An ERROR or WARNING message should generally indicate that something has gone
wrong in Asterisk. In the case of writing to a file descriptor, Asterisk is not
in control of when the far end closes its reading on a file descriptor. If the
far end does close the file descriptor in an unclean fashion, this isn't a bug
or error in Asterisk, particularly when the situation can be gracefully
handled in Asterisk.

Currently, when this happens, a user would see the following somewhat cryptic
ERROR message:

  "utils.c: write() returned error: Broken pipe"

There's a few problems with this:
(1) It doesn't provide any context, other than 'something broke a pipe'
(2) As noted, it isn't actually an error in Asterisk
(3) It can get rather spammy if the thing breaking the pipe occurs often, such
    as a FastAGI server
(4) Spammy ERROR messages make Asterisk appear to be having issues, or can even
    mask legitimate issues

This patch changes ast_carefulwrite to only log an ERROR if we actually had one
that was reasonably under our control. For debugging purposes, we still emit
a debug message if we detect that the far side has stopped reading.

Change-Id: Ia503bb1efcec685fa6f3017bedf98061f8e1b566

pjproject < 2.5.0 will segfault on a tls transport if async_operations
is greater than 1.  A runtime version check has been added to throw
an error if the version is < 2.5.0 and async_operations > 1.

To assist in the check, a new api "ast_compare_versions" was added
to utils which compares 2 major.minor.patch.extra version strings.

ASTERISK-25615 #close

Change-Id: I8e88bb49cbcfbca88d9de705496d6f6a8c938a98
Reported-by: George Joseph
Tested-by: George Joseph

Adds the TCP Keep Alive option to TCP and TLS server sockets. Previously
this option was only being set on session sockets.
http://www.tldp.org/HOWTO/html_single/TCP-Keepalive-HOWTO/
According to the link above, the SO_KEEPALIVE option is useful for knowing
when a TCP connected endpoint has severed communication without indicating
it or has become unreachable for some reason. Without this patch, keep
alive is not set on the socket listening for incoming TCP sessions and
in Komatsu's report this resulted in the thread listening for TCP becoming
stuck in a waiting state.

ASTERISK-25364 #close
Reported by: Hiroaki Komatsu

Change-Id: I7ed7bcfa982b367dc64b4b73fbd962da49b9af36

The default value was never set for audio_buffers, causing bad
audio quality.  This ensures the default is always set.

ASTERISK-25569 #close

Change-Id: I2d2ee3e644120b0f9f6ea6ab9286d7d590942a44

Fixed a bug that originally would show a negative number of
active calls occuring in Asterisk. A gauge is persistent so
incrementing and decrementing it results in a more consistent
performance. Also changed to the call to StatsD to use
ast_statsd_log_string() so that a "+" could be sent to StatsD.

ASTERISK-25619 #close

Change-Id: Iaaeff5c4c6a46535366b4d16ea0ed0ee75ab2ee7

Both transport and endpoint now check for the existence and readability
of tls certificate and key files before passing them on to pjproject.
This will cause the object to not load rather than waiting for pjproject
to discover that there's a problem when a session is attempted.

NOTE: chan_sip also uses ast_rtp_dtls_cfg_parse but it's located
in build_peer which is gigantic and I didn't want to disturb it.
Error messages will emit but it won't interrupt chan_sip loading.

ASTERISK-25618 #close

Change-Id: Ie43f2c1d653ac1fda6a6f6faecb7c2ebadaf47c9
Reported-by: George Joseph
Tested-by: George Joseph

The current logic for ICE negotiation starts it
when receiving an SDP with ICE candidates. This is
incorrect as ICE negotiation can only start when each 
call party have at least one pair of local and remote 
candidate. Starting ICE negotiation early would result 
in negotiation failure and ultimately no audio.

This change makes it so ICE negotiation is only started
when a response with SDP is received or when a response
with SDP is sent.

ASTERISK-24146

Change-Id: I55a632bde9e9827871b09141d82747e08379a8ca

Asterisk may crash when calling ast_channel_get_t38_state(c)
on a locked channel which is being hung up.

ASTERISK-25609 #close

Change-Id: Ifaa707c04b865a290ffab719bd2e5c48ff667c7b

See ASTERISK-25615.
If the transport protocol is tls and async_operations > 1, pjproject
will segfault if more than one operation is attempted on the same socket.
Until this is fixed upstream, a check has been added to throw an error
if a tls transport config has async_operations set to > 1.

ASTERISK-25615

Change-Id: I76b9a5b2a5a0054fe71ca5851e635f2dca7685a6
Reported-by: George Joseph
Tested-by: George Joseph

ASTERISK-25599 #close

Change-Id: Idbd187f711b2ec63dda949ca0f79aa0c1a0a0b6e

ASTERISK-25616 #close

Change-Id: Ibe729aaf2e6e25506cff247cec5149ec1e589319

A crash happens sometimes when performing a CLI "sip reload".  The bogus
peer gets refreshed while it is in use by a new call which can cause the
crash.

* Protected the global bogus peer object with an ao2 global object
container.

ASTERISK-25610 #close

Change-Id: I5b528c742195681abcf713c6e1011ea65354eeed

Current support for reason header did work only in SIP responses.
According to RFC3336 the reason header might appear in any SIP request.
But it seems to make most sence in BYE and CANCEL so parasing is done
there too (if use_q850_reason=yes).

Change-Id: Ib6be7b34c23a76d0e98dfd0816c89931000ac790

This reverts commit f42d22d3.

Unfortunately, using a bridge to manage T.38 state will cause severe deadlocks
in core_unreal/chan_local. Local channels attempt to reach across both their
peer and the peer's bridge to inspect T.38 state. Given the propensity of
Local channel chains, managing the locking situation in such a scenario is
practically infeasible.

Change-Id: I932107387c13aad2c75a7a4c1e94197a9d6d8a51

It will never be perfect or even pretty, mostly because of the differences
between static and dynamic contacts.

Created:

Can't use the contact or contact_status alloc functions
because the objects come and go regardless of the actual state.

Can't use the contact_apply_handler, ast_sip_location_add_contact or
a sorcery created handler because they only get called for dynamic
contacts.  Similarly, permanent_uri_handler only gets called for
static contacts.

So, Matt had it right. :)  ast_res_pjsip_find_or_create_contact_status is
the only place it can go and not have duplicated code.  Both
permanent_uri_handler and contact_apply_handler call find_or_create.

Removed:

Can't use the destructors for the same reason as above.  The only
place to put this is in persistent_endpoint_contact_deleted_observer
which I believe is the "correct" place but even that will handle only
dynamic contacts.  This doesn't called on shutdown however.  There is
no hook to use for static contacts that may be removed because of a
config change while asterisk is in operation.

I moved the cleanup of contact_status from ast_sip_location_delete_contact
to the handler as well.

Status Change and RTT:

Although they worked fine where they were (in update_contact_status) I
moved them to persistent_endpoint_contact_status_observer to make it
more consistent with removed.  There was logic there already to detect
a state change.

Finally, fixed a nit in permanent_uri_handler rmudgett reported
eralier.

ASTERISK-25608 #close

Change-Id: I4b56e7dfc3be3baaaf6f1eac5b2068a0b79e357d
Reported-by: George Joseph
Tested-by: George Joseph

ASTERISK-25584 #close

Change-Id: Iae00071b4ff1ae76f24995aeac4d00284fd14f91

When 4875e5ac was merged, it fixed several issues with a direct media bridge
transitioning to handling a T.38 fax. However, it uncovered a race condition
caused by the bridging core. When a channel involved in a T.38 fax leaves a
bridge, the frame queued by the channel driver that should inform the far side
that it is no longer in a T.38 fax may not make it across the bridge. The
bridging framework is *extremely* aggressive in tearing down the bridge, and
control frames that are currently in flight *may* get dropped.

This patch adds a new module to the bridging framework, bridge_t38. This module
maintains some notion of the T.38 state for the two channels in a bridge. When
the bridge detects that it is being torn down or when one of the two channels
leaves, it informs the respective channel(s) that they should stop faxing. This
ensures that channels switch back to audio if they survive and are ejected out
of a bridge while faxing.

ASTERISK-25582

Change-Id: If5b0bb478eb01c4607c9f4a7fc17c7957d260ea0

Beside that, the format-attribute module sends only non-default values in the
line fmtp, now. This avoids unnecessary overhead in SDP messages. Furthermore,
previously the parameter stereo was not parsed when being the first parameter.

ASTERISK-25583 #close

Change-Id: Iae85ba3e5960bfd5d51cf65bcffad00dd4875a73

This patch fixes a crash which would occur when an audiohook was
applied to a channel using an audio codec that could not be translated
to signed linear (such as when using pass-through codecs like OPUS or
when the codec translator module for the format in use is not loaded).

ASTERISK-25498 #close
Reported by: Ben Langfeld

Change-Id: Ib6ea7373fcc22e537cad373996136636201f4384

When 90d9a707 was merged, it mostly tested dynamic contacts created as
a result of registering a PJSIP endpoint. Contacts generated in this
fashion typically have a long alphanumeric string as their object identifier,
which maps reasonably well for StatsD. Unfortunately, this doesn't work in the
general case. StatsD treats both '.' and ':' characters as special characters.
In particular, having a ':' appear in the middle of a StatsD metric will
result in the metric being rejected.

This causes some obvious issues with SIP URIs.

The StatsD API should not be responsible for escaping the metric name passed
to it. The metric is treated as a single long string, and it would be
challenging to know what to escape in the string passed to the function.
Likewise, we don't want to escape the metric in PJSIP, as that involves
overhead that is wasted when either res_statsd isn't loaded or enabled.

This patch takes an alternative approach. The Contact ID has been changed
to be "aor@@uri_hash" instead of "aor@@uri". This (a) won't contain any of the
aforementioned special characters, (b) can be done on Contact creation,
which has minimal impact on run-time performance, and (c) also conforms to an
earlier commit that changed the ID for dynamic contacts.

The downside of this is that StatsD users will have to map SHA1 hashes back to
the Contacts that are emitting the statistics. To that end, the CLI commands
have been updated to include the first 10 characters of the MD5 hash, which
should be enough to match what is shown in Graphite (or some other StatsD
backend).

ASTERISK-25595 #close

Change-Id: Ic674a3307280365b4a45864a3571c295b48a01e2
Reported-by: Matt Jordan
Tested-by: George Joseph