| Age | Commit message (Collapse) | Author |
|---|
|
When a PPP session terminates on its own (i.e. not on behalf of a L2TP
tunnel/session request), the l2tp_ppp_finished() callback calls the
L2TP session disconnection function without updating the session's
state.
Session disconnection code then works like if the PPP session was still
up: it tries to disconnect the PPP by calling ap_session_terminate().
But since the PPP is already terminated, it returns immediately,
without calling the l2tp_ppp_finished() callback; so session cleanup
won't happen.
This patch updates the session's state upon PPP disconnection, so that
the session disconnection code won't try to disconnect the PPP in
__l2tp_session_free().
The stat_active counter is now updated inside the l2tp_ppp_finished()
callback so that it gets called upon any PPP disconnection.
Sending of the EV_CTRL_FINISHED event has also been moved. It is now
performed in the generic part of __l2tp_session_free() because it has
to be done every time the session really terminates.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
PPP and L2TP session contexts are the same. So since l2tp_ppp_finished()
is called within the PPP context, it doesn't need to switch to the
session context for calling l2tp_ppp_session_disconnect().
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Freeing a session requires four steps:
1-The session asynchronously requests its parent tunnel to
remove the session (the tunnel must start the process).
2-The parent tunnel removes its reference to this session,
then makes an asynchronous call to the session cleanup
function.
3-The session performs its cleanup, frees its memory and
asynchronously notifies its parent tunnel.
4-The parent tunnel decrements its sessions counter (so that
it knows when it doesn't have any session depending on it).
If two requests, A and B, for removing the same session, arrive
almost simultaneously, then the following scenario may occur:
1a-The session handles request A and transmits it to its
parent tunnel.
2a-The parent tunnel removes its reference to this session
and requests the session to perform cleanup.
1b-In the mean time, the session handles request B and
transmit it to its parent tunnel.
3a-The session continues to handle request A, cleans up the
session and frees memory.
2b-The parent tunnel handles request B: it removes its
reference to the session and tells it to perform cleanup.
But the session pointer used has just been freed at step 3a.
This patch modifies step 1, so that the session transmit the
session-id instead of a pointer to the session structure. That
way, the tunnel can check if the session has already been
deleted (or is pending deletion), without having to manipulate a
potentially invalid pointer.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Use a session counter for each tunnel to avoid freeing a tunnel
while still having sessions depending on it.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
* l2tp: Fix allocation checking when adding octets AVP
* cli, tcp: Fix non-NULL terminated string reception
* Fix va_end() missing calls
* chap-secrets: implemented encryption
* auth_pap: make messages like other auth modules
* cli: check xmit_buf is not null at enter to write function
* pppoe: implemented regular expression support
* chap-secrets: implemented encryption
* ippool: fixed initialization order
* optional shaper compiling
* ppp: dns/wins code cleanup
|
|
|
|
|
|
|
|
|
|
Use a triton context per session so that log_ppp_*() functions can
prepend log messages with PPP session information.
Since functions must now be called within the right context,
allocating and freeing sessions is done in the following manner:
* Tunnels allocate new sessions within their own context and
activate the new session's context just before answering
the ICRQ.
* Freeing sessions is slightly more complex. The session is
first removed from its tunnel session list (done within the
tunnel context), then the session itself if disconnected and
freed (done within the session context).
The l2tp_tunnel_*() and the l2tp_session_*() functions must now be
used within tunnel and session contexts respectively.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Accept to create several sessions for each tunnel. Session IDs
are generated randomly and stored in a binary search tree in
their corresponding tunnel (i.e. field "sessions" of the
l2tp_conn_t structure). A new mempool is defined for session
allocations.
Generation of the session IDs is simple but quite limited: it
selects a 16 bits random ID and checks if a session with this
ID already exists. If so the allocation fails and the session
is closed. Otherwise the selected ID is used for the new
session. So tunnels that handle many sessions may reject new
ones, even if unused session IDs are available (just because
the randomly choosen ID is already in use).
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Deactivate the tunnel connection establishment timer upon
reception of SCCCN messages and use a new session specific
timer for session establishment.
This new session timer follows the same behaviour as the
tunnel timeout timer: it is activated when sending the ICRP
message and deactivated upon reception of the corresponding
ICCN message.
This approach is necessary for handling several sessions per
tunnel, but it generates the following side effect: if a
tunnel is created but no session establishment is requested
from the LAC, the tunnel will no longer be automatically torn
down (since the tunnel is correctly set up, its timer is no
longer running, but since no session establishment process
has been started, there is no session timer neither).
Later on, tunnel and session timers could be turned into
inactivity timeouts to address this limitation.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Separate tunnels and sessions allocation: initialise tunnels
after receiving SCCRQ messages (with no session inside), then
initialise sessions upon reception of ICRQ messages.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Pass a struct l2tp_sess_t as parameter of functions acting on
sessions, instead of retrieving the session from an l2tp_conn_t
structure (which will no longer be possible once tunnels will
handle several sessions).
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
When a PPP connection is torn down, close the underlying session
but not the tunnel (unless its last session has been closed).
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Make HELLO messages independent from PPP connections (HELLO
messages are tunnel specific).
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Split StopCCN message building/sending from l2tp_terminate to
make it re-usable.
Rename l2tp_terminate() to l2tp_tunnel_disconnect() for
consistency with other tunnel/session management functions.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Implement CDN message sending in order to give the possibility
to close a session without closing the underlying tunnel (close
a tunnel only when closing its last session, which is always the
case as long as session-multiplexing is not implemented).
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Use the "state" field (struct l2tp_conn_t) for tracking tunnel
states and "state1" (struct l2tp_sess_t) for sessions. The
meaning of the STATE_* has been slightly modified:
* STATE_WAIT_SCCN (tunnel): no modification.
* STATE_WAIT_ICRQ (session): now unused. Acceptation of ICRQ
messages now depend on the tunnel state (STATE_ESTB).
* STATE_WAIT_ICCN (session): no modification.
* STATE_ESTB (tunnel and session): for tunnels, means that the
tunnel is ready to accept new sessions. For sessions, means
that the L2TP session is established, but that it doesn't
carry any PPP session yet.
* STATE_PPP (session): the session is in use (i.e. it is
established and carries a PPP session).
* STATE_FIN (tunnel): the tunnel is being closed.
* STATE_CLOSE (tunnel and session): default state after
allocation/free. For session, this is also the state
used once disconnected.
Since outgoing calls are not implemented yet, STATE_WAIT_OCRP and
STATE_WAIT_OCCN are not used.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Establish tunnel and session connections upon reception of SCCCN
and ICCN messages respectively.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Rename l2tp_disconnect() to l2tp_tunnel_free() for consistency
with the new l2tp_session_free() function.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Move session specific operations out of l2tp_disconnect().
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Separate session allocation from the rest of tunnel allocation
operations. This implies a few extra modifications:
* Store destination address of the SCCRQ message in the
connection structure (and rename the "addr" field to
"lac_addr" for consistency). This information is required
for allocating the session.
* No more PPP information in log prefix: with session
multiplexing, tunnels are no longer tied to a single PPP
session, so there is no struct ppp_t to pass as parameter
to the log_switch() function.
Session allocation is currently still done inside
l2tp_tunnel_alloc(). It should rather be done at session
establishment once tunnel/session separation will be terminated.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Use struct l2tp_sess_t to separate session data from other
connection information. This is required in order to implement
session multiplexing.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Fix several errors and compiler warnings in format string
arguments.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
ssh://accel-ppp.git.sourceforge.net/gitroot/accel-ppp/accel-ppp
|
|
|
|
Stop sending messages on tunnels for which no peer is listening.
Discard retransmissions too.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
|
|
Add error detection to ensure the FD_CLOEXEC flag gets set for
every new socket.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
* commit 'e04b9b2b35ed36ceb89d6991d2322210f6dd6abc':
L2TP: Close sockets opened by l2tp_connect() upon failure
|
|
* commit '0b99fef73bce3f1abed63e581594de9d1f132312':
L2TP: Fix socket() error handling in l2tp_connect()
iprange: accept network with null mask (which actually disables iprange module)
|
|
Add error detection to ensure the FD_CLOEXEC flag gets set for
every new socket.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Centralise error management to ensure full cleanup upon failure.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
Check if the tunnel file descriptor has been successfully created.
Explicitely check for negative values to detect socket() errors.
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
|
|
|
|
|
|
Low interface creation/deletion rate is kernel issue and should be fixed in 3.5.
This reverts commit 9ae4a0151805229face3385e6c966de90c7fec29.
|
|
|
|
to ignore them, instead of refusing connection.
Signed-off-by: Denys Fedoryshchenko <denys@visp.net.lb>
|
|
|
|
|
|
|
|
|
|
to flood of connections
|
|
l2tp: add Vendor-Name attribute to SCCRP
|
|
|
|
|
|
|
|
|
