Define l2tp_packet_add_int64() to create attributes of 64 bits long
integers.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Set Challenge attribute using a random length so that its size can't
be guessed when hide-avps is on.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Add option "hide-avps" in the "l2tp" section for hiding attributes
sent to peer. This same option is also made available on accel-ppp's
command line interface:
  accel-ppp# tunnel create tunnel peer-addr 192.0.2.1 hide-avps 1

Attribute hiding is performed upon attribute creation (in the
l2tp_packet_add_*() functions family) rather than upon packet sending.
This avoid running the cipher for every retransmission; the counterpart
is that l2tp_packet_print() can't dump original attributes of hidden
AVPs.

Currently, only one random vector is used for all hidden AVPs in a
packet. This is easily extensible though, as the 'last_RV' field in
struct l2tp_packet_t may be overridden to use new vectors for next
AVPs.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Define and export the u_randbuf() function that fills a buffer with
random data.
Convert L2TP's challenge generation code for using it.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Decode hidden AVPs on reception. This is transparent for functions in
l2tp.c (except for the presence of the Random Vector AVP).
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Secret length is used quite often especially when handling hidden AVPs.
Store conf_secret length together with conf_secret to avoid calling
strlen(conf_secret) every time.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Received attributes of type ATTR_TYPE_INT64 are transferred to upper
layer in network byte order while any other integer type uses host
byte order.
This patch converts int64 values to host byte order so that they can be
used like other integer types.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

When adding a string AVP to an L2TP packet, the attribute value is
allocated and set using strdup(). There's no need to memcpy() it
again afterwards.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Use "info2" log messages to inform about packets sent or processed.
For HELLO and ZLB messages, the log level is set to "debug" as these
are transmitted quite often and don't bring much information (this same
logging policy is used for logging packets when conf_verbose is on).
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Log message for any packet retransmission action using log_tunnel()
and level "info2".
Remove l2tp_conn_log() since it is no longer used.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Add a log message (warn level) for any discarded message type.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Use info1 messages to log information about main session establishment
and disconnection steps. These messages are:
  * establishment requests and their origin (CLI, ICRQ or OCRQ)
  * establishment confirmation (when session is really established)
  * PPP establishment
  * disconnection requests and their reason

Other information of interest use info2. These include:
  * Peer Session ID setting
  * Session internal structure freeing
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Use info1 messages to log information about main tunnel establishment
and disconnection steps. These messages are:
  * establishment requests and their origin (CLI or SCCRQ)
  * establishment confirmation (when tunnel is really established)
  * disconnection requests and their reason

Other information of interest use info2. These include:
  * Peer port update
  * Peer Tunnel ID setting
  * Tunnel internal structure freeing
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Adapt packet logging policy to make it consistent across L2TP packet
handling functions:
  * Don't log messages if conf_verbose is off.
  * Log HELLO and ZLB messages with log_debug and use log_info2 for
    any other message type.
  * Log retransmissions with log_info2, no matter the message type.
  * Log outgoing messages right before sending them.
  * Log incoming messages right before analysing their message type.
    This unifies the way l2tp_conn_read() and l2tp_udp_read() log
    incoming messages.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Implement minimal WEN messages processing, so that they get properly
acknowledged.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Handle incomming messages differently depending on their Message Type:
  * Close tunnel only when handling tunnel establishment messages
    fails (SCCRP and SCCCN) and everytime a StopCCN is received.
  * Never close tunnel after reception of HELLO, SLI or session
    establishment messages (ICRQ and OCRQ).
  * Ignore messages with invalid first attribute type.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Interpret AVPs present in CDN to provide user with better diagnostic.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Allocate PPP channel name using the string size to be stored and check
for allocation failure. Use two snprintf() and a _malloc() instead of
asprintf() so that allocated memory will be visible by memdebug.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

In all reception or sending functions, return 0 on success or if
incomming message has been ignored (l2tp_recv_*() only). Return -1
upon any processing error.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

There is no need to warn about missing Challenge when handling SCCRQ
comming from a host for which a secret exists: this will be done
automatically upon SCCRP generation (by l2tp_tunnel_genchallresp()).
Using the log message from l2tp_tunnel_genchallresp() has the advantage
of working for missing Challenge AVPs in both SCCRQ and SCCRP messages.

This patch also removes a redundant log message in l2tp_session_recv()
when handling unknown message types.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

  * Avoid retrieving tunnel and session contexts manually; use
    l2tp_{tunnel,session}_self() instead.
  * Use l2tp_session_disconnect() instead of using an
    l2tp_send_CDN(); l2tp_session_free() sequence.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

This reverts commit df6157ae.

This reverts commit 3108115e.

If a HELLO message can't be acknowledged just log an error as tunnel
doesn't need to get closed. If too many HELLO messages aren't
acknowledged, the peer will close the tunnel itself.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

If tunnel or session establishment timeout expires, send StopCCN or CDN
to notify peer before closing tunnel/session.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Always disconnect tunnel once l2tp_tunnel_disconnect() is called,
even if sending StopCCN failed.
Change the way unknown message types are handled to comply with
new l2tp_tunnel_disconnect() signature.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

Don't terminate tunnel in case of error while handling a duplicate
message. Such errors aren't critical enough to justify termination
of a tunnel and all of its sessions.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

When starting a tunnel created upon user request, free it in case
of error.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>

A side effect of setting peer port to zero when sending SCCRQ messages
is that the tunnel can't send any new message before receiving first
peer's answer (this doesn't apply to retransmissions as peer port is
set in the original packet). It is then impossible to send StopCCN
messages within this time frame.

This patch fixes this issue using a new tunnel flag which indicates if
tunnel's peer port has a default value or if it has been set after
reception of peer's first message.
Signed-off-by: NGuillaume Nault <g.nault@alphalink.fr>