[Opensim-dev] secure_inventory_server ??
Michael Wright
michaelwri22 at yahoo.co.uk
Sat Jul 26 10:56:42 UTC 2008
Sorry one more email,
Just noticed your latest reply on http://opensimulator.org/mantis/view.php?id=1828. And your suggestion on there, seems a good step forward. Can see some weakness in the public key scheme. But it at least shouldn't break anything.
I guess a lot of the problems do come from the language barrier, as that suggestion is very similar to one approach I have been suggesting in this thread. Which you kept rejecting.
Michael Wright <michaelwri22 at yahoo.co.uk> wrote:
liu xiaolu <lulurun at gmail.com> wrote:
These my final points:
1. Objectively, once more, inventoryserver contains no information that can make sure if a request
is or not from a valid user. only userserver contains that information, and we all do not want
such kind of "identity" information to be copied everywhere.
2. according to 1, no matter push or pull, inventoryserver has to validate the request or anyone can
use it. => inventoryserver has to "pull" or be "pushed" authed information from 1 or many
userservers.
3. that doesn't mean inventoryserver has to rely on some userservers, inventoryserver just rely on
if the user want it to validate and how the user want it to validate requests. as I said,
inventoryserver exists for its user, not userserver, why inventoryserver have to "pull" or be
"pushed" something is because "users" *want* do that to save their inventory. inventoryserver
works for users, it does not always trust 1 or some userservers, it trusts what the user specified.
It acts as what the user want it to do.
Your approach isn't creating inventory for users. Its creating inventory for a single grid and the users on that grid.
Y. "push" has security issue that any remote peer can "push" to inventoryserver. to set access
permissions on inventoryserver side or make a "safe peers" list ?
I think the "permission" should not be "user" based, because it is unsafe either, then if the
"permission" is "url" based, doesn't that mean the same as inventoryserver rely on the
"safe peers" ?
Again if you have a authentication scheme in place so that only "trusted" user servers could push the sessionID. Then its no different.
Currently the user server can do things like create new inventory set and request all the folders for a user. And the only security is checking that the reported ip address in the request matches the one stored for the user server.
But having said that, if we could at least get a pull system that works and doesn't break use cases then thats great. But at the moment the current system does break perfectly valid use cases.
Really, I think "push" is much more dangerous, and "pull" does not have to be designed "rigid".
if you can take time to write some of the usecases that should be support, but current "pull"
inventory breaks them, I would like to do what ever I can do to think and implement.
I just don't understand why you haven't read anything I have said or if you have read it. Decided to ignore it.
I have repeatedly descriped valid use cases that we should support.
liu xiaolu <lulurun at gmail.com> wrote: After all, please understand that I am not "selling" how my "secure inventory" is great,
I made it because I found my interop module is an ease-to-use tool to modify others' inventory(at OSGrid.org).
if there are bugs or broken things please let me know.
hope it is fixed or I would like to offer what I can do to help the fix.
What I am trying to do is to share my point of view on "interop", that is what I called avatar portability.
maybe I replied to a wrong mail.
So, about the usecase: (I finanlly got your idea, from the quoted mail)
>The use case (or one of many) that I'm talking about, is if a user has inventory on a inventory server, that user's inventory shouldn't be tied to any user server/grid. So the user should be able to use the same inventory set on multiple grids. At the
>moment that means they have the same userid on each grid but thats no problem (and in the future I hope we will support a better way of identifying a inventory belongs to a user).
>
>So lets call the Inventory server InvN like you used. My single inventory set (of folders/items) on that server is called InvSetA. Then we have Grid A and Grid B. With user server A and B. When I log into grid A it should use InvN to receive InvSetA.
>But also if I log out of grid A and log into grid B then it should also use InvN to receive InvSetA. The inventory set (of folders/items) isn't tied to a single grid/avatar. It belongs to the user. The "grid" is just using a guid (avatarID) to access the correct
>inventory set. So as long as both grids use the same id, they should be able to use the same inventory.
When I read the above sentences, what I was thinking is "why it is required to be like such a case?"
Still, There are 3 situations I can imagine:
*1*. UserA and UserB do not know each other, by (small) chance, thier UUID are the same.
uuid collision, hard to say anything.
yes, we do should care about this, though I learned that uuid wouldnot duplicate in common life, but we cannot trust the implementations.
I think this is possibly not what you mean, even if you really mean this ...
the "userserver_dictionary" should be "uuid at grid_identifier", "userserver_url", grid_identifier can be the address(host,ip) of the grid service, hmmm ...
*2*. UserA and UserB are the same person.
In order to login to both GridA and GridB, an User copied his/her account both at GridA(UserA) and GridB(UserB), so they are having the same uuid.
I suggest to use the way I mentioned: he/she can always keep one account at GridA, and travel to both GridA and GridB.
reasons for the suggestion:
1. coping account is not a forever solution for grid-grid travel.
2. there is an advanced way, we should encourage ourself to use that
*3*. UserA and UserB are completly different accounts, they are friends/groupmember(or they are owned by the same person who intend to do such thing),
they want to share InvSetA(both of them can CRUD InvSetA)
InvSetA only can be specified by user uuid, so they *have* to have the same uuid to share InvSetA,
this can be solved in another level - inventory does not belongs to a user, it belongs to a group.
* a group has 1 uuid used to identify inventory
* a group contains mutilple/single user
this may introduce "join" SQL statements to "select" the inventory, instead of just "select" by "uuid". ... no deeper thoughts.
I want it be a "pulling" system, because
* "push" has security issues, session_id or what ever things that is used as a secret to prove identity, shouldn't be "pushed" to everywhere
* I still do not think the case you are concerning is a major case, or even I think it is strange, except you are meaning *3*. but *3* is another issue.
* "pulling" reduces the responsibility of inventoryserver, in the case, inventoryserver couldn't get auth infomation for the specified userserver, because userserver
is specified by user itself, so it the problem of the user, not the inventory, but "push" is the opposite, to think inventory/asset server will play a important roll
in the future, the less responssibility it has the more it is reliable.
Wrote too much english today ... see you tomorrow
thank you for your detailed explanations.
regards,
lulurun
2008/7/26 Michael Wright <michaelwri22 at yahoo.co.uk>:
I might be missing something but I don't see how this solution addresses some of the problems. Maybe I wasn't very clear on some of the use cases that would currently be broke.
The use case (or one of many) that I'm talking about, is if a user has inventory on a inventory server, that user's inventory shouldn't be tied to any user server/grid. So the user should be able to use the same inventory set on multiple grids. At the moment that means they have the same userid on each grid but thats no problem (and in the future I hope we will support a better way of identifying a inventory belongs to a user).
So lets call the Inventory server InvN like you used. My single inventory set (of folders/items) on that server is called InvSetA. Then we have Grid A and Grid B. With user server A and B. When I log into grid A it should use InvN to receive InvSetA. But also if I log out of grid A and log into grid B then it should also use InvN to receive InvSetA. The inventory set (of folders/items) isn't tied to a single grid/avatar. It belongs to the user. The "grid" is just using a guid (avatarID) to access the correct inventory set. So as long as both grids use the same id, they should be able to use the same inventory.
So in this case if we have the pull from user server approach to authenticating then it needs to switch between user server A and B, depending on what grid the user is currently logged into. So again it would require a push somewhere to update which is the current user server for that user.
As far as I can tell, your solution just stops a inventory server being tied to a single grid/user server and instead ties each inventory set to a grid/user server. That is better than tying the whole server. But it doesn't support the above case.
There are a number of other use cases that would require multiple grids/instances to access a single inventory set.
I find it helps if we think of User server and the grid server as seperate to the inventory and asset server. The user server defines what users can access a grid (plus other data like what inventory server that user is using). The grid server defines and manages what regions are on a grid.
But asset and inventory server can (in part) be thought of more as end user servers. In that a user could use the same servers on multiple grids and that can mean using the same inventory on all grids. So then if we think about defining on the asset and inventory servers what grids/regions can access them.
liu xiaolu <lulurun at gmail.com> wrote:
First, I want to clarify that, I am not against what you said
To release inventory, asset server become independent ones is also what I have been working on.
I am just trying to introduce my thoughts:
>>[solution:]
>>* add a new table for inventoryserver, 2 fields, useruuid, userserver_url, everytime inventoryserver
>> extract "session_id", "user_id" from the request, get "userserver_url" by "user_id", then check the
>> identity of "user_id" from "userserver_url" (call check_auth_session)
>I really think rather than the inventory server calling a authenticate method on the user server. That
>the user server should send a sessionid to the inventory server when a new login happens. I guess
>your idea of the userserver_url was something silmilar in that the userserver would update it when
>a user logs in? Otherwise if it was just "hardcoded" in the db then it wouldn't solve anything.
sorry for the lack of explanation of the new table, temporary let me call it "userserver_dictionary".
* "userserver_dictionary" is a table always together with inventoryfolers, inventoryitems
* "userserver_dictionary" has 2 column: user_id pkey, userserver_url
// user registration
when a new user registered at a GridService G1, the user will possibly have its account on G1's
userserver U1, then, no matter which inventoryserver the user is using(or going to use), assume it is called InvN,
InvN's "userserver_dictionary" would be added 1 record <"uid", U1's url>,
// user login to VW
* user get authentication at U1, get "session_id"
* user login into a regionserver, regionserver gets user's profile and extract user's inventoryserver InvN from its profile.
* regionserver requests getInventory("session_id", user.uuid) from InvN.
* InvN gets "U1's url" by user.uuid from "userserver_dictionary"
* InvN requests check_auth_session("session_id", user.uuid) from "U1's url"
//
in this case, dose InvN satisfies what you want ?
* InvN is separate from any other server
* InvN is independent, not rely on any userserver, it pull identity under user's specification(get "userserver_url" by uuid)
* InvN can store users from different gridsevices
--
Lulurun
_______________________________________________
Opensim-dev mailing list
Opensim-dev at lists.berlios.de
https://lists.berlios.de/mailman/listinfo/opensim-dev
---------------------------------
Not happy with your email address?
Get the one you really want - millions of new email addresses available now at Yahoo!
_______________________________________________
Opensim-dev mailing list
Opensim-dev at lists.berlios.de
https://lists.berlios.de/mailman/listinfo/opensim-dev
--
Lulurun
_______________________________________________
Opensim-dev mailing list
Opensim-dev at lists.berlios.de
https://lists.berlios.de/mailman/listinfo/opensim-dev
---------------------------------
Not happy with your email address?
Get the one you really want - millions of new email addresses available now at Yahoo!
_______________________________________________
Opensim-dev mailing list
Opensim-dev at lists.berlios.de
https://lists.berlios.de/mailman/listinfo/opensim-dev
--
Lulurun
_______________________________________________
Opensim-dev mailing list
Opensim-dev at lists.berlios.de
https://lists.berlios.de/mailman/listinfo/opensim-dev
---------------------------------
Not happy with your email address?
Get the one you really want - millions of new email addresses available now at Yahoo!_______________________________________________
Opensim-dev mailing list
Opensim-dev at lists.berlios.de
https://lists.berlios.de/mailman/listinfo/opensim-dev
---------------------------------
Not happy with your email address?
Get the one you really want - millions of new email addresses available now at Yahoo!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://opensimulator.org/pipermail/opensim-dev/attachments/20080726/a9917c61/attachment-0001.html>
More information about the Opensim-dev
mailing list