Each VLR requesting auth tuples should use a distinct IND pool for 3G
auth. So far we tied the IND to the GSUP peer connection; MSC and SGSN
were always distinct GSUP peers, they ended up using distinct INDs.
However, we have implemented a GSUP proxy, so that, in a distributed
setup, a remotely roaming subscriber has only one direct GSUP peer
proxying for both remote MSC and SGSN. That means as soon as a
subscriber roams to a different site, we would use the GSUP proxy name
to determine the IND instead of the separate MSC and SGSN. The site's
MSC and SGSN appear as the same client, get the same IND bucket, waste
SQNs rapidly and cause auth tuple generation load.
So instead of using the local client as IND, persistently keep a list of
VLR names and assign a different IND to each. Use the
gsup_req->source_name as indicator, which reflects the actual remote
VLR's name (remote MSC or SGSN).
Persist the site <-> IND assignments in the database.
Add an IND test to db_test.c
There was an earlier patch version that separated the IND pools by
cn_domain, but it turned out to add complex semantics, while only
solving one aspect of the "adjacent VLR" problem. We need a solution not
only for CS vs PS, but also for 2,3G vs 4G, and for sites that are
physically adjacent to each other. This patch version does not offer any
automatic solution for that -- as soon as more than 2^IND_bitlen
(usually 32) VLRs show up, it is the responsibility of the admin to
ensure the 'ind' table in the hlr.db does not have unfortunate IND
assignments. So far no VTY commands exist for that, they may be added in
the future.
Related: OS#4319
Change-Id: I6f0a6bbef3a27507605c3b4a0e1a89bdfd468374
Checking for existence of a subscriber and seeing that there is none is not
inherently an error. However, osmo-hlr currently logs on all occasions:
DAUC ERROR Cannot read subscriber from db: MSISDN='1001': No such subscriber
This spams the ERROR log level. Particularly when a D-GSM setup does subscriber
existence checks for every incoming mslookup request, that potentially creates
constant ERROR logging.
The "No such subscriber" part comes from db_sel(), which might also return an
sqlite3_errmsg(). We still want those sqlite3_errmsg()es in the ERROR log.
Hence print an ERROR log only if db_sel() returns an rc != -ENOENT.
Change-Id: I5044e9b4519b948edc4e451cef0f7830d315619b
These are seemingly orthogonal changes in one patch, because they are in fact
sufficiently intertwined that we are not willing to spend the time to separate
them. They are also refactoring changes, unlikely to make sense on their own.
** lu_fsm:
Attempting to make luop.c keep state about incoming GSUP requests made me find
shortcomings in several places:
- since it predates osmo_fsm, it is a state machine that does not strictly
enforce the order of state transitions or the right sequence of incoming
events.
- several places OSMO_ASSERT() on data received from the network.
- modifies the subscriber state before a LU is accepted.
- dead code about canceling a subscriber in a previous VLR. That would be a
good thing to actually do, which should also be trivial now that we record
vlr_name and sgsn_name, but I decided to remove the dead code for now.
To both step up the LU game *and* make it easier for me to integrate
osmo_gsup_req handling, I decided to create a lu_fsm, drawing from my, by now,
ample experience of writing osmo_fsms.
** osmo_gsup_req:
Prepare for D-GSM, where osmo-hlr will do proxy routing for remote HLRs /
communicate with remote MSCs via a proxy:
a) It is important that a response that osmo-hlr generates and that is sent
back to a requesting MSC contains all IEs that are needed to route it back to
the requester. Particularly source_name must become destination_name in the
response to be able to even reach the requesting MSC. Other fields are also
necessary to match, which were so far taken care of in individual numerous code
paths.
b) For some operations, the response to a GSUP request is generated
asynchronously (like Update Location Request -> Response, or taking the
response from an EUSE, or the upcoming proxying to a remote HLR). To be able to
feed a request message's information back into the response, we must thus keep
the request data around. Since struct osmo_gsup_message references a lot of
external data, usually with pointers directly into the received msgb, it is not
so trivial to pass GSUP message data around asynchronously, on its own.
osmo_gsup_req is the combined solution for both a and b: it keeps all data for
a GSUP message by taking ownership of the incoming msgb, and it provides an
explicit API "forcing" callers to respond with osmo_gsup_req_respond(), so that
all code paths trivially are definitely responding with the correct IEs set to
match the request's routing (by using osmo_gsup_make_response() recently added
to libosmocore).
Adjust all osmo-hlr code paths to use *only* osmo_gsup_req to respond to
incoming requests received on the GSUP server (above LU code being one of
them).
In fact, the same should be done on the client side. Hence osmo_gsup_req is
implemented in a server/client agnostic way, and is placed in
libosmo-gsupclient. As soon as we see routing errors in complex GSUP setups,
using osmo_gsup_req in the related GSUP client is likely to resolve those
problems without much thinking required beyond making all code paths use it.
libosmo-gsupclient is hence added to osmo-hlr binary's own library
dependencies. It would have been added by the D-GSM proxy routing anyway, we
are just doing it a little sooner.
** cni_peer_id.c / osmo_ipa_name:
We so far handle an IPA unit name as pointer + size, or as just pointer with
implicit talloc size. To ease working with GSUP peer identification data, I
require:
- a non-allocated storage of an IPA Name. It brings the drawback of being
size limited, but our current implementation is anyway only able to handle
MSC and SGSN names of 31 characters (see struct hlr_subscriber).
- a single-argument handle for IPA Name,
- easy to use utility functions like osmo_ipa_name_to_str(), osmo_ipa_name_cmp(), and copying
by simple assignment, a = b.
Hence this patch adds a osmo_ipa_name in cni_peer_id.h and cni_peer_id.c. Heavily
used in LU and osmo_gsup_req.
Depends: libosmocore Id9692880079ea0f219f52d81b1923a76fc640566
Change-Id: I3a8dff3d4a1cbe10d6ab08257a0138d6b2a082d9
Despite LTE/EUTRAN using the same authentication procedure (UMTS AKA)
as 3G, there's one difference: The "operator defined" 16bit AMF field
is reduced to 15 bits, with the first bit now being used as 'separation
bit'. That bit should be '0' for 2G/3G (as it is right now) and '1'
for authentication vectores generated for authentication over
EUTRAN/EPS.
Depends: libosmocore I93850710ab55a605bf61b95063a69682a2899bb1 (OSMO_GSUP_RAT_TYPES_IE)
Change-Id: Ic766bc40f6126bb479bd0a05b0e96bec3e240008
Previous commit changed db_test to output code names to fix issues on
some platforms (I guess due to different error code values), but somehow
this log line was not updated.
Fixes: 8b860e54be
Change-Id: I598de6f83a86d528174d3d188596314572f5d70d
Replace commented out size check for Ki with a real check, and use it
consistently for Ki, K, OP and OPC. Add a test that sets all keys to the
wrong size and tries to read them.
Related: OS#2565
Change-Id: Ib8e8e9394fb65c6e7932ce9f8bebc321b99f7696
Check if a subscriber exists without generating an error log entry if
it does not. This is cheaper than db_subscr_get_by_msisdn(), as it
does not fetch the subscriber entry.
subscriber-create-on-demand will use this function to generate
a random unique MSISDN for new subscribers.
Related: OS#2542
Change-Id: Ibfbc408c966197682ba2b12d166ade4bfeb7abc2
Check if a subscriber exists without generating an error log entry if
it does not. This is cheaper than db_subscr_get_by_imsi(), as it does
not fetch the subscriber entry. subscriber-create-on-demand will use
this function.
Related: OS#2542
Change-Id: I63818c0dd4fd22b41dadeeba2a07a651b5454c54
Allow creating new subscribers without giving them access to CS or PS.
This will be used by the create-subscriber-on-demand feature.
Related: OS#2542
Change-Id: I1a6dd85387723dab5487c53b33d2d9ec6d05d006
Extend the database scheme, add imei to the hlr_subscriber struct and
create db_subscr_update_imei_by_imsi() and db_subscr_get_by_imei(). The
new functions are used in db_test, and in follow-up commits [1], [2].
Upgrade DB schema to version 2. SQLite can only insert new columns at
the end of the table, so this happens when upgrading the database. In
new databases, the column is placed after the IMEISV column (where it
makes more sense in my opinion). This should not have any effect, as
we never rely on the order of the columns in the tables.
Follow-up commit [1] will make use of this column to save the IMEI as
received from the MSC/VLR with the Check-IMEI Procedure. It was
decided to use Check-IMEI instead of the recent Automatic Device
Detection Procedure (which would send the IMEISV) in OS#3733, because
with Check-IMEI we don't need to rely on very recent releases of the
specification.
[1] change-id I09274ecbed64224f7ae305e09ede773931da2a57
"Optionally store IMEI in subscriber table"
[2] change-id I1af7b573ca2a1cb22497052665012d9c1acf3b30
"VTY: integrate IMEI"
Depends: Id2d2a3a93b033bafc74c62e15297034bf4aafe61 (libosmocore)
Related: OS#2541
Change-Id: If232c80bea35d5c6864b889ae92d477eeaa3f45d
Verify that it returns -ENOENT on non-existing IMSI and -ENOKEY for no auth
data.
Move the auc_compute_vectors() stub to the top near the db_get_auc() call, and
just return num_vec to get a successful return value when auth data is present.
Change-Id: Ic0158228afbd78b8ca21f62dffa9f868674682b9
Differentiate between "IMSI unknown" and "IMSI has no auth data": in case of
known IMSI lacking auth data, return -ENOKEY instead of -ENOENT.
Fix API doc comments to reflect what the functions actually return, on top of
adding the -ENOKEY detail.
Adjust db_test expectations from -ENOENT to -ENOKEY where appropriate.
Adjust VTY and CTRL command rc evaluation.
A subsequent patch will use these return values to log details on each of these
situations.
Change-Id: Icf6304d23585f2ed45e050fa27c787f2d66fd3f7
A user on openbsc@ complained that with SQLite 3.8.2, the db_test fails with
--- expected
+++ stderr
-DDB (2067) abort at 18 in [INSERT INTO subscriber (imsi) VALUES ($imsi)]: UNIQUE constraint failed: subscriber.imsi
+DDB (2067) abort at 35 in [INSERT INTO subscriber (imsi) VALUES ($imsi)]: UNIQUE constraint failed: subscriber.imsi
i.e. a trivial difference in the error message issued by SQLite.
For db_test, don't output any SQLite error messages: Add argument
enable_sqlite_logging, pass as true, except in db_test.c.
Remove the SQLite error messages from expected output.
(Note that there is a src/db_test.c program that's not of interest here, this
is about the tests/db/db_test.c)
Change-Id: I2513d71cc0072aef8d08f47d0a1959f311176229
Use the common db_bind_int64() so that the stmt bindings are cleared for any
errors and to get error logging for free.
On error with sqlite3_step(), log the SQL error message, and make sure the stmt
is cleared of bindings and reset.
After sqlite3_step(), verify that exactly one row was modifed, log and return
errors otherwise.
After this patch, the DB interaction closely matches the other (refactored) DB
functions.
Change-Id: I0d870d405e2e0a830360d9ad19f0a3f9e09d8cf2
Adopt the error handling of the other db functions: return -ENOENT on unknown
subscriber and -EIO on SQL failure. Return 0 for no error, instead of the
number of rows modified.
Adjust the single caller: db_get_auc()
(and db_test.c).
Change-Id: I006f471962bdad95d00a3a4c41a28ebbc9740884
Add ability to add and remove auc_2g and auc_3g table rows with
db_subscr_update_aud_by_id().
In db_subscr_delete_by_id(), make sure that when deleting a subscriber, also
all auth data associated with that user ID is removed as well. A newly created
subscriber must not obtain the same auth tokens just by getting the same id.
Depends: libosmocore Idf75946eb0a84e145adad13fc7c78bb7a267aa0a
Change-Id: Icb11b5e059fb920447a9aa414db1819a0c020529
Use named parameters in the SQL statements.
Use db_bind_* functions to drop some code dup.
Adopt error handling (rc and logging) to match the other db functions: return
-ENOENT for unknown subscriber, -EIO for SQL failures.
Change-Id: Iad49d29b90a708c6cf55bfb3bcc02d9e29001a15
Use named parameters in the SQL statement.
Use db_bind_* functions to drop some code dup.
Use explicit subscriber id arg instead of subscriber struct.
Match return values and error logging to other db functions.
Change-Id: I35665e84ddbe54a6f218b24033df969ad2e669a0
Allow to set nam_ps and nam_cs from this same function, by adding the is_ps
arg.
Combine both NAM_PS stmts to DB_STMT_UPD_NAM_PS_BY_IMSI, add another such stmt
for CS. Use named parameters instead of parameter indexes.
Improve error return values as well as error logging to clearly indicate
whether the operation could not find the requested IMSI, or other errors
occured.
Adjust the single caller.
This prepares for upcoming VTY and possibly CTRL commands, and the error
handling introduced here has been or will be adopted by other functions in
previous or subsequent patches.
Change-Id: I6e70e15228f5bb10bee6758ae5dc9687d65839bd
Factor out the selected SQL columns as SEL_COLUMNS macro, so that each of the
new DB_STMTs will select identical columns: the old DB_STMT_SEL_BY_IMSI as well
as the new DB_STMT_SEL_BY_MSISDN and DB_STMT_SEL_BY_ID.
Add the new functions db_subscr_get_by_msisdn() and db_subscr_get_by_id() and
factor out common parts with db_subscr_get_by_imsi() to static db_sel().
Change-Id: I6d0ddd1b7e3f6b180b4b1b2663c5725d2a4a9428
db_remove_reset() needs to be called after each stmt run, whether it succeeded
or not.
In case sqlite3_clear_bindings() would fail to unbind a stmt, we would anyway
be beyond recovery. There seem to be no plausible situations where such failure
would occur, unless there have been no bindings in the first place.
In case there was an SQL stmt failure, sqlite3_reset() will re-barf the same
error message, we will always have logged it earlier already in the proper
context.
We are never evaluating the return value, nor would we know how to recover from
non-success.
The conclusions:
- db_remove_reset() does not need to log any errors.
- db_remove_reset() does not need to return success.
Change-Id: I21678463e59f607f5f5c5732963e274392f0fffd
Neels Hofmeyr
Harald Welte
Pau Espin
Oliver Smith