigned i, pos = 0;
Change-Id: I17df6f9263bee06676309c00837f12220803c814
The following can be seen when building with CC=clang:
utils/utils_test.c:1239:2: runtime error: applying non-zero offset 99 to null pointer
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior utils/utils_test.c:1239:2 in
utils/utils_test.c:1241:3: runtime error: applying non-zero offset 99 to null pointer
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior utils/utils_test.c:1241:3 in
utils/utils_test.c:1242:2: runtime error: applying non-zero offset 99 to null pointer
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior utils/utils_test.c:1242:2 in
44. testsuite.at:274: 44. utils (testsuite.at:274): FAILED (testsuite.at:278)
This makes utils_test fail due to unexpected UBSan's output.
Even though passing NULL to the strbuf API is relatively safe, it makes
no sense and the API user should ensure that this never happens. And
so we should not be testing this case.
Change-Id: Icd2323e93ec64afc1822d48e5e1d090083edf539
This can be seen when building with CC=clang:
utils.c:150:22: runtime error: applying non-zero offset 100 to null pointer
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior utils.c:150:22 in
utils.c:150:33: runtime error: addition of unsigned offset to 0x000000000064 overflowed to 0x000000000063
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior utils.c:150:33 in
The *dst pointer may be NULL (e.g. bcd2str_test() is passing it).
This makes tests/utils/utils_test fail. Let's fix this.
Change-Id: I542aef1ac220891b6bbdb0c60c39232f0df0a43c
We're seeing tons of -Wsign-compare warnings since I merged 0f59cebf:
include/osmocom/core/utils.h: In function 'size_t _osmo_strbuf_char_count(const osmo_strbuf*)':
include/osmocom/core/utils.h:24:29: error: comparison of integer expressions of different
signedness: 'long int' and 'long unsigned int'
[-Werror=sign-compare]
24 | #define OSMO_MIN(a, b) ((a) >= (b) ? (b) : (a))
| ~~~~^~~~~~
include/osmocom/core/utils.h:309:16: note: in expansion of macro 'OSMO_MIN'
309 | return OSMO_MIN(sb->pos - sb->buf, sb->len - 1);
| ^~~~~~~~
Interestingly enough, this -Wsign-compare problem has always been the
case, even before commit 0f59cebf. And somehow this did not show up
when building libosmocore.git, but only when building C++ projects
(osmo-pcu and osmo-trx).
Perhaps it has something to do with how g++ compiles extern "C" code.
Change-Id: I8e396459409e4260b8715f9e890e8972d4609a31
Fixes: 0f59cebf "utils: improve readability of OSMO_STRBUF_CHAR_COUNT"
Similarly to OSMO_STRBUF_REMAIN, let's improve the code readability
by adding a static inline function. We should generally prefer using
static inline functions over macros, unless there is something that
only the proprocessor can do.
Change-Id: I71f24b87c13fd83952029171a6993f8da5e32e5b
Currently, OSMO_STRBUF_REMAIN would return a huge number if the given
strbuf has its .pos pointer set to NULL. This macro is safe against
the .buf pointer being NULL, but not against .pos being NULL.
Fix this by adding a static inline function (for the sake of code
readability) and handle .pos being NULL properly by returning length
of the buffer. Add a unit test.
Change-Id: I294a74a99c40995cf7fb5520d61f697d967be5a4
Extend the existing testing coverage to check per-state enter/leave
callbacks. An interesting behavior can be seen from the test output:
when allocating an FSM instance, the .onenter callback is not being
called for the initial FSM state (ST_NULL). Likewise, the .onleave
callback is not being called when free()ing an FSM instance.
Change-Id: I22edcf91375a09854f0dab1e2e02e034629310f7
The code so far only supported 240bit RLP frames; Add support for
576bit in this patch. We still only support versions 0+1 and not
version 2.
Change-Id: Idfdcabb19fe8733fb9c5ee76a39b0bf4cdf60c2c
- Use defines instead of repeating string literals
- Use enum for indexing same data everywhere consistently
Related: OS#5915
Change-Id: I11c926dd4125c6657ae3cd21d8038f161d9fd9c3
We should not be doing the actual logic inside the assert statements.
Change-Id: I3bbf4e602c25c5aaced609e9834d6b053688194d
Fixes: CID#274664, CID#274679
These values end up being used by API users of
osmo_sock_multiaddr_get_name_buf() and
osmo_multiaddr_ip_and_port_snprintf().
Change-Id: I18a0e1a652a3e8ef3e97154355eb1d07a14ef0bd
As outlined in the test case, we pull a total of 50 bits from the
transmitter in two rounds, pulling 25 bits at a time. In the default
8-N-1 configuration, 50 bits should ideally comprise 5 characters.
However, as observed, only a total of 4 characters are retrieved
from the application, leaving the remaining 10 bits (5 + 5) unused.
Change-Id: Ic2539681a4adf6c1822e0bc256e4c829813d0e21
This problem can only happen if the user is flush()ing the Rx buffer
manually by calling osmo_soft_uart_flush_rx(). Let's demonstrate it
in the unit test, so that we don't forget about it (add FIXME).
Change-Id: Iad932a505d6fd98360f90510651501f8708ff5d2
Coverity tells us that with the current logic it's possible (in theory)
that we may dereference NULL pointer in osmo_soft_uart_flush_rx(). This
is highly unlikely, because the Rx buffer gets allocated once when the
Rx is enabled and remains even after the Rx gets disabled. The Rx flags
cannot be anything than 0x00 before the Rx gets enabled.
Even though this NULL pointer dereference is unlikely, the Rx flushing
logic is still not entirely correct. As can be seen from the unit test
output, the Rx callback of the application may be called with an empty
msgb if the following conditions are both met:
a) the osmo_soft_uart_flush_rx() is invoked manually, and
b) a parity and/or a framing error has occurred previously.
We should not be checking suart->rx.flags in osmo_soft_uart_flush_rx(),
since this is already done in suart_rx_ch(), which is calling it.
Removing this check also eliminates a theoretical possibility of the
NULL pointer dereference, so we're killing two birds with one stone.
- Do not check suart->rx.flags in osmo_soft_uart_flush_rx().
- Add a unit test for various flush()ing scenarios.
Change-Id: I5179f5fd2361e4e96ac9bf48e80b99e53a7e4712
Fixes: CID#336545
This is a convenience helper to reetrieve the whole set of remote
addresses and call getsockopt() on them, making it easy for users to
analyse the full set of remote addresses of a socket simply providing an
fd.
Related: SYS#6636
Change-Id: I3e1c84526b006baff435bbbca49dc6cf7d201cf5
In the _output_buf() we explicitly initialize only the 'buf' and 'len'
fields of the struct osmo_strbuf, leaving the 'pos' field implicitly
initialized to NULL. Later, in this function, 'sb.pos' is passed to
ctime_r() and strlen(), leading to a NULL pointer dereference (segfault)
in certain scenarios.
This situation can occur when color logging is disabled or when
a specific logging subsystem has no associated color. Any application
using libosmocore's logging API would crash with the following config:
log stderr
logging filter all 1
logging timestamp 1
logging color 0
Fix this by initializing the 'pos' field explicitly.
Change-Id: I7ec9badf525e03e54e10b725d820c636eaa3fd1c
Fixes: d71331bc "logging: fix nul octets in log output / use osmo_strbuf"
Fixes: CID#336550
The goto tag was wrong, probably due to a copy-paste mistype while
reimplementing the function.
Closes: Coverity CID#336546
Change-Id: I06b810fde7bf750fcb42d6d9e6223883e26f5f0b
According to Coverity, we check return value of this function in
all other cases except this one (9 out of 10 times), so let's add
the missing assert(), just to be sure.
Change-Id: I675f4089cc990be5fcda792276b6808742f6f0d7
Fixes: CID#336557
If the given queue is empty, queue->list.next points to &queue->list.
Current implementation would call llist_del() on the queue's llist_head,
decrement queue->current_length (which will be 0), and return a pointer
to &queue->list to the caller. This is completely wrong.
- Use the existing item_dequeue(), which does exactly what we need.
- Do not decrement the current_length if nothing was dequeued.
- Uncomment code in the unit test, we should not crash anymore.
Change-Id: I63094df73b166b549616c869ad908e9f4f7d46d1
Fixes: CID#336557
This patch is adding a simple testcase, which does the following:
* enqueue up to the limit (12 items),
* dequeue up to the limit (12 items).
Everything works as expected, unless you attempt to dequeue from
an empty queue: the test binary segfaults. The problem is explained
and fixed in a subsequent patch.
Change-Id: Ie0edbf00e656fbe231952bdbccfd37d143e8b2b1
Related: CID#336557
I am seeing this when building with gcc v13.2.1:
tests/coding/coding_test.c: In function ‘test_pdtch’:
tests/coding/coding_test.c:444:23: warning: ‘*result[<unknown>]’
may be used uninitialized
444 | result[len - 1] &= 0x7f;
| ~~~~~~^~~~~~~~~
tests/coding/coding_test.c:448:23: warning: ‘*result[39]’
may be used uninitialized
448 | result[len - 1] &= 0x07;
| ~~~~~~^~~~~~~~~
The idea here is to pre-clear some bits in the resulting buffer,
because they're not going to be set during decoding of the burst
bits. The problem is that result[] holds uninitialized data, so
we're basically taking a 'garbage' octet and clear some of its
bits. The remaining 'garbage' bits of that octet are overwritten
by the decoder, so in the end we still get deterministic results.
Let's make GCC happy by clearing all bits in the last octet.
Change-Id: I24d79de8b3a5f4184b71414504657e5857498e0e
This patch brings decoding of the CS1 in consistency with the other
three coding schemes, for which we support decoding USF independently.
Change-Id: I61a3628741c0ac68374fc7f077cf3a07e51277c3
The USF (Uplink State Flag) field is present in the MAC header of all
Downlink PDCH blocks. It is used by the network to indicate which MS
can transmit on subsequent Uplink PDCH block(s). This field is of a
high importance for the MS, thus the decoder API allows the caller
to obtain USF value separately from the actual data bits.
In the case of gsm0503_pdtch_decode(), if the 'usf_p' pointer is not
NULL, the USF value would be assigned for CS2/CS3/CS4 (but not CS1)
even if the CRC check fails (negative return value). A subsequent
patch is to bring the CS1 in consistency with CS2/CS3/CS4.
In the case of gsm0503_pdtch_egprs_decode(), decoding of the USF
field separately from data bits is not implemented, and moreover
the function itself cannot be used for decoding Downlink blocks.
Change-Id: I43e8bfb4003f34766ace7c5c6080ca583ce5efbb
An extra osmo_multiaddr_ip_and_port_snprintf() API is introduced which
is used by osmo_sock_multiaddr_get_name_buf() but which will also be
used by other app uers willing to use
osmo_sock_multiaddr_get_ip_and_port() according to its needs (eg. only
printing the local side).
Related: SYS#6636
Change-Id: I48950754ed6f61ee5ffa04a447fab8903f10acc0
This API will be used internally by osmo_sock_multiaddr_get_name_buf()
in a follow-up patch.
This API can also be used directly by user who wish to obtain a list of
local/remote IP addresses and port from an SCTP socket.
Related: SYS#6636
Related: OS#5581
Change-Id: I19d560ab4aadec18a4c0f94115675ec1d7ab14d7
Patch [1] merged few weeks ago, for yet unknown reasons, sprinkles nul
characters at seemingly randomly chosen log line ends.
Trying to figure out why that happens, i got tired of the unreadable
cruft, and decided to migrate the _output_buf() implementation to
osmo_strbuf first.
With osmo_strbuf in use and implementing 1:1 what the previous code did,
the odd nul octets have disappeared. So the bug was caused by unreadable
code.
[1] 11a416827d
Ia7de9d88aa5ac48ec0d5c1a931a89d21c02c5433
"logging: ensure ANSI color escape is sent in same line/before newline"
Related: OS#6284
Related: Ia7de9d88aa5ac48ec0d5c1a931a89d21c02c5433
Change-Id: Ib577a5e0d7450ce93ff21f37ba3262704cbf4752
Upcoming patch adopts osmo_strbuf in logging.c, which sometimes needs to
steal and re-add trailing newline characters, and also needs to let
ctime_r() write to the buffer before updating the osmo_strbuf state.
Related: OS#6284
Related: Ib577a5e0d7450ce93ff21f37ba3262704cbf4752
Change-Id: I997707c328eab3ffa00a78fdb9a0a2cbe18404b4
This code already exists twice, and upcoming patch will need this as
well in logging.c. Add a macro to remove the code dup.
Related: OS#6284
Related: Ib577a5e0d7450ce93ff21f37ba3262704cbf4752
Change-Id: I6f2991125882bff948708bbb4ae218f9f3d1e50c
Those parameters are not related to binding and hence should be
applicable before binding. This allows a caller setting them while not
caring about explicit binding (OSMO_SOCK_F_BIND).
Until recently calling this function without OSMO_SOCK_F_BIND was not
really supported, so the previous placement setting these params in the
function didn't matter much. It does now.
Change-Id: Ia32510e8db1de0cc0dc36cebf8a94f09e44fda70
This is an attempt to fix several downsides of current
osmo_sock_init2_multiaddr() API, mainly the requirement to pass an explicit
local address (!NULL). It also now works fine if OSMO_SOCK_F_BIND flag
is not used.
This reimplementation is based on the follwing logic:
- If caller passed family=AF_INET or family=AF_INET6, that same family
is used and kernel will fail if something is wrong.
- If caller passes family=AF_UNSPEC, the function will try to find the
required family to create the socket. The decision is taken on the
assumption that an AF_INET6 socket can handle both AF_INET6 and AF_INET
addresses (through v4v6 mapping). Hence, if any of the addresses in the
local or remote set of addresses resolves through getaddrinfo() to an
IPv6 address, then AF_INET6 is used; AF_INET is used otherwise.
Related: OS#6279
Change-Id: I2641fbaca6f477404b094dbc53c0c1a3dd3fd2fd
Vadim Yanitskiy
Philipp Maier
Daniel Willmann
Harald Welte
Pau Espin
arehbein
Neels Hofmeyr