To be consistent, lets use msec for this timeout as well.

Note: This define value is a minimum scan time taken from BT Core spec 4.0,
Vol 3, Part C, chapter 9.2.6
Signed-off-by: Lukasz Rymanowski <lukasz.rymanowski@tieto.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

With this patch it is possible to control discovery interleaved
timeout value from debugfs.

It is for fine tuning of this timeout.
Signed-off-by: Lukasz Rymanowski <lukasz.rymanowski@tieto.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

Keep msec instead of jiffies in this define. This is needed by following
patch where we want this timeout to be exposed in debugfs.

Note: Value of this timeout comes from recommendation in BT Core Spec.4.0,
Vol 3, Part C, chapter 13.2.1.
Signed-off-by: Lukasz Rymanowski <lukasz.rymanowski@tieto.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

This patch improves readability and makes future changes easier.
Signed-off-by: Bing Zhao <bzhao@marvell.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

The ssp parameter is supposed to be a mandatory one and there are no
callers that would pass NULL to this function. Removing this unnecessary
NULL check also makes (false positive) static analyzer warnings go away.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

When sending a pending device found event triggered by disabling LE
scanning we should use the stored RSSI instead of sending a zero value.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

Especially in crowded environments it can become frequent that we have
to send out whatever pending event there is stored. Since user space
has its own filtering of small RSSI changes sending a 0 value will
essentially force user space to wake up the higher layers (e.g. over
D-Bus) even though the RSSI didn't actually change more than the
threshold value.

This patch adds storing also of the RSSI for pending advertising reports
so that we report an as accurate RSSI as possible when we have to send
out the stored information to user space.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

Occasionally, during active scanning we will receive duplicate ADV_IND
reports from the same device before receiving the SCAN_RSP from them. In
order to not wake up user space unnecessarily it's better not to send
these extra events as they do not contain any new information.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

The line was incorrectly split between the variable type and its name.
This patch fixes the issue.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

All callers of hci_inquiry_cache_update() pass a non-NULL ssp pointer to
it and even the function itself assumes in another place that the
pointer is non-NULL. Therefore, remove the redundant check.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

To avoid too many events being sent to user space and to help parsing of
all available remote device data it makes sense for us to wait for the
scan response and send a single merged Device Found event to user space.

This patch adds a few new variables to hci_dev to track the last
received ADV_IND/ADV_SCAN_IND, i.e. those which will cause a SCAN_REQ to
be send in the case of active scanning. When the SCAN_RSP is received
the pending data is passed together with the SCAN_RSP to the
mgmt_device_found function which takes care of merging them into a
single Device Found event.

We also need a bit of extra logic to handle situations where we don't
receive a SCAN_RSP after caching some data. In such a scenario we simply
have to send out the pending data as it is and then operate on the new
report as if there was no pending data.

We also need to send out any pending data when scanning stops as
well as ensure that the storage is empty at the start of a new active
scanning session. These both cases are covered by the update to the
hci_cc_le_set_scan_enable function in this patch.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

When we're in peripheral mode (HCI_ADVERTISING flag is set) the most
natural mapping of connect() is to perform directed advertising to the
peer device.

This patch does the necessary changes to enable directed advertising and
keeps the hci_conn state as BT_CONNECT in a similar way as is done for
central or BR/EDR connection initiation.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

In preparation for being able to merge ADV_IND/ADV_SCAN_IND and SCAN_RSP
together into a single device found event add a second parameter to the
mgmt_device_found function. For now all callers pass NULL as this
parameters since we don't yet have storing of the last received
advertising report.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

Passive LE scanning is only used by the kernel-internal connection
establishment procedure. It makes therefore little sense to send device
found events to user space.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

When a timeout occurs using directed advertising a 0x3c error gets
generated. Since the operation is analogous to conventional connection
creation map this to the usual EHOSTDOWN error.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

Once directed advertising is brought into the picture simply the lack of
an hci_conn object when an le_conn_complete event occurs is no longer a
reliable indication that the responder & initiator values need to be
set based on our advertising address type.

This patch moves the code for setting these values outside of the
"if (!conn)" branch and ensures that they get set for any connection
where we are in the slave role.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

Any time hci_conn_add is used for an LE connection we need to ensure
that the local identity address is correctly described in the src and
src_type variables. This patch moves setting these values directly into
hci_conn_add so that callers don't have to duplicate the effort
themselves.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

This patch does a small code simplification replacing the tipical
kmalloc-copy_from_user sequence by memdup_user() helper.

Cc: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Andre Guedes <andre.guedes@openbossa.org>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

As preparation for merging ADV_IND/ADV_SCAN_IND and SCAN_RSP together
into a single mgmt Device Found event refactor individual advertising
report handling into a separate function. This will help keep the code
more readable as more logic gets added.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

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: Andrzej Kaczmarek <andrzej.kaczmarek@tieto.com>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Tested-by: Andrzej Kaczmarek <andrzej.kaczmarek@tieto.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

The smp variable in smp_conn_security is not used anywhere before the
smp = smp_chan_create() call in the smp_conn_security function so it
makes no sense to assign any other value to it before that.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

The smp pointer might not be initialized for jumps to the "done" label
in the smp_conn_security function. Furthermore doing the set_bit after
done might "overwrite" a previous value of the flag in case pairing was
already in progress. This patch moves the call to set_bit before the
label so that it is only done for a newly created smp context (as
returned by smp_chan_create).
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

Due to several devices being unable to handle this procedure reliably
(resulting in forced disconnections before pairing completes) it's
better to remove it altogether.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

In the case that the just-works model would be triggered we only want to
confirm remotely initiated pairings (i.e. those triggered by a Security
Request or Pairing Request). This patch adds the necessary check to the
tk_request function to fall back to the JUST_WORKS method in the case of
a locally initiated pairing.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

For remotely initiated just-works pairings we want to show the user a
confirmation dialog for the pairing. However, we can only know which
side was the initiator by tracking which side sends the first Security
Request or Pairing Request PDU. This patch adds a new SMP flag to
indicate whether our side was the initiator for the pairing.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

In the case that a local pairing confirmation (JUST_CFM) has been
selected as the method we need to use the user confirm request mgmt
event for it with the confirm_hint set to 1 (to indicate confirmation
without any specific passkey value). Without this (if passkey_notify was
used) the pairing would never proceed. This patch adds the necessary
call to mgmt_user_confirm_request in this scenario.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

It is possible that the HCI_LE_Start_Encryption command fails in an
early stage and triggers a command status event with the failure code.
In such a case we need to properly notify the hci_conn object and
cleanly bring the connection down. This patch adds the missing command
status handler for this HCI command.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

If a sudden disconnection happens the l2cap_conn pointer may already
have been cleaned up by the time hci_conn_security gets called,
resulting in the following oops if we don't have a proper NULL check:

BUG: unable to handle kernel NULL pointer dereference at 000000c8
IP: [<c132e2ed>] smp_conn_security+0x26/0x151
*pde = 00000000
Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
CPU: 1 PID: 673 Comm: memcheck-x86-li Not tainted 3.14.0-rc2+ #437
Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
task: f0ef0520 ti: f0d6a000 task.ti: f0d6a000
EIP: 0060:[<c132e2ed>] EFLAGS: 00010246 CPU: 1
EIP is at smp_conn_security+0x26/0x151
EAX: f0ec1770 EBX: f0ec1770 ECX: 00000002 EDX: 00000002
ESI: 00000002 EDI: 00000000 EBP: f0d6bdc0 ESP: f0d6bda0
 DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
CR0: 80050033 CR2: 000000c8 CR3: 30f0f000 CR4: 00000690
Stack:
 f4f55000 00000002 f0d6bdcc c1097a2b c1319f40 f0ec1770 00000002 f0d6bdd0
 f0d6bde8 c1312a82 f0d6bdfc c1312a82 c1319f84 00000008 f4d81c20 f0e5fd86
 f0ec1770 f0d6bdfc f0d6be28 c131be3b c131bdc1 f0d25270 c131be3b 00000008
Call Trace:
 [<c1097a2b>] ? __kmalloc+0x118/0x128
 [<c1319f40>] ? mgmt_pending_add+0x49/0x9b
 [<c1312a82>] hci_conn_security+0x4a/0x1dd
 [<c1312a82>] ? hci_conn_security+0x4a/0x1dd
 [<c1319f84>] ? mgmt_pending_add+0x8d/0x9b
 [<c131be3b>] pair_device+0x1e1/0x206
 [<c131bdc1>] ? pair_device+0x167/0x206
 [<c131be3b>] ? pair_device+0x1e1/0x206
 [<c131ed44>] mgmt_control+0x275/0x2d6
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

Make sure no timer callback is running before releasing the
datastructure which contains it.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

The LE scan type paramter defines if active scanning or passive scanning
is in use. Track the currently set value so it can be used for decision
making from other pieces in the core.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>

If vendor specific HCI commands are received from application,
we should send corresponding events to stack.
These events should be consumed in driver, only if they are for
the internal HCI commands generated by driver.

This patch fixes the vendor command 0x3f stuck problem with
above mentioned change. For example,

hcitool cmd 3f 22 fe 06 22 21 20 43 50 00
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Bing Zhao <bzhao@marvell.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

We need to ensure that we do not send events to user space with the
identity address if we have not yet notified user space of the IRK. The
code was previously trying to handle this for the mgmt_pair_device
response (which worked well enough) but this is not the only connection
related event that might be sent to user space before pairing is
successful: another important event is Device Disconnected.

The issue can actually be solved more simply than the solution
previously used for mgmt_pair_device. Since we do have the identity
address tracked as part of the remote IRK struct we can just copy it
over from there to the hci_conn struct once we've for real sent the mgmt
event for the new IRK.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

The passkey_notify and user_confirm functions in mgmt.c were expecting
different endianess for the passkey, leading to a big endian bug and
sparse warning in recently added SMP code. This patch converts both
functions to expect host endianess and do the conversion to little
endian only when assigning to the mgmt event struct.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

In Secure Connections Only mode, it is required that Secure Connections
is used for pairing and that the link key is encrypted with AES-CCM using
a P-256 authenticated combination key. If this is not the case, then new
connection shall be refused or existing connections shall be dropped.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>

It is possible that pairing fails after we've already received remote
identity information. One example of such a situation is when
re-encryption using the LTK fails. In this case the hci_conn object has
already been updated with the identity address but user space does not
yet know about it (since we didn't notify it of the new IRK yet).

To ensure user space doesn't get a Pair Device command response with an
unknown address always use the same address in the response as was used
for the original command.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

When performing SMP pairing with MITM protection one side needs to
enter the passkey while the other side displays to the user what needs
to be entered. Nowhere in the SMP specification does it say that the
displaying side needs to any kind of confirmation of the passkey, even
though a code comment in smp.c implies this.

This patch removes the misleading comment and converts the code to use
the passkey notification mgmt event instead of the passkey confirmation
mgmt event.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

In some cases the current 250ms delay is not enough for the remote to
receive the keys, as can be witnessed by the following log:

> ACL Data RX: Handle 64 flags 0x02 dlen 21               [hci1] 231.414217
      SMP: Signing Information (0x0a) len 16
        Signature key: 555bb66b7ab3abc9d5c287c97fe6eb29
< ACL Data TX: Handle 64 flags 0x00 dlen 21               [hci1] 231.414414
      SMP: Encryption Information (0x06) len 16
        Long term key: 2a7cdc233c9a4b1f3ed31dd9843fea29
< ACL Data TX: Handle 64 flags 0x00 dlen 15               [hci1] 231.414466
      SMP: Master Identification (0x07) len 10
        EDIV: 0xeccc
        Rand: 0x322e0ef50bd9308a
< ACL Data TX: Handle 64 flags 0x00 dlen 21               [hci1] 231.414505
      SMP: Signing Information (0x0a) len 16
        Signature key: bbda1b2076e2325aa66fbcdd5388f745
> HCI Event: Number of Completed Packets (0x13) plen 5    [hci1] 231.483130
        Num handles: 1
        Handle: 64
        Count: 2
< HCI Command: LE Start Encryption (0x08|0x0019) plen 28  [hci1] 231.664211
        Handle: 64
        Random number: 0x5052ad2b75fed54b
        Encrypted diversifier: 0xb7c2
        Long term key: a336ede66711b49a84bde9b41426692e
> HCI Event: Command Status (0x0f) plen 4                 [hci1] 231.666937
      LE Start Encryption (0x08|0x0019) ncmd 1
        Status: Success (0x00)
> HCI Event: Number of Completed Packets (0x13) plen 5    [hci1] 231.712646
        Num handles: 1
        Handle: 64
        Count: 1
> HCI Event: Disconnect Complete (0x05) plen 4            [hci1] 232.562587
        Status: Success (0x00)
        Handle: 64
        Reason: Remote User Terminated Connection (0x13)

As can be seen, the last key (Signing Information) is sent at 231.414505
but the completed packets event for it comes only at 231.712646,
i.e. roughly 298ms later.

To have a better margin of error this patch increases the delay to
500ms.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

This is a trivial coding style simplification by instead of having an
extra early return to instead revert the if condition and do the single
needed queue_work() call there.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

The pairing process initiated through mgmt sets the conn->auth_type
value regardless of BR/EDR or LE pairing. This value will contain the
MITM flag if the local IO capability allows it. When sending the SMP
pairing request we should check the value and ensure that the MITM bit
gets correctly set in the bonding flags.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

The SMP specification is written with the assumption that both key
information, plaintextData and encryptedData follow the same little
endian byte ordering as the rest of SMP.

Since the kernel crypto routines expect big endian data the code has had
to do various byte swapping tricks to make the behavior as expected,
however the swapping has been scattered all around the place.

This patch centralizes the byte order swapping into the smp_e function
by making its public interface match what the other SMP functions expect
as per specification. The benefit is vastly simplified calls to smp_e.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>