This is addressing multiple issues regarding the loss of the
OML/RSL link to the BTS.
1.) When we lose the OML link, close down all RSL connections
on all TRXs (only tested with one TRX) and free the e1inp_line
allocated for the OML connection.
2.) When we lose the RSL link on any TRX and we know to which
lines this connection belongs, we will close down the OML connection
as we have a problem to just reactivate one RSL link.
3.) When we lose the RSL link on any TRX and we do not know
where it belongs to we will free the bfd we have allocated in the
rsl listen/accept method and we properly close the socket (i could
not test this one properly). This is made under the assumption
the BTS has not responded to the ID request.
4.) When we already have a bts->oml_link we will throw it away
and use the new link (it should not happen) and the same applies
to the rsl link.
Reducing the throttling to this value created a regression with
bringing up RSL on the nanoBTS 900. We do seem to have a bug/issue
in the bsc_init code and might send a command too early without this
longer wait period and then the state transition does not happen.
For now it is agreed that reverting is the best thing to do.
Debugged-by: Sylvain Munaut <246tnt@gmail.com>
This reverts commit f5284ae1cf.
The length field of the IPA header allows to have 16bit numbers
and I just ran into the 300 byte limit with MGCP messages. Make it
three times the size and see how long this is going to be enough.
The code in 354ef81d80 checked
for fd >= 0 but on startup the struct is memset to 0 so this
test is true. Initialize the fds to -1 to make the code work
and be able to handle all ranges of the fd.
Attempt to read the three byte IPA header. If we read 0 then
the socket can be closed otherwise we need to read three bytes, if
we don't we do have a failure.
After having parsed the header we need to evaluate the length,
if the length is bigger than would fit into our buffer we will
ignore that and print an error.
This is fixing a crash when the BTS is crashing...
This code used to be a sleep, it was changed to be a timer by Andreas
but this timer does not seem to have any use. When doing the sw load
this timer is increasing the upload time dramatically, reduce it to
make it work faster.
Fix a infinite loop when establishing a new RSL connection and the
BSC is identifying itself with a unit id of an already established
RSL connection. The infinite loops happens because we are corrupting
the the linuxlist inside the bsc_fd when registering the bfd twice.
Due the lack of proper authentication favor the new RSL connection
as the real one and close the previous one.
This library is intended to collect all generic/common funcitionality
of all Osmocom.org projects, including OpenBSC but also OsmocomBB
The library currently includes the following modules:
bitvec, comp128, gsm_utils, msgb, select, signal, statistics, talloc, timer,
tlv_parse, linuxlist
msgb allocation error debugging had to be temporarily disabled as it depends on
'debug.c' functionality which at the moment remains in OpenBSC
* Close the socket when the bind is failing.
* Close the socket when the listen is failing.
* Close the socket then the bsc_register_fd is failing.
* Return an error when the socket call is not returning a socket.
we need to set newbfd->priv_nr to 2+trx_id, rather than keeping
it '2' all the time, as it is used to look-up the e1i_ts when
we receive a packet. A constant '2' would always match to TRX 0.
we also need to keep one separate bit for each TRX state in order
to properly generate the EVT_E1_TEI_UP event for trx > 0.
This is confirmed by looking at the source of their dissector.
The length can go up to 273 bytes apparently (again, according
to the source of their dissector).
Commit 9bd7bc1b4c caused the RSL
initialization to happen too early, resulting in the BTS
bootstrapping but missing [some of?] the system information
types.
In order to support multi-TRX configurations, we need to be able
to cope with multiple RSL streams (each with their own stream identifier)
inside one ip.access TCP connection.
Since this is very similar to using the TEI on a E1 line, we simply
recycle the logic and data fields that are used for the TEI.
this fixes the delay of audio caused by stalling of the openbsc process.
the use of 'usleep(100000)' for slowing down transmission to nanoBTS is
replaced by the tx-delay timer. i did this on bs11 code, so i did it the
same way. it actually queues frames for transmission not nanoBTS. on
transmission a timer is started and when this timer expires, the next
frame in the queue is transmitted (timer restarted) until the queue is
empty.
This is a regression coming from the recent split of the
handle_ts1_read method in two.
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
Signed-off-by: Holger Hans Peter Freyther <zecke@selfish.org>
For the MSC<->BSC connection we are going to use the same header
as used from BTS<->BSC but we are not having an E1-Link, a gsm network
or a gsm_bts available and can not use this part of the code.
In order to use multiple mISDN cards, we need to:
1) move driver initialization out of line initialization
2) make sure we allow partial (virtual) E1 cards with < 30 B-channels
Holger Hans Peter Freyther
Harald Welte
Sylvain Munaut