This meant that if a user exited the region, the UploadTexture handler would be effectively removed for everyone, causing subsequent failures.
This hopefully resolves the recent UploadTexture LLSD problems
This was a regression in 5640f2e (Thu Dec 1 23:24:15 2011 +0000)
This is required for the substitution of different HTTP servers or the newer HttpServer.dll without having to commit to a particular implementation.
This is also required to write regression tests that involve the HTTP layer.
If you need to recompile, all you need to do is replace OSHttpRequest/OSHttpResponse references with IOSHttpRequest/IOSHttpResponse.
Enabling this by setting Cap_FetchInventoryDescendents2 = "localhost" in the [ClientStack.LindenCaps] section of OpenSim.ini downloads inventory via http rather than udp in later viewers.
OutPacket() must be called within the m_killRecord lock. Otherwise the following event sequence is possible
1) LLClientView.ProcessEntityUpdates() passes the kill record check for a particular part suspends before OutPacket()
2) Another thread calls LLClientView.SendKillObject() to delete the same part and modifies the kill record
3) The same thread places the kill packet on the Task queue.
4) The earlier thread resumes and places the update packet on the Task queue after the kill packet.
This results in a ghost part in the sim that only goes away after client relog.
This commit also removes the unnecessary m_entityUpdates.SyncRoot locking in SendKillObject.
UpdateFlag is now referenced/used only within SOP and SOG. Outsiders are
using ScheduleFullUpdate, ScheduleTerseUpdate or ClearUpdateSchedule on
SOP consistently now. Also started working toward eliminating those
calls to ScheduleFullUpdate, ScheduleTerseUpdate or ClearUpdateSchedule
from outside SOP in favor of just setting properties on SOP and let SOP
decide if an update should be scheduled. This consolidates the update
policy within SOP and the client rather than everywhere that makes
changes to SOP. Some places forget to call update while others call it
multiple times, "just to be sure".
UpdateFlag and Schedule*Update will both be made private shortly.
UpdateFlag is intended to be transient and internal to SOP so it has
been removed from XML serializer for SOPs.
setting position at the same time as taint appears to undermine the whole purpose of taint
testing doesn't reveal any obvious regressions in doing this
This is necessary so that code in HttpServer can use framework facilities such as the thread watchdog for monitoring purposes.
Doing this shuffle meant that MainServer was moved into OpenSim/Framework/Servers
Also had to make OpenSim.Framework.Console rely on OpenSim.Framework rather than the other way around since it in turn relies on HttpServer
MainConsole and some new interfaces had to be moved into OpenSim/Framework to allow this. This can be reverted if parts of OpenSim.Framework stop relying on console presence (cheifly RegionInfo)
It's never possible for SOG to have no RootPart, except in the first few picosends of the big bang when it's pulled from region persistence or deserialized
The only times when ParentGroup might be null is during regression tests (which might not be a valid thing) and when scene objects are being constructed from the database.
At all other times it's not possible for a SOP not to have a SOG parent.
The root part state is the canonical value, so always send that instead.
Sending conflicting attachments states for non-root parts of a rezzed object is enough to crash the client.
Fixes http://opensimulator.org/mantis/view.php?id=5664.
Many thanks to mewtwo0641 for some fantastic qa work on this one.
This is done by introducing a PresenceType enum into ScenePresence which currently has two values, User and Npc.
This seems better than a SaveAttachments flag in terms of code comprehension, though I'm still slightly uneasy about introducing these semantics to core objects
this is to allow walking on prims. it will be up to the script writer to be sure that there is a continuous path.
currently implemented in osNpcMoveToTarget(), but none of this is final.
This now works again except that it requires a click or avatar mvmt to get going
This is because the ScenePresence.HandleAgentUpdate() method doesn't trigger until the client does something significant, at which point autopilot takes over.
Even clicking is enough to trigger.
This will be improved presently.
However, what happens now is that undo just doesn't do anything when the root prim is selected on its own. This requires more code than just fiddling with undo states.
This involves implementing a boolean in UndoState to signal whether the undo needs to be done for an entire group/linkset or just a single prim
Resizing individual components of linksets is still dodgy.
Resizing still has to be down twice, since for some reason the client is sending two multiobjectupdate packets on every resize except the very first. This applies to single prims and linksets. Need to look into this.
Undo rotation and position appear to be working.
Resizing a single prim appears to be working, though the undo has to be done twice.
Resizing a group of prims still does not work properly - possibly because in the UndoState we don't store a knowledge of when we're resizing a whole group rather than individual prims.
This needs to be addressed.
Unable to get to the bottom of why resizing a mesh fails to properly reset the physics proxy, when toggling phantom does
After a mesh is generated, the existing sculptdata is set to zero in PrimitiveBaseShape to save memory
When phantom is toggled, the sculptdata is regenerated before remeshing.
But on resize, the sculptdata is not regenerated.
So clearly, resetting sculptdata is possible, but haven't quite been able to pin down how this is being done when phantom is toggled.
Many thanks to the aurora project for pioneering this.
This code is almost certainly not bug free, but it does at least appear to handle simple meshes (except when the viewer crashes - but it is beta!).
Changed the experimental capability introduced a couple of commits ago: now sending that extra information as part of the response in the SimulatorFeatures cap.
It appears that if the viewer requests a folder containing links, we must also send the folders that contain the link targets first.
This was tested with Kokua 0.1.0 WIP though I predict it will also work with other viewer 2s
limits because the only ones used now are the defaults (which are overwritten
by the client throttles anyway). Updated the default rates to correspond to
about 350kbps.
Also added a configuration to disable adaptive throttle. The default
is the previous behavior (no adaptation).
command to look at the entity update priority queue. Added a "name" parameter
to show queues, show pqueues and show throttles to look at data for a specific
user.
per Melanie's very good suggestion. The immediate queue is
serviced completely before all others, making it a very good
place to put avatar updates & attachments.
Moved the priority queue out of the LLUDP directory and
into the framework. It is now a fairly general utility.
clients. If the sent packets are ack'ed successfully the throttle
will open quickly up to the maximum specified by the client and/or
the sims client throttle.
This still needs a lot of adjustment to get the rates correct.
Often, by the time the UDPServer realizes that an entity update packet
has not been acknowledged, there is a newer update for the same entity
already queued up or there is a higher priority update that should be
sent first. This patch eliminates 1:1 packet resends for unacked entity
update packets. Insteawd, unacked update packets are decomposed into the
original entity updates and those updates are placed back into the
priority queues based on their new priority but the original update
timestamp. This will generally place them at the head of the line to be
put back on the wire as a new outgoing packet but prevents the resend
queue from filling up with multiple stale updates for the same entity.
This new approach takes advantage of the UDP nature of the Linden protocol
in that the intent of a reliable update packet is that if it goes
unacknowledge, SOMETHING has to happen to get the update to the client.
We are simply making sure that we are resending current object state
rather than stale object state.
Additionally, this patch includes a generalized callback mechanism so
that any caller can specify their own method to call when a packet
expires without being acknowledged. We use this mechanism to requeue
update packets and otherwise use the UDPServer default method of just
putting expired packets in the resend queue.
this appears to cause problems with the system timer resolution.
This caused a problem with tokens going into the root throttle as
bursts leading to some starvation.
Also changed EnqueueOutgoing to always queue a packet if there
are already packets in the queue. Ensures consistent ordering
of packet sends.
types of property updates to be specified. Not sure if one form
of property update should supercede another. But for now the old
OpenSim behavior is preserved by sending both.
to the entity update queue. The number of property packets can
become significant when selecting/deselecting large numbers of
objects.
This is experimental code.
when client and simulator throttles are set. This algorithm also uses
pre-defined burst rate of 150% of the sustained rate for each of the
throttles.
Removed the "state" queue. The state queue is not a Linden queue and
appeared to be used just to get kill packets sent.
types of property updates to be specified. Not sure if one form
of property update should supercede another. But for now the old
OpenSim behavior is preserved by sending both.
to the entity update queue. The number of property packets can
become significant when selecting/deselecting large numbers of
objects.
This is experimental code.
when client and simulator throttles are set. This algorithm also uses
pre-defined burst rate of 150% of the sustained rate for each of the
throttles.
Removed the "state" queue. The state queue is not a Linden queue and
appeared to be used just to get kill packets sent.
time to wait to retransmit packets) always maxed out (no retransmissions
for 24 or 48 seconds.
Note that this is going to cause faster (and more) retransmissions. Fix
for dynamic throttling needs to go with this.
Handling these synchronously kills the inbound packet loop if many requests are made for remote land and those requests are handled slowly or timeout (timeout is 10s)
This can happen if a user searches for "land for sale" and then clicks many of the parcels in the list (or just presses down arrow to move through every entry).
This applies to adding/removing estate users, groups, managers and bans.
This is the application of the AllEstates_0.5.patch from http://opensimulator.org/mantis/view.php?id=5420
Thanks very much, Snoopy!
time to wait to retransmit packets) always maxed out (no retransmissions
for 24 or 48 seconds.
Note that this is going to cause faster (and more) retransmissions. Fix
for dynamic throttling needs to go with this.
In order to pass ILandObject into IClientAPI.SendLandProperties(), had to push ILandObject and IPrimCounts into OpenSim.Framework from OpenSim.Region.Framework.Interfaces, in order to avoid ci
Counts are showing odd behaviour at the moment, this will be addressed shortly.
This is in a very crude state, currently.
The LindenUDPModule was renamed LindenUDPInfoModule and moved to OptionalModules
OptionalModules was given a direct reference to OpenSim.Region.ClientStack.LindenUDP so that it can inspect specific LindenUDP settings without having to generalize those to all client views (some of which may have no concept of the settings involved).
This might be ess messy if OpenSim.Region.ClientStack.LindenUDP were a region module instead, like MXP, IRC and NPC
* Added Calculating Time Dilation in the OdePlubin
* When multiple object updates are stuffed into one packet, average the time dilation between them as a compromise.
* Time Dilation on the update is calculated when the EntityUpdate object is created. The pre-calc-ed TD is stored in the Entity update and used when it goes out on the wire. Previously, it was 1.0 all the time. The time dilation is tied to when the update is created, not when the update is sent.
* Adds an item that checks to see if the top request has been there for longer then 30 seconds without an update and sends an AbortXfer if it encounters one. This allows the client to cancel the Xfer on it's side so you can re-select the prim and get the inventory when it fails the first time.
* Some interesting locking... Using NewFiles to lock the rest of them. We'll see how that goes.
* The goal of this is to ensure that Xfers are restartable when they fail. The client will not do that on it's own.
These locks are necessary to avoid a delete/update race condition for scene objects.
However, since we're now locking on m_killRecord this shouldn't cause delays to m_entityUpdates reprioritization
Remove the too coarse CanEditParcel method in favor of a CanEditParcelProperties
method that takes a GroupPowers argument to specify what action is to be
taken. Also, make the method to set parcel data much more granular. Permissions
in a deeded setting should now work.
Revert "This may have been the biggest, baddest bug in OpenSim ever... confusion between bytes per second and bytes per millisecond."
This reverts commit 870bbcfc6c.
If an LL 1.23.5 client (and possibly earlier and later) receives an object update after a kill object packet, it leaves the deleted prim in the scene until client relog
This is possible in LLUDPServer if an object update packet is queued but a kill packet sent immediately.
Beyond invasive tracking of kill sending, most expedient solution is to always queue kills, so that they always arrive after updates.
In tests, this doesn't appear to affect performance.
There is probably still an issue present where an update packet might not be acked and then resent after the kill packet.
For each agent, this command shows how many packets have been sent/received and how many bytes remain in each of the send queues (resend, land, texture, etc.)
Sometimes useful for diagnostics
This will cause visual params to be persisted along with worn items. With
this, alpha and tattoo laters will be saved. Multiple layers MAY work, but
not tested because I don't use Viewer 2.
Setting this to true avoids a 500ms or so client freeze when the LLUDP server thread is taken up with processing a UseCircuitCode packet synchronously.
Extensive testing on Wright Plaza appeared to show no bad effects and this seems to reduce login lag considerably.
Of course, a lot of login lag is still coming from other sources.
It contains a major interface version bump and will NOT work with earlier grid
services. This is preliminary work that will lead to layers support.
Rest appearance services are commented out completely, they will have to be
adapted by someone who actually uses them. Remote admin is working, but has
no layers support. There is no layers support in the database. Login likely
won't work. You have been warned.
AvatarService -- add two new methods, GetAppearance and SetAppearance
to get around the lossy encoding in AvatarData. Preseve the old
functions to avoid changing the behavior for ROBUST services.
AvatarAppearance -- major refactor, moved the various encoding
methods used by AgentCircuitData, ClientAgentUpdate and
ScenePresence into one location. Changed initialization.
AvatarAttachments -- added a class specifically to handle
attachments in preparation for additional functionality
that will be needed for viewer 2.
AvatarFactory -- removed a number of unused or methods duplicated
in other locations. Moved in all appearance event handling from
ScenePresence. Required a change to IClientAPI that propogated
throughout all the IClientAPI implementations.
Object updates are sent on the task queue. It's possible for an object update to be placed on the client queue before a kill packet comes along.
The kill packet would then be placed on the state queue and possibly get sent before the update
If the update gets sent afterwards then client get undeletable no owner objects until relog
Placing the kills in the task queue should mean that they are received after updates. The kill record prevents subsequent updates getting on the queue
Comments state that updates are sent via the state queue but this isn't true. If this was the case this problem might not exist.
This is necessary because it was still possible for an entity update packet to be constructed, the thread to pause, a kill to be sent on another thread, and then the original thread to resume and send the update
This would result in an update being received after a kill, which results in undeletable ghost objects until the viewer is relogged
Extending the lock looks okay since its only taken by kill, update and reprioritize, and both kill and update do not take further locks
However, evidence suggests that there is still a kill/update race somewhere
* Removed StorageManager
* CONFIG CHANGE: There are no more database settings in OpenSim.ini. Check the config-include configuration files for region store and estate store database settings
These levels correspond to packets that one isn't usually interested in when debugging (e.g. regular outgoing SimStats packets)
This is equivalent to what we had a year ago before it was removed. It's extremely crude since it doesn't allow one to pick individual clients or packets. However, it can still be useful when debugging packet race conditions.
On Linden Lab clients and some derivatives, receiving an entity update after an entity deletion notice results in an undeleteable prim that disappears upon client relog.
This check was dropped in 0.7 for unknown reasons but renewed testing demonstrates that queued updates can still be present after a scene object has been deleted.
At least on stock Linden clients, updating the texture on all faces of the prim will actually send an ObjectImage packet for each update.
There is a race condition if these are handled async, meaning that occasionally not all of the faces are correctly updated.
DisableFacelights option to OpenSim.ini to finally kill those immersion-
breaking, silly vanity lights that destroy nighttime RP. Girls, you look
just fine without them. Guys, you too. Thank you. Melanie has left the building.
nonworking ownership assignment in SOG, which messed things up before.
No longer trust the client to send the ID of the person something is copied
as, since it allows to run a script with someone else's permissions. Properly
adjust inventory ownership and perms.
This is one step towards reducing hud glitches on region crossing, since the viewer fails to display prims if it receives child full updates before the root prim full update
This commit also introduces a mechanism in LLClientView to stop child attachment updates ever going out before the root one
This is a very temporary mechanism and will be commented out when the next step of the fix (to give root prims higher udpate priority) is committed
This code is a foreport from the equivalent changes in 0.6.9-post-fixes
* Simplified the interest management code to make it easier to add new policies. Prioritization and reprioritization share code paths now
* Improved the distance and front back policies to always give your avatar the highest priority
This implements the 'share with group' flag for notecards and scripts in prim inventory since the PermissionsModule checks group membership and permissions.
Other than that, the code in PermissionsModule duplicates the checks in LLClientView so there should be no change other than allowing group members to edit embedded notecards and scripts.
For all other asset types, the permission checking code in LLClientView continues to be used, pending refactoring of suitable permissions code
This means that 'share with group' will not yet work for prim inventory items other than notecards and scripts
* Adds GetTextureModule that implements the "GetTexture" capability, aka HTTP texture fetching. This is a significantly optimized path that does not require any server-side JPEG2000 decoding, texture priority queue, or UDP file transfer
* Sanity check for null reference in LLClientView.RefreshGroupMembership()
If serverside permissions are off then this works as expected. Previously, it was impossible for more than one person to edit such items even if permissions were off.
If serverside permissions are on then this works as expected if the object was created by an avatar who had the required group active.
However, if the group for the object is later set then the contained item is still not editable. This may be linked to a wider bug where the object is still not modifiable by the group anyway
If a full update is sent after the kill, the object remains as in the linden viewer but in an undeletable and unowned state until relog
This patch prevents this by recording kills in LLClientView
* Moved the SL asset type to content type conversion methods from ServerUtils to OpenSim.Framework.SLUtil
* Linked content type to asset type in AssetMetadata
Fixes: Undo, T-pose of others on login, modifiedBulletX works again, feet now stand on the ground instead of in the ground, adds checks to CombatModule. Adds: Redo, Land Undo, checks to agentUpdate (so one can not fall off of a region), more vehicle parts. Finishes almost all of LSL (1 function left, 2 events).
Direct flames and kudos to Revolution, please
Signed-off-by: Melanie <melanie@t-data.com>
These patch should allow people using systems that do not have their locale set to En_US or similar to use OpenSim without suffering effects such as being a million miles up in the air on login.
The problem was caused by parsing strings without forcing that parse to be En_US (hence different decimal and digit group symbols were causing problems).
Thanks very much to VikingErik for doing the legwork on this fix and phacelia for spotting it in the first place.
This resolves the problem where eyes and hair would turn white on standalone configurations
When a client receives body part information, for some insane reason or other it always ends up uploading this back to the server and then immediately re-requesting it.
This should have been okay since we stored that asset in cache. However, the standalone asset service connector was not checking this cache properly, so every time the client made the request for the asset it has just loaded it would get a big fat null back in the face, causing it to make clothes and hair white.
This bug did not affect grids since they use a different service connector.
* Handle logout properly. This needed an addition to IClientAPI, because of how the logout packet is currently being handled -- the agent is being removed from the scene before the different event handlers are executed, which is broken.
When an object was deleted, the remove script instance call was aggregating the scripting events as normal.
This would queue a full update of the prim before the viewer was notifed of the deletion of that prim (QuitPacket)
On some occasions, the QuitPacket would be sent before the full update was dequeued and sent.
In principle, you would think that a viewer would ignore updates for deleted prims. But it appears that in the Linden viewer (1.23.5),
a prim update that arrives after the prim was deleted instead makes the deleted prim persist in the viewer. Such prims have no properties
and cannot be removed from the viewer except by a relog.
This change stops the prim event aggregation call if it's being deleted anyway, hence removing the spurious viewer-confusing update.
This merge was very conflicted. I think I got them all, but I can't be sure.
I had to merge to master or risk divergence to the point of unmergeability.
* Make several packets not asynchronous (such as AgentUpdate). In theory, all fast returning packet handling methods should not be asynchronous. Ones that wait on an external resource or a long held lock, should be asynchronous.
* Marking CoarseLocationUpdate as *not* zerocoded. Zerocoding can only save space when a packet contains three or more contiguous zeroes, and will use more space if it contains single zeroes randomly scattered through the packet (which is what you see when you send a long list of UUIDs)
InventoryDescendents packet. Testing has shown that UDP inventory now
works flawlessly and, unlike CAPS inventory, doesn't download the entire
agent inventory on start. Neither does it incessantly re-request folder
NULL_KEY. Therefore, I have disabled CAPS inventory.
* Unified the way region handles are stored and used in ScenePresence
* Fixed camera position for child agents
* CheckForSignificantMovement now checks avatar and camera position (both are important for scene prioritization)
* Removing debug code from the previous commit
* Fixing a bug where the max burst rate for the state category was being set as unlimited, causing connections to child agents to saturate bandwidth
* Upped the example default drip rates to 1000 bytes/sec, the minimum granularity for the token buckets
* Added a new [Startup] config option called use_async_when_possible to signal how to run operations that could be either sync or async
* Changed Scene.ForEachClient to respect use_async_when_possible
* Fixing a potential deadlock in Parallel.ForEach by locking on a temporary object instead of the enumerator (which may be shared across multiple invocations on ForEach). Thank you diva
* Lock the LLUDPClient RTO math * Add a helper function for backing off the RTO, and follow the optional advice in RFC 2988 to clear existing SRTT and RTTVAR values during a backoff
* Removing the unused PrimitiveBaseShape.SculptImage parameter * Improved performance of SceneObjectPart instantiation * ZeroMesher now drops SculptData bytes like Meshmerizer, to allow the texture data to be GCed * Improved typecasting speed in MySQLLegacyRegionData.BuildShape()
* Improved the instantiation of PrimitiveBaseShape
* Avoid allocating an Action<IClientAPI> object every round of the OutgoingPacketHandler
* Removed unnecessary semi-colon endings from OpenSim.ini.example [InterestManagement] section
* Changed Util.FireAndForget() to use any of five different methods set with async_call_method in the [Startup] section of OpenSim.ini. Look at the example config for possible values
* Change the OnQueueEmpty firing to use a minimum time until next fire instead of a sleep
* Set OutgoingPacket.TickCount = 0 earlier to avoid extra resends when things are running slowly (inside a profiler, for example)
* Removed the unused PacketSent() function
* Switched UnackedPacketCollection from a SortedDictionary to a Dictionary now that the sorting is no longer needed. Big performance improvement for ResendUnacked()
* Changed HandleQueueEmpty()'s Monitor.TryEnter() calls to locks. We want to take our time in this function and do all the work necessary, since returning too fast will induce a sleep anyways
* Clamp retransmission timeout values between three and 10 seconds
* Log outgoing time for a packet right after it is sent instead of well before
* Loop through the entire UnackedPacketCollection when looking for expired packets
* Changed the timer tracking numbers for each client to not have "memory". It will no longer queue up calls to functions like ResendUnacked
* Reverted Jim's WaitHandle code. Although it was technically more correct, it exhibited the exact same behavior as the old code but spent more cycles. The 20ms has been replaced with the minimum amount of time before a token bucket could receive a drip, and an else { sleep(0); } was added to make sure the outgoing packet handler always yields at least a minimum amount
* Added TokenBucket.cs to OpenSim, with some fixes for setting a more accurate MaxBurst value and getting a more accurate Content value (by Drip()ing each get)
* Send terrain data in a spiral pattern instead of a typewriter pattern (placeholder until terrain data becomes part of the interest list management)
* Added a debug line when resent packets are being sent
* Replaced calls to ThreadPool.QueueUserWorkItem() with ThreadPool.UnsafeQueueUserWorkItem() since OpenSim does not use Code Access Security sandboxing
* Initialize the LLClientView prim full update queue to the number of prims in the scene for a big performance boost
* Reordered some comparisons on hot code paths for a minor speed boost
* Removed an unnecessary call to the expensive DateTime.Now function (if you *have* to get the current time as opposed to Environment.TickCount, always use DateTime.UtcNow)
* Don't fire the queue empty callback for the Resend category
* Run the outgoing packet handler thread loop for each client synchronously. It seems like more time was being spent doing the execution asynchronously, and it made deadlocks very difficult to track down
* Rewrote some expensive math in LandObject.cs
* Optimized EntityManager to only lock on operations that need locking, and use TryGetValue() where possible
* Only update the attachment database when an object is attached or detached
* Other small misc. performance improvements
* Changed the Send*Data structs in IClientAPI to use public readonly members instead of private members and getters
* Made Parallel.ProcessorCount public
* Started switching over packet building methods in LLClientView to use Util.StringToBytes[256/1024]() instead of Utils.StringToBytes()
* More cleanup of the ScenePresences vs. ClientManager nightmare
* ScenePresence.HandleAgentUpdate() will now time out and drop incoming AgentUpdate packets after three seconds. This fixes a deadlock on m_AgentUpdates that was blocking up the LLUDP server
* Handle the AgentFOV packet
* Bypass queuing and throttles for ping checks to make ping times more closely match network latency
* Only track reliable bytes in LLUDPCLient.BytesSinceLastACK
* Changed the throttling logic to obey the requested client bandwidth limit but also share bandwidth between some of the categories to improve throughput on high prim or heavily trafficked regions
Replaced the update lists with a priority queue implementation in LLClientView.
The priority queues are based on the MinHeap implementation also included in
this commit within the OpneSim.Framework namespace. Initially setup to exactly
mimic the behavior beofre the change which was a first come first serve queue.
* Removed two redundant parameters from SceneObjectPart
* Changed some code in terse update sending that was meant to work with references to work with value types (since Vector3 and Quaternion are structs)
* Committing a preview of a new method for sending object updates efficiently (all commented out for now)
* Apparently the LLClientView should have been doing this previously.. Also fixed the 'You' on the index block.. so the client doesn't display an extra green dot.
* Thanks lkalif for bringing it to our attention.
Justin Clark-Casey (justincc)