There's no reason why all users of L2CAP would need to worry about
initializing chan->nesting to L2CAP_NESTING_NORMAL (which is important
since 0 is the same as NESTING_SMP). This patch moves the initialization
to the common place that's used to create all new channels, i.e. the
l2cap_chan_create() function.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The teardown callback for L2CAP channels is problematic in that it is
explicitly called for all types of channels from l2cap_chan_del(),
meaning it's not possible to hard-code a nesting level when taking the
socket lock. The simplest way to have a correct nesting level for the
socket locking is to use the same value as for the chan. This also means
that the other places trying to lock parent sockets need to be update to
use the chan value (since L2CAP_NESTING_PARENT is defined as 2 whereas
SINGLE_DEPTH_NESTING has the value 1).
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

By default lockdep considers all L2CAP channels equal. This would mean
that we get warnings if a channel is locked when another one's lock is
tried to be acquired in the same thread. This kind of inter-channel
locking dependencies exist in the form of parent-child channels as well
as any channel wishing to elevate the security by requesting procedures
on the SMP channel.

To eliminate the chance for these lockdep warnings we introduce a
nesting level for each channel and use that when acquiring the channel
lock. For now there exists the earlier mentioned three identified
categories: SMP, "normal" channels and parent channels (i.e. those in
BT_LISTEN state). The nesting level is defined as atomic_t since we need
access to it before the lock is actually acquired.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Any code calling bt_accept_dequeue() to get a new child socket from a
server socket should use lock_sock_nested to avoid lockdep warnings due
to the parent and child sockets being locked at the same time. The
l2cap_sock_accept() function is already doing this correctly but a
second place calling bt_accept_dequeue() is the code path from
l2cap_sock_teardown_cb() that calls l2cap_sock_cleanup_listen().

This patch fixes the proper nested locking annotation and thereby avoids
the following style of lockdep warning.

[  +0.000224] [ INFO: possible recursive locking detected ]
[  +0.000222] 3.17.0+ #1153 Not tainted
[  +0.000130] ---------------------------------------------
[  +0.000227] l2cap-tester/562 is trying to acquire lock:
[  +0.000210]  (sk_lock-AF_BLUETOOTH-BTPROTO_L2CAP){+.+...}, at: [<c1393f47>] bt_accept_dequeue+0x68/0x11b
[  +0.000467]
but task is already holding lock:
[  +0.000186]  (sk_lock-AF_BLUETOOTH-BTPROTO_L2CAP){+.+...}, at: [<c13b949a>] lock_sock+0xa/0xc
[  +0.000421]
other info that might help us debug this:
[  +0.000199]  Possible unsafe locking scenario:

[  +0.000117]        CPU0
[  +0.000000]        ----
[  +0.000000]   lock(sk_lock-AF_BLUETOOTH-BTPROTO_L2CAP);
[  +0.000000]   lock(sk_lock-AF_BLUETOOTH-BTPROTO_L2CAP);
[  +0.000000]
 *** DEADLOCK ***
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Now that SMP has been converted to use fixed channels we've got a bit of
a problem with the hci_conn reference counting. So far the L2CAP code
has kept a reference for each L2CAP channel that was notified of the
connection. With SMP however this would mean that the connection is
never dropped even though there are no other users of it. Furthermore,
SMP already does its own hci_conn reference counting internally,
starting from a security or pairing request and ending with the key
distribution.

This patch makes L2CAP fixed channels default to the L2CAP core not
keeping a hci_conn reference for them. A new FLAG_HOLD_HCI_CONN flag is
added so that L2CAP users can declare an exception to this rule and hold
a reference even for their fixed channels. One such exception is the
L2CAP socket layer which does want a reference for each socket (e.g. an
ATT socket which uses a fixed channel).
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The l2cap_add_scid function is used for registering a fixed L2CAP
channel. Instead of having separate initialization of the channel type
and outgoing MTU in l2cap_sock.c it's more intuitive to do these things
in the l2cap_add_scid function itself (and thereby make the
functionality available to other users besides l2cap_sock.c).
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The LE ATT socket uses a special trick where it temporarily sets
BT_CONFIG state for the duration of a security level elevation. In order
to not require special hacks for going back to BT_CONNECTED state in the
l2cap_core.c code the most reasonable place to resume the state is the
resume callback. This patch adds a new flag to track the pending
security level change and ensures that the state is set back to
BT_CONNECTED in the resume callback in case the flag is set.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The EOPNOTSUPP and ENOTSUPP errors are very similar in meaning, but
ENOTSUPP is a fairly new addition to POSIX. Not all libc versions know
about the value the kernel uses for ENOTSUPP so it's better to use
EOPNOTSUPP to ensure understandable error messages.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

We're interested in whether an authentication request is because of a
remote or local action. So far hci_conn_security() has been used both
for incoming and outgoing actions (e.g. RFCOMM or L2CAP connect
requests) so without some modifications it cannot know which peer is
responsible for requesting authentication.

This patch adds a new "bool initiator" parameter to hci_conn_security()
to indicate which side is responsible for the request and updates the
current users to pass this information correspondingly.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

If the current process is exiting, lingering on socket close will make
it unkillable, so we should avoid it.

Reproducer:

  #include <sys/types.h>
  #include <sys/socket.h>

  #define BTPROTO_L2CAP   0
  #define BTPROTO_SCO     2
  #define BTPROTO_RFCOMM  3

  int main()
  {
          int fd;
          struct linger ling;

          fd = socket(PF_BLUETOOTH, SOCK_STREAM, BTPROTO_RFCOMM);
          //or: fd = socket(PF_BLUETOOTH, SOCK_DGRAM, BTPROTO_L2CAP);
          //or: fd = socket(PF_BLUETOOTH, SOCK_SEQPACKET, BTPROTO_SCO);

          ling.l_onoff = 1;
          ling.l_linger = 1000000000;
          setsockopt(fd, SOL_SOCKET, SO_LINGER, &ling, sizeof(ling));

          return 0;
  }
Signed-off-by: NVladimir Davydov <vdavydov@parallels.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>
Cc: stable@vger.kernel.org

We have all the necessary remote information for getpeername() when we
are in the BT_CONFIG state so this should be allowed. This is
particularly important for LE sockets where changing the security level
will temporarily move the socket into BT_CONFIG state.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The highly optimized TX path for L2CAP channels and its fragmentation
within the HCI ACL packets requires to copy data from user provided
IO vectors and also kernel provided memory buffers.

This patch allows channel clients to provide a memcpy_fromiovec callback
to keep this optimized behavior, but adapt it to kernel vs user memory
for the TX path. For all kernel internal L2CAP channels, a default
implementation is provided that can be referenced.

In case of A2MP, this fixes a long-standing issue with wrongly accessing
kernel memory as user memory.

This patch originally by Marcel Holtmann.
Signed-off-by: NJukka Rissanen <jukka.rissanen@linux.intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

When allocating the L2CAP SKB for transmission, provide the upper layers
with a clear distinction on what is the header and what is the body
portion of the SKB.
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>

The SKB for L2CAP sockets are all allocated in a central callback
in the socket support. Instead of having to pass around the socket
priority all the time, assign it to skb->priority when actually
allocating the SKB.
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>

The struct l2cap_ops field should not allow any modifications and thus
it is better declared as const.
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>

Kernel supports SMP Security Request so don't block increasing security
when we are slave.
Signed-off-by: NMarcin Kraglak <marcin.kraglak@tieto.com>
Acked-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>
Cc: stable@vger.kernel.org

-[0x01 Introduction

We have found a programming error causing a deadlock in Bluetooth subsystem
of Linux kernel. The problem is caused by missing release_sock() call when
L2CAP connection creation fails due full accept queue.

The issue can be reproduced with 3.15-rc5 kernel and is also present in
earlier kernels.

-[0x02 Details

The problem occurs when multiple L2CAP connections are created to a PSM which
contains listening socket (like SDP) and left pending, for example,
configuration (the underlying ACL link is not disconnected between
connections).

When L2CAP connection request is received and listening socket is found the
l2cap_sock_new_connection_cb() function (net/bluetooth/l2cap_sock.c) is called.
This function locks the 'parent' socket and then checks if the accept queue
is full.

1178         lock_sock(parent);
1179
1180         /* Check for backlog size */
1181         if (sk_acceptq_is_full(parent)) {
1182                 BT_DBG("backlog full %d", parent->sk_ack_backlog);
1183                 return NULL;
1184         }

If case the accept queue is full NULL is returned, but the 'parent' socket
is not released. Thus when next L2CAP connection request is received the code
blocks on lock_sock() since the parent is still locked.

Also note that for connections already established and waiting for
configuration to complete a timeout will occur and l2cap_chan_timeout()
(net/bluetooth/l2cap_core.c) will be called. All threads calling this
function will also be blocked waiting for the channel mutex since the thread
which is waiting on lock_sock() alread holds the channel mutex.

We were able to reproduce this by sending continuously L2CAP connection
request followed by disconnection request containing invalid CID. This left
the created connections pending configuration.

After the deadlock occurs it is impossible to kill bluetoothd, btmon will not
get any more data etc. requiring reboot to recover.

-[0x03 Fix

Releasing the 'parent' socket when l2cap_sock_new_connection_cb() returns NULL
seems to fix the issue.
Signed-off-by: NJukka Taimisto <jtt@codenomicon.com>
Reported-by: NTommi Mäkilä <tmakila@codenomicon.com>
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Cc: stable@vger.kernel.org

Several spots in the kernel perform a sequence like:

	skb_queue_tail(&sk->s_receive_queue, skb);
	sk->sk_data_ready(sk, skb->len);

But at the moment we place the SKB onto the socket receive queue it
can be consumed and freed up.  So this skb->len access is potentially
to freed up memory.

Furthermore, the skb->len can be modified by the consumer so it is
possible that the value isn't accurate.

And finally, no actual implementation of this callback actually uses
the length argument.  And since nobody actually cared about it's
value, lots of call sites pass arbitrary values in such as '0' and
even '1'.

So just remove the length argument from the callback, that way there
is no confusion whatsoever and all of these use-after-free cases get
fixed as a side effect.

Based upon a patch by Eric Dumazet and his suggestion to audit this
issue tree-wide.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We should let user space request the peer address also in the pending
connect states, i.e. BT_CONNECT and BT_CONNECT2. There is existing user
space code that tries to do this and will fail without extending the set
of allowed states for the peer address information.

This patch adds the two states to the allowed ones in the L2CAP and
RFCOMM sock_getname functions, thereby preventing ENOTCONN from being
returned.
Reported-by: NAndrzej Kaczmarek <andrzej.kaczmarek@tieto.com>
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Tested-by: NAndrzej Kaczmarek <andrzej.kaczmarek@tieto.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The use of __constant_<foo> has been unnecessary for quite awhile now.

Make these uses consistent with the rest of the kernel.
Signed-off-by: NJoe Perches <joe@perches.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Now that the LE L2CAP Connection Oriented Channel support has undergone a
decent amount of testing we can make it officially supported. This patch
removes the enable_lecoc module parameter which was previously needed to
enable support for LE L2CAP CoC.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The l2cap_chan->psm value is always set to a valid value for a
connection oriented channel. The l2cap_chan->sport is used for tracking
local PSM allocations but will not always have a proper value, such as
with connected sockets derived from a listening socket. This patch fixes
the sock_getname callback to always use chan->psm when returning address
information.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

When we're not connected the peer address information is undefined. This
patch fixes the remote address getting to return a proper error in case
the state is anything else than BT_CONNECTED.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Fixed channels have the same source and destination CID. Ensure that the
values get properly initialized when receiving incoming connections and
deriving values from the parent socket.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The BT_SECURITY option should also be allowed for fixed channels, so
punch the appropriate hole for it when checking for the channel type.
The main user of fixed CID user space sockets is right now ATT (which is
broken without this patch).
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

ATT channels are not connection oriented so having them use
L2CAP_CHAN_CONN_ORIENTED is quite confusing. Instead, use the new
L2CAP_CHAN_FIXED type and ensure that the MTU and CID values get
properly set.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

With the introduction of security level 4, the L2CAP sockets need to
be made aware of this new level. This change ensures that the pairing
requirements are set correctly for these connections.
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>

This is a follow-up patch to f3d33426 ("net: rework recvmsg
handler msg_name and msg_namelen logic").

DECLARE_SOCKADDR validates that the structure we use for writing the
name information to is not larger than the buffer which is reserved
for msg->msg_name (which is 128 bytes). Also use DECLARE_SOCKADDR
consistently in sendmsg code paths.
Signed-off-by: NSteffen Hurrle <steffen@hurrle.net>
Suggested-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

L2CAP RAW sockets can be used for things which do not involve
establishing actual connection oriented L2CAP channels. One example of
such usage is the l2ping tool. The default security level for L2CAP
sockets is LOW, which implies that for SSP based connection
authentication is still requested (although with no MITM requirement),
which is not what we want (or need) for things like l2ping. Therefore,
default to one lower level, i.e. BT_SECURITY_SDP, for L2CAP RAW sockets
in order not to trigger unwanted authentication requests.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The values in l2cap_chan that are used for actually transmitting data
only need to be initialized right after we've received an L2CAP Connect
Request or just before we send one. The only thing that we need to
initialize though bind() and connect() is the chan->mode value. This way
all other initializations can be done in the l2cap_le_flowctl_init
function (which now becomes private to l2cap_core.c) and the
l2cap_le_flowctl_start function can be completely removed.

Also, since the l2cap_sock_init function initializes the imtu and omtu
to adequate values these do not need to be part of l2cap_le_flowctl_init.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

LE PSM values have different ranges than those for BR/EDR. The valid
ranges for fixed, SIG assigned values is 0x0001-0x007f and for dynamic
PSM values 0x0080-0x00ff. We need to ensure that bind() and connect()
calls conform to these ranges when operating on LE CoC sockets.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Setting the BT_SK_SUSPEND socket flag from the L2CAP core is causing a
dependency on the socket. So instead of doing that, use a channel
callback into the socket handling to suspend.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

This patch adds new socket options for LE sockets since the existing
L2CAP_OPTIONS socket option is not usable for LE. For now, the new
socket options also require LE CoC support to be explicitly enabled to
leave some playroom in case something needs to be changed in a backwards
incompatible way.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Most of the values in L2CAP_OPTIONS are not applicable for LE and those
that are have different semantics. It makes therefore sense to
completely block this socket option for LE and add (in a separate patch)
a new socket option for tweaking the values that do make sense (mainly
the send and receive MTU). Legacy user space ATT code still depends on
getsockopt for L2CAP_OPTIONS though so we need to plug a hole for that
for backwards compatibility.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

This patch adds tracking of L2CAP connection oriented channel local and
remote credits to struct l2cap_chan and ensures that connect requests
and responses contain the right values.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The LE connection oriented channels have their own mode with its own
data transfer rules. In order to implement this properly we need to
distinguish L2CAP channels operating in this mode from other modes.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Along with the L2CAP Connection Oriented Channels features it is now
allowed to use both custom fixed CIDs as well as PSM based (connection
oriented connections). Since the support for this (with the subsequent
patches) is still on an experimental stage, add a module parameter to
enable it.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

There is no access to chan->sk in L2CAP core now. This change marks the
end of the task of splitting L2CAP between Core and Socket, thus sk is now
gone from struct l2cap_chan.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Adding the channel to the skb private data makes possible to us know which
channel the skb we have came from.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The parent socket is not used inside the L2CAP core anymore. We only lock
it to indirect access through the new_connection() callback. The hold of
the socket lock was moved to the new_connection() callback.

Inside L2CAP core the channel lock is now used in l2cap_le_conn_ready()
and l2cap_conn_ready() to protect the execution of these two functions
during the handling of new incoming connections.

This change remove the socket lock usage from L2CAP core while keeping
the code safe against race conditions.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>