|Age||Commit message (Collapse)||Author||Files||Lines|
Do the same as we do in 2G: release the connection as soon as nothing else
is pending for a given subscriber.
Before, osmo-cscn would wait for the UE "to get bored" and send an Iu
release. But the CN should stay lean on connections. Also, 25.413 in section
7, 6th point states:
"While the Iu release is managed from the CN, the RNC has the capability to
request the release of all Iu connection resources from the corresponding Iu
So far we did not manage Iu release from osmo-cscn at all.
Use the same mechanism we use in 2G: from msc_release_connection(), just before
freeing the gsm_subscriber_conn, invoke a CN initiated Iu Release command to
This works around OS#1816 ("USSD only works when IuCS is released", on nano3G),
because the Iu conn is now released right after every signalling, so that
typically no two requests will use the same conn.
In iu.h/iu.c, add iu_tx_release(), absorbing almost all of the code from
Add stub to db_test.c, necessary to build it without linking libiu.
 3GPP TS 25.413 v12.4.0 Release 12 / ETSI TS 125 413 V12.4.0 (2015-04)
Send IuCS RAB Activation upon MNCC_CALL_PROC_REQ.
Implement function msc_call_assignment(): decide between sending A-iface BSSMAP
Assignment Request or IuCS RAB Assignment Request.
Implement iu_rab_act_cs() to send the IuCS RAB Assignment Request. The IP
address and port of the MGCPGW sent in the RAB Assignment are still hardcoded.
The A-interface extension is not implemented yet.
Declare ranap_new_msg_rab_assign_voice() to avoid including
ranap_msg_factory.h, which would require adding ASN1 CFLAGS to Makefile.am.
The mgcpgw_client as well as some more osmo-iuh functions are now linked from
libmsc, hence add some dummy stubs to libiudummy and db_test.c.
Add libiu function to send a CommonID message down a UE connection,
iu_tx_common_id(); add also a corresponding stub to libiudummy for linking with
Add libmsc function msc_tx_common_id() to call the above. Add this mostly to
clearly indicate in msc_ifaces.h that libmsc is calling out of the MSC; also
to do conn->via_iface checking.
Call msc_tx_common_id() after ciphering is established, in
_gsm48_rx_mm_serv_req_sec_cb()'s GSM_SECURITY_SUCCEEDED case.
The recent shifts and cuts have made some library linking for bsc, channel and
db tests unnecessary.
Temporarily disable all paging to be able to link libmsc without libbsc.
Skip the paging part of channel_test because the paging is now disabled.
In osmo-nitb, paging is done on BSC level and MSC level "at the same time".
When the new CSCN is fully operational, paging will be controlled separately on
the MSC level, and the BSC (RNC) level will be instructed over an IuCS or
A-interface to negotiate paging with the MS (UE). This MSC level paging does
not yet exist and will be added in subsequent commits.
Time zone used to be configurable per-BTS. In the upcoming MSC-split, no BTS
structures will be available on the MSC level. To simplify, drop the ability to
manage several time zones in a core network and place the time zone config on
the network VTY level, i.e. in gsm_network. If we are going to re-add fine
grained time zone settings, it should probably be tied to the LAC.
Adjust time zone VTY config code (to be moved to libxsc in subsequent commit).
Adjust time zone Ctrl Interface code.
Implement connection_for_subscr() from a completely different angle: instead of
looking up lchans in bts structs, look up the subscriber in the global list of
gsm_subscriber_connection. static lchan_find() is thus obsoleted.
All callers of connection_for_subscr() live in libmsc, so move to libmsc.
The move and edit are done in a single commit since the old and new
implementation have nothing in common.
Future: osmo-cscn will use this, without bts being present.
Remove implementation of connection_for_subscr() from channel_test.c -- it is
possible that the abort() in there was intended for a regression test, but
actually it seems the implementation was merely added for linking reasons, and
the abort() added to guard against the NULL return value: no comment nor the
commit log indicate that the abort() is test critical; the addition was the
only change in channel_test.c for that commit; at the same time a
connection_for_subscr() call was added in libmsc.
Because of libxsc, tests/gsm0408,subscr,trau no longer need libmsc.
libbsc and libmsc will have separate subscriber connection structs. Hence don't
rely on gsm_subscriber_connection, but work on a direct pointer to the counter
for the next RP reference.
The only very thin function in gsm_04_11_helper.c thus becomes obsolete: drop
the entire file.
Reincarnate gsm_network_init() as the parts not specific to libbsc.
Move from bsc_network_init() those bits that are not BSC specific (and useful
for upcoming osmo-cscn).
Add libxsc to all linkages that use gsm_network_init().
Note: the only requirement to allow linking gsm_network_init() without libbsc
is to keep the call to gsm_net_update_ctype() out of libxsc. The other items
are kept out of libxsc because it makes sense semantically. But the separation
is not strong in that the BSC specific data members are of course still
omnipresent in struct gsm_network. If bsc_network_init() is not called, these
are not initialized properly -- for now no users of uninitialized members
So this is just a first step towards a sensible split of the BSC and MSC
gsm_network structs. The long term aim should be to have entirely separate
structs with some common general items.
bsc_network_init() is more fit to live in a BSC specific header.
The gsm_network_init() function initializes a whole lot of BSC specific stuff.
Aiming to move some of it to libxsc, first rename it to bsc_network_init().
This will retain the BSC specific stuff when the move is done.
Adjust all callers.
Future: osmo-cscn will call the more generic part and not the BSC specific
the OML attribute tables are hardcoded. To set variable parameters,
the hardcoded data structure (tlv) is patched on byte level during
runtime. This patch replaces this mechanism.
- Replace hardcoded OML attribute tables with dynamically
generated TLV structures.
- Add unit tests to check if the OML attribute tables are
- Put OML attribute table generator code in a separate file:
- missing break in gprs_sndcp_pcomp.c, line 143
- string overflow in slhc_test.c, line 211
- sizeof mismatch in gprs_sndcp_xid.c, line 1369 and 1378
- mismatching signedness in gprs_sndcp_xid.c, line 1377
- needless < 0 comparison in gprs_sndcp_xid.c, line 477
- needless < 0 comparison in gprs_sndcp_xid.c, line 209
- missing returncode check in v42bis_test.c, line 320
- wrong pointer dereferentialization in gprs_sndcp_comp.c, line 73
Drop extern definitions of talloc_msgb_ctx and use msgb_talloc_ctx_init()
In sgsn_test.c, use a local variable msgb_ctx to do the talloc report
from the return value of msgb_talloc_ctx_init().
Add msgb_talloc_ctx_init() call to many main() functions still lacking a
msgb talloc context.
Tests for V.42bis, slhc, llc-xid and sndcp-xid are built and
executed on all build configurations, but are only needed when
the sgsn (gprs) is built. This patch adds conditions check
if the tests mentioned abvove are needed or not.
Most of this file uses four spaces of indenting. Replace all tabs with spaces.
Remove the erratic 'set' from the recently added vim comment at the bottom.
Add option to TextTestRunner that shows print output on stdout.
It's better to see everything in our jenkins runs and not hide test output.
In nat_msc_test(), upon socket timeout, retry up to six times. Reduce the
timeout between retries. This should get rid of sporadic test failures that
we've been seeing a lot on jenkins lately.
Raise an exception upon unexpected vty response.
Print more detail to stdout. Since we would actually want as much output as we
can get in a test suite, remove the 'if (verbose)' and just always print the
connection source. unittest is keeping all stdout silent by default anyway.
In gsm_data_shared.c, add ts_pchan() to determine actual pchan type for dynamic
and non-dynamic TS.
Use in ts_subslots() to fix the value returned for TCH/F_PDCH in PDCH mode.
Adjust the assertion in channel_test.c accordingly.
Drop GSM_PCHAN_TCH_F_PDCH, which is now handled in ts_pchan().
Explicitly add GSM_PCHAN_PDCH as zero in subslots_per_pchan (cosmetic).
Adjust the comment in subslots_per_pchan.
The fix for the number of subslots affects only one caller: bts_chan_load() in
chan_alloc.c. Before this, it would always include a TCH/F_PDCH in the
load_counter->total, now it is skipped when in PDCH mode. Whether this is the
way bts_chan_load() should handle dynamic TS is a separate discussion, so far
I'm only making sure that the two dyn TS kinds act in the same way:
TCH/F_TCH/H_PDCH is only counted when in TCH mode, and TCH/F_PDCH should match.
Add test_dyn_ts_subslots() and call from main(). Update channel_test.ok.
This includes erratic assert to show a bug for TCH/F_PDCH in PDCH mode: the nr
of subslots should be the same as for a normal PDCH, i.e. zero. This will be
adjusted along with the fix in an upcoming commit.
Move the main() guts to test_request_chan(), so that I can add another test in
an upcoming commit.
- Adding Testcases for UNCOMPRESSED_TCP and TYPE_IP
- Minor cosmetic changes
Instead of below error, raise an exception to describe what's happening.
Seen in a jenkins run on https://gerrit.osmocom.org/#/c/945/2:
ERROR: testBSCreload (__main__.TestVTYNAT)
Traceback (most recent call last):
File "./vty_test_runner.py", line 787, in testBSCreload
msc = nat_msc_test(self, ip, port)
File "./vty_test_runner.py", line 1251, in nat_msc_test
UnboundLocalError: local variable 'conn' referenced before assignment
Decouple the root talloc context from libbsc's global talloc_bsc_ctx.
This allows to define the root talloc ctx from a main() scope, which in turn
helps decouple libmsc from libbsc.
- Add compression control for V.42bis Add code to handle compression
- Add Adjustments in SNDCP
- Add VTY commands
- Edit previously committed V.42bis implementation to function
- Add unit test to verify the correct function of V.42bis
- Add module to handle compression entities
- Add module to control header compression
- Introduce VTY commands for heade compression configuration
- Add changes in sndcp and llc to integrate header compression
The previously pushed slhc implementation has been modified to compile
and function outside of the kernel. Also debug log messages were added
and datatypes ware matched. The implementation is now ready to be used
The SNDCP-XID (or layer-3 xid) is used to exchange layer-3 parameters
such as compression. The encoder encodes a bytestream that is then
sent as regular XID field from LLC.
We will need the SNDCP-XID to negotiate the parameters for our
upcomming GPRS data and header compression features
Some fixes for build environments where dependencies are installed in
In gsm48_gmm_sendmsg(), redirect to iu_tx() for both cases of MM context
present or not.
In gsm48_rx_gmm_att_req(), compose an MM context marked as Iu for messages
coming in from a ue_conn_ctx (passed in msg->dst). Also make sure cid is
initialized to avoid introducing a compiler warning.
In gsm48_rx_gmm_ra_upd_req(), look up an Iu MM context based on the presence of
the ue_conn_ctx in msg->dst.
In sgsn-test, add libiu and libasn1c, libosmo-sigtran, libosmo-ranap, which are
now needed for an --enable-iu build.
With this commit the already existing XID mechanism has been
modified to suit the needs for the upcomming SNDCP-XID patches.
This commit should not break anything since it does not alter
the current behaviour (incoming XID is still just echoed, on
GMM-Reset a basic XID message is still echoed)
The lle-xid encoder/decoder is needed to encode and decode llc
xid parameter messages. We need this to exchange sndcp-parameters
(SNDCP-XID) and also simple parameters such as encryption IOVs
Changing the test to allow still allocated block from the rate
Add GSM_PCHAN_TCH_F_TCH_H_PDCH in gsm_pchant_names and gsm_pchant_descs: the
VTY and CTRL can now handle the new pchan type.
Adjust the CTRL iface test to expect the new PCHAN type in the output.
Fixes make check with --enable-external-tests after libosmocore commit
fd80f5a04239c2ab7b561401476dd89f2861748b that adds GSM_PCHAN_TCH_F_TCH_H_PDCH.
According to 3GPP TS 24.008 § 188.8.131.52 some GMM frames are not supposed
to be ciphered. Propagate information about the necessity for
encryption between MM <-> LLC to ensure only proper frames are
The A&C reference number specified in 3GPP TS 24.008 § 10.5.5.19
identifies particular request sent by network with the related response
sent by MS. The value transparently copied from request to response by
MS: the spec do not specify what exactly should be in there so we use
rand() to decrease chance for collisions.
Note: variable named 'rand' clashes with standard function rand() so it
Previously it was possible to start osmo-sgsn with "auth-policy remote"
but without "gsup remote-*" which resulted in broken setup: no MS could
perform GPRS ATTACH. Add consistency check to vty code to fix this.
Previously if subscriber was automatically created it got assigned
random MSISDN number. Make it optional (defaulting to previous behavior)
by adding following:
* new optional no-extension argument for subscriber-create-on-demand vty
* db unit tests
* vty test
Note: using the db made with new code might result in subscribers with
empty extension. Such subscribers cannot be deleted using old
code. Make sure not to mix db versions or manually fix it by editing
sqlite with external program.