During production of a Thing, decisions have to be made whether the following parameters should be pre-configured, manually entered after installation or automatically found and/or created by the device if possible (zero-configuration networking):

• Address and domain of XMPP Server.
• JID of the Thing.
• JID of Thing Registry, if separate from the XMPP Server.
• JID of first Provisioning Server, if separate from Thing Registry or XMPP Server.

A decision has to be made at this point if global/manufacturer/customer servers should be used, or if local resources should be searched for and used if found. The first option is easy to configure in a production environment and might have commercial significance, but cannot use local resources where available. The second leaves much responsibility to the Thing for finding local resources, but has the advantage of allowing for a more decentralized network architecture. A detailed discussion of the two alternatives goes beyond the scope of this specification, and will not be presented here.

Apart from physical installation, and connection to power and communication infrastructure, the installation phase of a Thing might also require manual entry of values that could not be set in the production environment. Since Things might have very limited human user interfaces, external tools might be required to provide this information. Due to its complexity, any manual entry of configuration parameters should be avoided, if possible. However, manual entry of some parameters might allow for Things to use local resources where such cannot be found nor set in a production environment.

If the address of an XMPP Server is not preconfigured, the Thing must attempt to find one in its local surroundings. This can be done using one of several methods:

• DHCP
• Multicast DNS
• SSDP/UPnP

The following sections describe them in more detail.

Note: If server-less messaging is to be used, as described in &xep0174; this step can be used to find the Thing Registry and optionally the Provisioning Server and other peers it want to connect to. The next section can thus be skipped.

Once an XMPP Server has been found, a connection can be made. If multiple XMPP Servers are found, the client is free to choose the one that best suits its purposes.

If the Thing does not have an account already, an account can be registered along what is specified in &xep0077;. If multiple servers are available, the first XMPP server that allows account creation can be used.

If a Thing Registry is not preconfigured, one must be found. A Thing Registry can be hosted either as a server component using &xep0114; or as an XMPP Client accessible through a JID. The following lists methods to obtaining the Component Address or JID for the Thing Registry. Note that the last one has security considerations that need to be taken into account, if implemented.

1. Preconfigured Component Address of Thing Registry. A Component address is normally a subdomain to the domain of the XMPP Server that hosts the component.
2. Preconfigured bare JID of Thing Registry.
3. Preconfigured subdomain part of Component Address. This will be added to the domain of the XMPP Server used to connet to.
4. Preconfigured user name of JID. This will be added to the domain of the XMPP Server used to connected to.
5. Searching through Server Components on the XMPP Server currently connected to, as described in Determining Support.

Once a Thing Registry has been found and been befriended, the Thing can register itself with the registry, as follows:

There are two types of tags: Tags with string values and tags with numerical values. The distinction is important, since the type of value affects how comparisons are made later when performing searches in the registry.

The Thing should only register parameters required to be known by the owner of the Thing. Dynamic meta information must be avoided at this point. To claim the ownership of the Thing, the owner needs to present the same meta information as registered by the Thing. Before an owner has claimed ownership of the Thing, it will not be returned in any search results. A list of predefined meta tag names can be found in the Meta Tags section below.

The Thing can register itself as many times as it wants, and the response is always empty. Only one record per resource-less JID must be created. A new registration overrides any previous information, including meta tags previously reported but not available in the new registration. Once a Thing has been claimed by an owner, it should not register itself again, unless it is reset and the installation process restarted.

If the Thing tries to register itself even though the Thing has already been claimed in the registry, the registry must not update any meta data in the registry, and instead respond with the following response. When the thing receives this, it can safely extract the JID of the owner and switch its internal state to claimed.

Note: Meta Tag names are case insensitive. In this document, all tag names have been written using upper case letters.

If a thing is self-owned, it can register itself with the Registry as normal, with the addition of setting the attribute selfOwned to true, as is shown below. This registers the Thing directly as PUBLIC CLAIMED, with no need for an owner to claim ownership of the device. This can be useful if installing Things that should be publically available.

A Thing might reside behind a gateway or concentrator and might not be directly connected to the XMPP network itself, as is described in &xep0326;. In these cases, there are three optional attributes that can be used to identify the Thing behind the JID: The nodeId attribute gives the ID of the Thing (a.k.a. "Node"). The Node might reside in specific Data Source (large systems might have multiple sources of nodes). In this case, the data source is specified in the sourceId attribute. Normally, the Node ID is considered to be unique within the concentrator. If multiple data sources are available, the Node ID is unique within the data source. However, a third attribute allows the uniqueness to be restricted to a given cacheType. Finally, it is the triple (nodeId, sourceId, cacheType) which guarantees uniqueness within the concentrator.

For a Thing controlled by a concentrator to register itself in the Thing Registry, it simply adds the optional attributes nodeId, sourceId and cacheType as appropriate to the registration request, as follows:

If the Thing behind the concentrator is self-owned, it simply adds the selfOwned attribute to the request and sets it to true.

As mentioned above, the owner of the Thing must provide the information provided by the Thing to the Registry, in order to claim ownership over it. To avoid the possibility that somebody can guess the information, the information must necessarily be long. This creates the problem of transfer of information. One method to solve this is through the use of QR-codes. Such codes can be either printed on a sticker and put on the Thing itself, its wrapping, or displayed on its display when not claimed. This QR-code can then be photographed by a smart phone or tablet, decoded and the information retrieved can be used in the ownership claim call.

If QR-codes are used to transfer Thing meta data for ownership claims, it must be generated as follows: To the string "IoTDisco" is appended all meta tags in order. Each tag name is prefixed by a semi-colon (;), and if the tag is numeric, the tag is prefixed by an additional hash sign (#). Each tag value is prefixed by a colon (:). If the meta value contains semi-colons or back-slashes, each one is prefixed by a back-slash. When decoding the string, this allows the decoder to correctly differ between tag delimiters and characters belonging to tag values. A tag name must never contain colon, hash sign or white space characters.

The above meta data would therefore generate the string:

Using UTF-8 encoding when generating the QR-code, this string returns the following QR-code:

Once the client has the required meta information about the Thing to claim ownership, it sends itself the following request to the Thing Registry:

If this claim is successful, the Thing is marked as a public claimed Thing. The thing can always be removed later, but after the claim, the Thing is public. If you want to claim a private Thing, you can add the public attribute with value false to the claim, as follows:

In this case, if the claim is successful, the Thing will not be made public in the Thing Registry, after the claim.

If a claim is successful, i.e. there's a Thing that has not been claimed with EXACTLY the same meta data (however, the order is not important), the Thing is marked in the Registry as CLAIMED, and as public or private depending on the public attribute, and an empty result is returned as follows. If there's a claimed Thing with exactly the same meta data, and the JID of the claimant (without resource) matches the JID of the claimer (without resource), a success response is also returned, containing the resource-less JID of the Thing, as follows:

If the Thing that has been claimed resides behind a concentrator, the result will contain those of the attributes nodeId, sourceId and cacheType that are required to access the Thing in calls made using &xep0323;, &xep0324;, &xep0325; and &xep0326;. The following example illustrates a response where a Thing behind a Concentrator has been claimed:

If, on the other hand, no such Thing was found, or if such a Thing was found, but it is already claimed by somebody else, a failure response is returned. This response should avoid to inform the client in detail why the claim failed, as follows:

When the Thing has been successfully claimed, the Registry sends information about this to the Thing, to inform it that it has been claimed and the resource-less JID of owner. After receiving this information, it doesn't need to register itself with the Registry anymore.

If the Thing was claimed as a private Thing, this is shown using the public attribute in the response, as follows:

If the public attribute is present and has value false, it means no further meta data updates are necessary, since the device is not searchable through the Thing Registry.

If the Thing resides behind a concentrator, the request must contain those of the attributes nodeId, sourceId and cacheType that are required to access the Thing, as follows:

The Thing simply returns an empty response to acknowledge the receipt of the information, as follows:

After receiving this, the thing knows it can accept friendship requests from the corresponding owner. It can also safely send a friendship request to the owner.

Note: Meta Tag names are case insensitive. In this document, all tag names have been written using upper case letters.

After a Thing has been claimed and is registed as a PUBLIC CLAIMED Thing in the Registry, it implies the Thing is available in searches. The owner can choose to remove the Thing from the Registry, to avoid that the Thing appears in searches. To remove a Thing from the Registry the owner simply sends a removal request to the Registry with the resource-less JID of the Thing to remove, as follows:

If the Thing resides behind a concentrator, the request must contain those of the attributes nodeId, sourceId and cacheType that are required to access the Thing, as follows:

If such a Thing is found and is owned by the caller, it is removed from the Registry, and an empty response is returned, to acknowledge the removal of the Thing from the registry, as follows:

However, if such a thing is not found, or if the thing is owned by another, an item-not-found error is returned, as follows:

After successfully removing a Thing from the Registry, and if the Thing is friend to the Registry, the Registry informs the Thing it has been removed from the Registry. It does this, so the Thing can remove the friendship and stop any meta data updates to the Registry.

If the Thing lies behind a concentrator, the removal request would look as follows:

The Thing acknowledges the removal request by returning an empty response, as follows:

Up to this point only basic configuration and ownership and visibility of a Thing has been covered. For more advanced operations, a Thing might be required to use a Provisioning Server to whom it can delegate trust and allow making decisions, controlling access rights and privileges for the Thing, as described in &xep0324;. If a Provisioning Server is not preconfigured, one must be found. The following lists methods to obtaining the JID for the Provisioning Server.

1. Preconfigured Component Address of Provisioning Server. A Component address is normally a subdomain to the domain of the XMPP Server that hosts the component.
2. Preconfigured bare JID of Provisioning Server.
3. Preconfigured subdomain part of Component Address. This will be added to the domain of the XMPP Server used to connet to.
4. Preconfigured user name of JID. This will be added to the domain of the XMPP Server used to connected to.
5. The Thing Registry itself can be a Provisioning Server. This can be found out by sending a discovery request to the Thing Registry, as described in Determining Support.
6. The Owner itself can be a Provisioning Server. This can be found out by sending a discovery request to the Owner, as described in Determining Support.
7. Searching through Server Components on the XMPP Server currently connected to, as described in Determining Support.

Once a Provisioning Server has been found and been befriended, the Thing can delegate its trust to it, according to &xep0324;.

Once a Thing has been claimed and chooses to reside as a public Thing in the registry, it can update its meta information at any time. This meta information will be available in searches made to the registry by third parties and is considered public. However, the Thing should be connected to a provisioning server at this point, so that correct decisions can be made regarding to friendship, readout and control requests made by parties other than the owner.

Meta information updated in this way will only overwrite tags provided in the request, and leave other tags previously reported as is. To remove a string-valued tag, it should be updated with an empty value. It is also recommended that key meta information required to claim ownership of the Thing after a factory reset is either removed, truncated or otherwise modified after it has been claimed so that third parties with physical access to a public Thing cannot hijack it by searching for it, extracting its meta information from the registry, then resetting it and then claiming ownership of it.

To update meta data about itself, a Thing simply sends a request to the Thing Registry, as follows:

If the Thing resides behind a concentrator, the request must contain those of the attributes nodeId, sourceId and cacheType that are required to access the Thing, as follows:

If the Thing is found in the registry and it is claimed, the registry simply acknowledges the update as follows:

However, if the Thing is not found in the registry, probably because the owner has removed it from the registry, an error response is returned. When receiving such a response, the Thing should assume it is the owner who has removed it from the registry, and that further meta data updates are not desired. The Thing can then unfriend the registry and stop further meta data updates. The error response from the registry would look as follows:

If the Thing on the other hand is found in the Registry, but is not claimed, the registry must not update any meta data in the registry, and instead respond with the following response. When the thing receives this, the Thing can assume it has been disowned, and perform a new registration in the Registry so that it can be re-claimed.

Note: Meta Tag names are case insensitive. In this document, all tag names have been written using upper case letters.

An owner of a thing can also update the meta-data of a thing it has claimed. To do this, you simply add a jid attribute containing the JID of the thing to the update element. (If this attribute is not present, the JID is assumed to be that of the sender of the message.)

The owner can update meta-data of things behind concentrators also. To do this, the corresponding attributes nodeId, sourceId and cacheType must be used to identify the thing, as follows:

If the Thing is found in the registry and it is claimed by the sender of the current message (i.e. owner is the sender), the registry simply acknowledges the update as follows:

But if the owner is not the sender of the current message (i.e. owner is somebody else), or if the thing is not found at all, the server must report the node as not existing (i.e. not existing among the set of things claimed by the owner).

It is possible for anyone with access to the Thing Registry to search for public Things that have been claimed, including self-owned Things. Such searches will never return things that have not been claimed or have been removed from the registry.

A search is performed by providing one or more comparison operators in a search request to the registry. If more than one comparison operator is provided, the search is assumed to be performed on the intersection (i.e. AND) of all operators. If the union (i.e. OR) of different conditions is desired, multiple consecutive searches have to be performed.

The following table lists available search operators, their element names and meanings:

The following example shows how a search for specific devices within a specific geographic area can be found. More precisely, it searches for a certain kind of PLC produced by a certain manufacturer, but only versions 1.0 <= v < 2.0 and with serial numbers beginning with 39487. The PLCs must also lie within latitude 33 ad 34 degrees south and between longitude 70 and 72 west.

The offset attribute tells the registry the number of responses to skip before returning found things. It provides a mechanism to page result sets that are too large to return in one response. the maxCount attribute contains the desired maximum number of things to return in the response. The registry can lower this value, if it decides the requested maximum number is too large.

If tag names are not found corresponding to the names provided in the search, the result set will always be empty. There's a reserved tag named KEY that can be used to provide information shared only between things and their owners. If a search contains an operator referencing this tag name, the result set must also always be empty. Searches on KEY MUST never find things. Furthermore, search results must never return KEY tags.

The registry returns any things found in a response similar to the following:

If a Thing resides behind a concentrator, the response must contain those of the attributes nodeId, sourceId and cacheType that are required to access the Thing, as follows:

If more results are available in the search (accessible by using the offset attribute in a new search), the more attribute is present with value true.

Note: Meta Tag names are case insensitive. In this document, all tag names have been written using upper case letters.

A thing can unregister itself from the Registry. This can be done in an uninstallation procedure for instance. To unregister from the registry, it simply sends an un-registration request to the registry as follows.

If the Thing resides behind a concentrator, the request must contain those of the attributes nodeId, sourceId and cacheType that are required to access the Thing, as follows:

The registry always returns an empty response, simply to acknowledge the receipt of the request.

The owner of a Thing can disown the Thing, returning it to a state without owner. This is done by sending the following request to the Thing Registry:

If the Thing resides behind a concentrator, the request must contain those of the attributes nodeId, sourceId and cacheType that are required to access the Thing, as follows:

If such a Thing is not found, or if the thing is not owned by the caller, an item-not-found error is returned, as follows:

If such a Thing is found, and it is owned by the caller, but not online as a friend, the Thing cannot be disowned, since it would put the Thing in a state from which it cannot be re-claimed. Therefore, the Thing Registry must respond in the following manner:

Before returning a response to the caller, the Thing Registry informs the Thing it has been disowned. It does this, so the Thing can remove the friendship to the owner, and perform a new registration.

If the Thing lies behind a concentrator, the disowned request would look as follows:

The Thing acknowledges that it has been disowned by returning an empty response, as follows:

When receiving the acknowledgement from the Thing, the Thing is set as an unclaimed Thing in the Registry. Furthermore, all tags corresponding to the Thing are removed from the registry, and a random KEY tag is added of sufficient complexity to make sure other clients cannot claim the Thing by guessing. Finally, an empty response is returned, to acknowledge that the Thing has been disowned, as follows:

If for any reason, the Thing does not acknowledge the disowned request, or an error occurs, the Registry returns the same error as if the Thing would have been offline.

A client must treat the connection between a Thing Registry differently if it is hosted as a client, having a JID, or if it is hosted as a Jabber Server Component. If it is hosted as a server component, there's no need for the thing to become friends with the Thing Registry. Messages and requests can be made directly to the server component without having to add it to the roster or request presence subscriptions. If the Thing Registry is hosted as a client, having a JID (@ in the address), the Thing Registry must be added to the roster of the client before the client can communicate with the Thing Registry.

This document does not limit the number or names of tags used by Things to register meta information about themselves. However, it provides some general limits and defines the meaning of a few tags that must have the meanings specified herein.

The maximum length of a tag name is 32 characters. Tag names must not include colon (:), hash sign (#) or white space characters. String tag values must not exceed 128 characters in length.

The following table lists predefined tag names and their corresponding meanings.

It is up to the Thing Registry to choose which tags it persists and which tags it doesn't. To avoid the possibility of malicious reporting of tags, some limit should be imposed on what tags are supported. As a minimum, a Thing Registry must support all predefined tags, as listed above.

Note: Meta Tag names are case insensitive. In this document, all tag names have been written using upper case letters.

In the case the Thing Registry is not the XMPP Server to which the Thing is connected, a friendship relationship between the Thing and the Thing Registry needs to be handled. To minimize the number of concurrent friends the Thing Registry needs to maintain, a Thing must only maintain an active friendship with the registry if it needs to communicate with the registry. This means that unless updating meta data frequently, the Thing must unfriend the Registry when done with its communication. If only updating meta data intermittently, the friendship can be reestablished when needed, and removed when done.

The &xep0114; provides an elegant way to introduce external services as server components using a third port into the server (the first two being the client-to-server port and the server-to-server port). But since XEP-0114 is historical, meaning it is not guaranteed to conform to v1.0 of the XMPP specification, it has some serious security issues:

1. It lacks SSL/TLS support, or the starttls element to switch to TLS after connecting. This makes it possible to sniff traffic in this port.
2. It lacks SASL authentication. Instead a simple handshake is performed
3. There is no way to actually verify that the server is the server. This makes it possible to create a simple Man-in-the-middle attack.

For these reasons, it is not recommended that a Thing Registry service, publishing itself as a Jabber Server Component, does so from outside of the network. Instead, the Thing Registry should be installed on the same server or on a server in the same local area network, so that the Jabber Component protocol port is closed to the Internet.

Since it is not guaranteed that an XMPP Server operator allows installation of third party products (such as a Thing Registry), the option to host a Thing Registry using a normal JID is still available. It can be used in proof of concepts, etc. For scalability issues it is recommended that the Thing Registry be hosted as a Jabber Server Component when the population of Things grows.

If using predefined user names when searching for a Thing Registry or Provisioning Server, care must be taken to which XMPP Server things connect. It might be possible for third parties to register these predefined account names, and pretend to be a Thing Registry or Provisioning Server and in this way hijack unsuspecting Things. If installing things using this method of finding a Thing Registry or Provisioning Server, these accounts must be registered beforehand, to make sure the things cannot be hijacked.

The combination of visible key meta information (perhaps in a visible QR-code) and a factory default reset button on a Thing, opens up the possibility to hijack the Thing. To avoid this, at least one of the two should be removed after successful installation. Either the key meta information (QR-code) should be placed on the package or separate paper and not on the thing itself, or the factory default reset button should be sealed or hidden and only accessible by licensed maintenance personell. If using an electronic means to present the key meta information (for instance by displayed a QR-code on a display on the thing), care should be taken so that the information cannot be displayed without breaking a seal, or other means to protect the Thing.

Regardless the above security measures, a Thing can be hijacked by a third party in the time window between successful installation of the device and until the correct owner has claimed ownership of the device. Minimizing this time window, and using a shared secret (KEY tag) between the Thing and its owner, decreases the possibility of getting the thing hijacked.

Care should be taken what key meta information is used to accept an ownership claim. After a successful claim, this meta information is still available in the registry, at least until the Thing is removed from the registry. While public in the registry, the meta information can be searched and presented to third parties. Access to this information can help third parties to hijack Things, if they can reset them to factory default settings.

To avoid this, the Thing can do three things after a successful ownership claim:

• Including a KEY tag in the key meta information. The KEY tag is not searchable nor presented in search results.
• Remove, truncate or change some key meta information after a successful ownership claim. Partial information is not sufficient for a successful ownership claim.
• Remove the Thing from the registry.

The KEY tag is unique in that it is not searchable nor available is search results. For this reason it is ideal for providing secrets shared only between the Thing and the owner. By providing a sufficiently long KEY value in the key meta information required to claim the Thing, guessing the information even though the other meta information is available, will be sufficiently hard to make it practically impossible.

Even though the KEY tag is not searchable or available in search results, it should be emptied by the Thing after a successful claim, just to make sure the key cannot be learned by looking into the database of the registry, or by some other means.

This document does not limit tag names or the number of tags that can be used by Things. This opens up the possibility of tag spam. Malicious things could fill the database of the registry by reporting random tag names until the database is full.

To prevent such malicious attacks, the registry could limit the tags it allows to be stored in the database. The registry must however allow the storage of the predefined tag names defined in this document. If it has a configurable list of approved tags that can be stored, or if it allows any tags is an implementation decision.

If using external services when creating QR-codes, like the Google Charts API used in this document, make sure HTTPS is used and certificates validated. If HTTP is used, meta-data tags used in Thing Registry registrations can be found out by sniffing the network, making it possible to hijack the corresponding devices.

TBD

TBD

TBD