Force pushing all reply lumps (headers added to the reply) when using t_reply_with_body() - the function takes the msg from the transaction, not the real msg; so we need to be sure that the msg in transaction was fully cloned.

Reported by Giuseppe Cardone on mailing list.

Reported by David Sanders (@dsanders11).
Closes #644 .

(cherry picked from commit 566d4db)

(cherry picked from commit 684846d)

* proper error messages on compress
* correct buffer size aproximation on gzip compress(10% + 12)
* fix bug on decompression

(cherry picked from commit ec25c07)

(cherry picked from commit b9fe3c2)

(cherry picked from commit ffb2554)

The ACK CSEQ number does not depend on the number of pings or other in-dialog requests since the INVITE transaction, it must be the same as the INVITE
Closes issue #680

(cherry picked from commit 26a0a62)

See:
#375 (diff)

This affects timers and failure routes that are supposed to reuse a static context

This patch fixes a well-hidden TCP connection referencing bug which
would only reveal itself upon running into failed TCP send() operations
due to various reasons (e.g. full send buffer) in high-traffic environments.

Credits to Hieu Ta and Don Steul from Jibe Mobile for detailed reporting
and providing an adequate replication environment

(cherry picked from commit 3e1a1a7)

_tcpconn_find() returns a pointer which should never be used outside of
TCP lock. This patch completes commit 22f4c1a by fixing two
additional issues of the same type in tcpconn_add_alias()

(cherry picked from commit cb5eebe)

Also add a handy macro for future usage

(cherry picked from commit 48b88bd)

Conflicts:
	modules/emergency/emergency_methods.c
	modules/emergency/subscriber_emergency.c
	modules/usrloc/udomain.c

Credits go to 46Labs for discovering and providing test scenarios

(cherry picked from commit 28bcadf)

Credits go to 46labs for debugging this

(cherry picked from commit c18c8ea)

If the async engine in TM detects a SIP retransmission, break script and run the Post Script callbacks, othewise the transaction will stay ref'ed for ever.
Credits for spotting this problem go to Razvan Crainea.

(cherry picked from commit 3ae7e37172b5ce3142e8448972a84a49d0faa53a)

This fix prevents transactions from being unrefferenced when dangling
transaction pointers remain stored in static variables.

(cherry picked from commit 3f3b6fd)

This commit fixes context_alloc memory leak

(cherry picked from commit d69d085)

(cherry picked from commit 5286558)

* if a void string was given to the function ("") 0 value was returned which was not correct
* input values should be checked for NULL

(cherry picked from commit 19edf61)

In case the acc_evi_request() function was used on a failed request
(code < 300), send a missed event rather than the normal one.

Reported by Trevor Francis from 46Labs

(cherry picked from commit 18f5da0)

(cherry picked from commit f0a75d9)

…t record to NULL

(cherry picked from commit fb3923d)

 - suggested by Liviu Chircu

(cherry picked from commit 1cf5dd8)

(cherry picked from commit bd35916)

…ion records

(cherry picked from commit 889f353)

* database are now started in child_init
* avoid opening db conns in mod_init and leaving them in the db pool because
when fork() shall be done all the processes will use the same db
structure therefore the same tcp connection

(cherry picked from commit 47b5dc9)

(cherry picked from commit eccbf8d)

This fixes the charting of SHMEM related stats in Control Panel

(cherry picked from commit 52d5ef5)

(cherry picked from commit 0dca32b)

(cherry picked from commit 152f22e63cc3f045492367b18ed5357cf5a89f2b)

Maintain database result integrity by supporting null column values.
Closes #693 and #677

Presence_dialoginfo: Fix memory leaks in notify_body.c
(cherry picked from commit ab5cd61)

Beside SIP URI, use the SET ID too when performing FB updates to set the destination state.
(a SIP URI may be present in multiple SET IDs)
Closes  #711 , reported by @mishehu

For the entity that initiates a call, late SDP negociation is used, because
the SDP of the first entity is not known. This patch sets the late_sdp flag
for that entity, to flag the b2b logic to send the ACK with SDP.

This bug was reported on the mailing list by Sebastian Sastre.

(cherry picked from commit 20df836)

* stop inserting a contact even though received is OK
(call parse_uri() for each contact)
* at startup, if a bogus contact is found continue processing,
letting the user know that he has to remove that contact from
the database

Conflicts:
	modules/avpops/avpops.c

rtpproxy: patch: has_sdp() does not exist
(cherry picked from commit 2e59ba9)

Tested with mongo-c-driver v0.6. It turns out that the driver's
mongo_cmd_get_last_error() does not always return an iterable bson
structure. This patch adds a safety check, so that iteration is skipped.

Reported by Sasmita Panda
Closes issue #747

(cherry picked from commit 2e5160e)

(cherry picked from commit 3607208)

%*.s is a format specifier for string width
%.*s is a format specifier for string length

99% of the time, we want the latter one and _not_ the former.

(cherry picked from commit e0c0a54)

Conflicts:
	modules/event_flatstore/event_flatstore.c
	modules/proto_tls/tls_params.c
	modules/usrloc/dlist.c

This prevents OpenSIPS from blocking waiting for a connection to be
established.

Fixes #753 reported by Nick Altmann

(cherry picked from commit 76034ad)

Do not use the static counter as the timer routine gets executed in various processes now - better use a shm counter.
See all the details in #751
Based on the PR #751
Closing / outdating #751
Credits for finding the fix and comming up with a solutoin go to @aerringer.

Overlapping of retransmission timer and wait timer may lead to inconsistent timer list -> various crashes around TM timer lists.
Related to #749, #732, #723.
Many thanks for Trevor Francis ( @46labs )  for helping with the troubleshooting.

IMPORTANT NOTE: this is a temporary fix as it simply avoid the collision between the two timer routines - normally the routines themselves should be safe to run in parallel (more digging is required here).

(cherry picked from commit ea6e1fa)

Do not do anything about branches with no received reply; better continue the retransmission hoping to get something back; if still not, we will generate the 408 Timeout based on FR timer; this helps with better coping with missed/lated provisional replies in the context of cancelling the transaction

Many thanks to Maxim Sobolev for bringing this under discussion.
More to be read:
http://lists.opensips.org/pipermail/devel/2015-November/018623.html
http://lists.opensips.org/pipermail/devel/2015-December/018810.html

(cherry picked from commit 0dfb46d)

(cherry-picked from 6e7535207d5531e0fc2562d05b874acd30488368)

(cherry picked from commit e36453e)

(cherry picked from commit 02defafe421fa1ff4ab0f05f63cc490fa1717ff3)

This should fix a memory leak (reported in #743 by Eric Tamme)

(cherry picked from commit 29190f3)

(cherry picked from commit 666cf83)

…tions

(cherry picked from commit c537cb8)

Conflicts:
	modules/dialog/dlg_tophiding.c

(cherry picked from commit e09f09a)

…trol flow, invalid expressions

(cherry picked from commit b22844e)

Conflicts:
	modules/drouting/drouting.c

…, null dereference

(cherry picked from commit aeedb7d)

Conflicts:
	modules/dispatcher/dispatcher.c

(cherry picked from commit c004967)

Conflicts:
	db/db.c

… check

(cherry picked from commit b8e2318)

(cherry picked from commit 92a7361)

…t check, db_http

(cherry picked from commit d471234)

(cherry picked from commit b578dda)

(cherry picked from commit 906bb91)

Conflicts:
	cachedb/cachedb_id.c

(cherry picked from commit 81aeba4)

(cherry picked from commit fd57609)

(cherry picked from commit 939b343)

(cherry picked from commit 446dd58)

(cherry picked from commit 9bd4c77)

leave IPPROTO_* format in hep header(which is returned by
pipport2su() function) and after setting it use OpenSIPS
format for sending(PROTO_UDP)

- remove obsolete triggers
- change tarball name and url

 - closes #758 - uac_registrant wrong hash sequence

(cherry picked from commit be371da)

Now the calls per minute parameter is build as consisting of an initial
time(t0) and a 60 seconds window(as was before). Each window slot holds the
calls received in that second(as before). The difference is that now
these parameters are updated based on three conditions:
* we receive the call between [t0; t0 + WINDOW_SIZE(60s)] - we only update
the number of calls for that slot(t0 + current time) and the calls per minute
parameter
* we receive the call in the interval of [t0+WINDOW_SIZE; t0+2*WINDOW_SIZE] -
we do the following: move t0 in current call time - WINDOW_SIZE, invalidate
all the calls we knew of from t0 to current t0 (current call time - WINDOW_SIZE)
* we receive a call after 2 * WINDOW_SIZE(60 seconds) + t0 - this invalidates
all the calls we knew of since the window is not does not contains calls newer
than the last 60 seconds;

(cherry picked from commit dd25c63)

Reported by @ferrored on github
Closes #762

(cherry picked from commit 70e5a65)

(cherry picked from commit 8a51273)

Conflicts:
	modules/avpops/avpops.c

Reported by Søren Andersen.

dialog: Document that unset_dlg_profile can be used from request_route.
(cherry picked from commit 8d23d9b)

(cherry picked from commit 9750241)

Conflicts:
	modules/ldap/iniparser.c
	modules/usrloc/dlist.c

Credits to Ognjen Seslija for the bug report

This was braking the creation of extra tables (at least in postgres is did).
Reported by Nathaniel L. Keeling III on mailing.

(cherry picked from commit ce6e844)