File

mod_delegation/mod_delegation.lua @ 6301:fa45ae704b79

mod_cloud_notify: Update Readme diff --git a/mod_cloud_notify/README.md b/mod_cloud_notify/README.md --- a/mod_cloud_notify/README.md +++ b/mod_cloud_notify/README.md @@ -1,109 +1,106 @@ ---- -labels: -- 'Stage-Beta' -summary: 'XEP-0357: Cloud push notifications' ---- +# Introduction -Introduction -============ +This module enables support for sending "push notifications" to clients +that need it, typically those running on certain mobile devices. -This module enables support for sending "push notifications" to clients that -need it, typically those running on certain mobile devices. +As well as this module, your client must support push notifications (the +apps that need it generally do, of course) and the app developer's push +gateway must be reachable from your Prosody server (this happens over a +normal XMPP server-to-server 's2s' connection). -As well as this module, your client must support push notifications (the apps -that need it generally do, of course) and the app developer's push gateway -must be reachable from your Prosody server (this happens over a normal XMPP -server-to-server 's2s' connection). - -Details -======= +# Details Some platforms, notably Apple's iOS and many versions of Android, impose -limits that prevent applications from running or accessing the network in the -background. This makes it difficult or impossible for an XMPP application to -remain reliably connected to a server to receive messages. - -In order for messaging and other apps to receive notifications, the OS vendors -run proprietary servers that their OS maintains a permanent connection to in -the background. Then they provide APIs to application developers that allow -sending notifications to specific devices via those servers. +limits that prevent applications from running or accessing the network +in the background. This makes it difficult or impossible for an XMPP +application to remain reliably connected to a server to receive +messages. -When you connect to your server with an app that requires push notifications, -it will use this module to set up a "push registration". When you receive -a message but your device is not connected to the server, this module will -generate a notification and send it to the push gateway operated by your -application's developers). Their gateway will then connect to your device's -OS vendor and ask them to forward the notification to your device. When your -device receives the notification, it will display it or wake up the app so it -can connect to XMPP and receive any pending messages. +In order for messaging and other apps to receive notifications, the OS +vendors run proprietary servers that their OS maintains a permanent +connection to in the background. Then they provide APIs to application +developers that allow sending notifications to specific devices via +those servers. -This protocol is described for developers in [XEP-0357: Push Notifications]. +When you connect to your server with an app that requires push +notifications, it will use this module to set up a "push registration". +When you receive a message but your device is not connected to the +server, this module will generate a notification and send it to the push +gateway operated by your application's developers). Their gateway will +then connect to your device's OS vendor and ask them to forward the +notification to your device. When your device receives the notification, +it will display it or wake up the app so it can connect to XMPP and +receive any pending messages. -For this module to work reliably, you must have [mod_smacks], [mod_mam] and -[mod_carbons] also enabled on your server. +This protocol is described for developers in \[XEP-0357: Push +Notifications\]. + +For this module to work reliably, you must have \[mod_smacks\], +\[mod_mam\] and \[mod_carbons\] also enabled on your server. -Some clients, notably Siskin and Snikket iOS need some additional extensions -that are not currently defined in a standard XEP. To support these clients, -see [mod_cloud_notify_extensions]. +Some clients, notably Siskin and Snikket iOS need some additional +extensions that are not currently defined in a standard XEP. To support +these clients, see \[mod_cloud_notify_extensions\]. -Configuration -============= +# Configuration - Option Default Description - ------------------------------------ ----------------- ------------------------------------------------------------------------------------------------------------------- - `push_notification_important_body` `New Message!` The body text to use when the stanza is important (see above), no message body is sent if this is empty - `push_max_errors` `16` How much persistent push errors are tolerated before notifications for the identifier in question are disabled - `push_max_devices` `5` The number of allowed devices per user (the oldest devices are automatically removed if this threshold is reached) - `push_max_hibernation_timeout` `259200` (72h) Number of seconds to extend the smacks timeout if no push was triggered yet (default: 72 hours) - `push_notification_with_body` (\*) `false` Whether or not to send the real message body to remote pubsub node. Without end-to-end encryption, enabling this may expose your message contents to your client developers and OS vendor. Not recommended. - `push_notification_with_sender` (\*) `false` Whether or not to send the real message sender to remote pubsub node. Enabling this may expose your contacts to your client developers and OS vendor. Not recommended. + Option Default Description + -------------------------------------- ---------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- + `push_notification_important_body` `New Message!` The body text to use when the stanza is important (see above), no message body is sent if this is empty + `push_max_errors` `16` How much persistent push errors are tolerated before notifications for the identifier in question are disabled + `push_max_devices` `5` The number of allowed devices per user (the oldest devices are automatically removed if this threshold is reached) + `push_max_hibernation_timeout` `259200` (72h) Number of seconds to extend the smacks timeout if no push was triggered yet (default: 72 hours) + `push_notification_with_body` (\*) `false` Whether or not to send the real message body to remote pubsub node. Without end-to-end encryption, enabling this may expose your message contents to your client developers and OS vendor. Not recommended. + `push_notification_with_sender` (\*) `false` Whether or not to send the real message sender to remote pubsub node. Enabling this may expose your contacts to your client developers and OS vendor. Not recommended. -(\*) There are privacy implications for enabling these options. +(\*) There are privacy implications for enabling these options.[^1] -Internal design notes -===================== +# Internal design notes -App servers are notified about offline messages, messages stored by [mod_mam] -or messages waiting in the smacks queue. -The business rules outlined [here](//mail.jabber.org/pipermail/standards/2016-February/030925.html) are all honored[^2]. +App servers are notified about offline messages, messages stored by +\[mod_mam\] or messages waiting in the smacks queue. The business rules +outlined +[here](//mail.jabber.org/pipermail/standards/2016-February/030925.html) +are all honored[^2]. -To cooperate with [mod_smacks] this module consumes some events: -`smacks-ack-delayed`, `smacks-hibernation-start` and `smacks-hibernation-end`. -These events allow this module to send out notifications for messages received -while the session is hibernated by [mod_smacks] or even when smacks -acknowledgements for messages are delayed by a certain amount of seconds -configurable with the [mod_smacks] setting `smacks_max_ack_delay`. +To cooperate with \[mod_smacks\] this module consumes some events: +`smacks-ack-delayed`, `smacks-hibernation-start` and +`smacks-hibernation-end`. These events allow this module to send out +notifications for messages received while the session is hibernated by +\[mod_smacks\] or even when smacks acknowledgements for messages are +delayed by a certain amount of seconds configurable with the +\[mod_smacks\] setting `smacks_max_ack_delay`. -The `smacks_max_ack_delay` setting allows to send out notifications to clients -which aren't already in smacks hibernation state (because the read timeout or -connection close didn't already happen) but also aren't responding to acknowledgement -request in a timely manner. This setting thus allows conversations to be smoother -under such circumstances. +The `smacks_max_ack_delay` setting allows to send out notifications to +clients which aren't already in smacks hibernation state (because the +read timeout or connection close didn't already happen) but also aren't +responding to acknowledgement request in a timely manner. This setting +thus allows conversations to be smoother under such circumstances. -The new event `cloud-notify-ping` can be used by any module to send out a cloud -notification to either all registered endpoints for the given user or only the endpoints -given in the event data. +The new event `cloud-notify-ping` can be used by any module to send out +a cloud notification to either all registered endpoints for the given +user or only the endpoints given in the event data. -The config setting `push_notification_important_body` can be used to specify an alternative -body text to send to the remote pubsub node if the stanza is encrypted or has a body. -This way the real contents of the message aren't revealed to the push appserver but it -can still see that the push is important. -This is used by Chatsecure on iOS to send out high priority pushes in those cases for example. +The config setting `push_notification_important_body` can be used to +specify an alternative body text to send to the remote pubsub node if +the stanza is encrypted or has a body. This way the real contents of the +message aren't revealed to the push appserver but it can still see that +the push is important. This is used by Chatsecure on iOS to send out +high priority pushes in those cases for example. -Compatibility -============= - -**Note:** This module should be used with Lua 5.2 and higher. Using it with -Lua 5.1 may cause push notifications to not be sent to some clients. +# Compatibility ------- ----------------------------------------------------------------------------- - trunk Works - 0.12 Works - 0.11 Works - 0.10 Works - 0.9 Support dropped, use last supported version [675726ab06d3](//hg.prosody.im/prosody-modules/raw-file/675726ab06d3/mod_cloud_notify/mod_cloud_notify.lua) ------- ----------------------------------------------------------------------------- +**Note:** This module should be used with Lua 5.2 and higher. Using it +with Lua 5.1 may cause push notifications to not be sent to some +clients. + ------- ----------------------------------------------------------------- + trunk Works as of 25-06-13 + 13 Works + 0.12 Works + ------- ----------------------------------------------------------------- -[^1]: The service which is expected to forward notifications to something like Google Cloud Messaging or Apple Notification Service -[^2]: [business_rules.markdown](//hg.prosody.im/prosody-modules/file/tip/mod_cloud_notify/business_rules.markdown) +[^1]: The service which is expected to forward notifications to + something like Google Cloud Messaging or Apple Notification Service + +[^2]: [business_rules.md](//hg.prosody.im/prosody-modules/file/tip/mod_cloud_notify/business_rules.md)
author Menel <menel@snikket.de>
date Fri, 13 Jun 2025 10:36:52 +0200
parent 4953:7d6ae8bb95dc
line wrap: on
line source

-- XEP-0355 (Namespace Delegation)
-- Copyright (C) 2015-2016 Jérôme Poisson
--
-- This module is MIT/X11 licensed. Please see the
-- COPYING file in the source package for more information.

-- This module manage namespace delegation, a way to delegate server features
-- to an external entity/component. Only the admin mode is implemented so far

-- TODO: client mode

local jid = require("util.jid")
local st = require("util.stanza")
local set = require("util.set")

local new_id = require("util.id").short;

local delegation_session = module:shared("/*/delegation/session")

-- FIXME: temporarily needed for disco_items_hook, to be removed when clean implementation is done
local is_contact_subscribed = require "core.rostermanager".is_contact_subscribed;
local jid_split = require "util.jid".split;
local jid_bare = require "util.jid".bare;

if delegation_session.connected_cb == nil then
	-- set used to have connected event listeners
	-- which allow a host to react on events from
	-- other hosts
	delegation_session.connected_cb = set.new()
end
local connected_cb = delegation_session.connected_cb

local _DELEGATION_NS = 'urn:xmpp:delegation:2'
local _FORWARDED_NS = 'urn:xmpp:forward:0'
local _DISCO_INFO_NS = 'http://jabber.org/protocol/disco#info'
local _DISCO_ITEMS_NS = 'http://jabber.org/protocol/disco#items'
local _DATA_NS = 'jabber:x:data'

local _MAIN_SEP = '::'
local _BARE_SEP = ':bare:'
local _REMAINING = ':*'
local _MAIN_PREFIX = _DELEGATION_NS.._MAIN_SEP
local _BARE_PREFIX = _DELEGATION_NS.._BARE_SEP
local _DISCO_REMAINING = _BARE_PREFIX.."disco#info".._REMAINING
local _DISCO_ITEMS_REMAINING = _BARE_PREFIX.."disco#items".._REMAINING
local _PREFIXES = {_MAIN_PREFIX, _BARE_PREFIX}

local disco_nest

module:log("debug", "Loading namespace delegation module ")

--> Configuration management <--

local ns_delegations = module:get_option("delegations", {})

local jid2ns = {}
for namespace, ns_data in pairs(ns_delegations) do
	-- "connected" contain the full jid of connected managing entity
	ns_data.connected = nil
	if ns_data.jid then
		if jid2ns[ns_data.jid] == nil then
			jid2ns[ns_data.jid] = {}
		end
		jid2ns[ns_data.jid][namespace] = ns_data
		module:log("debug", "Namespace %s is delegated%s to %s", namespace, ns_data.filtering and " (with filtering)" or "", ns_data.jid)
	else
		module:log("warn", "Ignoring delegation for %s: no jid specified", tostring(namespace))
		ns_delegations[namespace] = nil
	end
end


local function advertise_delegations(session, to_jid)
	-- send <message/> stanza to advertise delegations
	-- as expained in § 4.2
	local message = st.message({from=module.host, to=to_jid})
					  :tag("delegation", {xmlns=_DELEGATION_NS})

	-- we need to check if a delegation is granted because the configuration
	-- can be complicated if some delegations are granted to bare jid
	-- and other to full jids, and several resources are connected.
	local have_delegation = false

	for namespace, ns_data  in pairs(jid2ns[to_jid]) do
		if ns_data.connected == to_jid then
			have_delegation = true
			message:tag("delegated", {namespace=namespace})
			if type(ns_data.filtering) == "table" then
				for _, attribute in pairs(ns_data.filtering) do
					message:tag("attribute", {name=attribute}):up()
				end
			end
			message:up()
		end
	end

	if have_delegation then
		session.send(message)
	end
end

local function set_connected(entity_jid)
	-- set the "connected" key for all namespace managed by entity_jid
	-- if the namespace has already a connected entity, ignore the new one
	local function set_config(jid_)
		for namespace, ns_data in pairs(jid2ns[jid_]) do
			if ns_data.connected == nil then
				ns_data.connected = entity_jid
				-- disco remaining and disco items remaining are special namespaces
				-- there is no disco nesting for them
				if namespace ~= _DISCO_ITEMS_REMAINING and namespace ~= _DISCO_REMAINING then
					disco_nest(namespace, entity_jid)
				end
			end
		end
	end
	local bare_jid = jid.bare(entity_jid)
	set_config(bare_jid)
	-- We can have a bare jid of a full jid specified in configuration
	-- so we try our luck with both (first connected resource will
	-- manage the namespaces in case of bare jid)
	if bare_jid ~= entity_jid then
		set_config(entity_jid)
		jid2ns[entity_jid] = jid2ns[bare_jid]
	end
end

local function on_presence(event)
	local session = event.origin
	local bare_jid = jid.bare(session.full_jid)

	if jid2ns[bare_jid] or jid2ns[session.full_jid] then
		set_connected(session.full_jid)
		advertise_delegations(session, session.full_jid)
	end
end

local function on_component_connected(event)
	-- method called by the module loaded by the component
	-- /!\ the event come from the component host,
	-- not from the host of this module
	local session = event.session
	local bare_jid = jid.join(session.username, session.host)

	local jid_delegations = jid2ns[bare_jid]
	if jid_delegations ~= nil then
		set_connected(bare_jid)
		advertise_delegations(session, bare_jid)
	end
end

local function on_component_auth(event)
	-- react to component-authenticated event from this host
	-- and call the on_connected methods from all other hosts
	-- needed for the component to get delegations advertising
	for callback in connected_cb:items() do
		callback(event)
	end
end

if module:get_host_type() ~= "component" then
	connected_cb:add(on_component_connected)
end
module:hook('component-authenticated', on_component_auth)
module:hook('presence/initial', on_presence)


--> delegated namespaces hook <--

local managing_ent_error
local stanza_cache = {} -- we cache original stanza to build reply

local function clean_xmlns(node)
    -- Recursively remove "jabber:client" attribute from node.
    -- In Prosody internal routing, xmlns should not be set.
    -- Keeping xmlns would lead to issues like mod_smacks ignoring the outgoing stanza,
    -- so we remove all xmlns attributes with a value of "jabber:client"
    -- note: this function comes from mod_privilege
    if node.attr.xmlns == 'jabber:client' then
        for childnode in node:childtags() do
            clean_xmlns(childnode)
        end
        node.attr.xmlns = nil
    end
end

local function managing_ent_result(event)
	-- this function manage iq results from the managing entity
	-- it do a couple of security check before sending the
	-- result to the managed entity
	local stanza = event.stanza
	if stanza.attr.to ~= module.host then
		module:log("warn", 'forwarded stanza result has "to" attribute not addressed to current host, id conflict ?')
		return
	end
	module:unhook("iq-result/host/"..stanza.attr.id, managing_ent_result)
	module:unhook("iq-error/host/"..stanza.attr.id, managing_ent_error)

	-- lot of checks to do...
	local delegation = stanza.tags[1]
	if #stanza ~= 1 or delegation.name ~= "delegation" or
		delegation.attr.xmlns ~= _DELEGATION_NS then
		module:log("warn", "ignoring invalid iq result from managing entity %s", stanza.attr.from)
		stanza_cache[stanza.attr.from][stanza.attr.id] = nil
		return true
	end

	local forwarded = delegation.tags[1]
	if #delegation ~= 1 or forwarded.name ~= "forwarded" or
		forwarded.attr.xmlns ~= _FORWARDED_NS then
		module:log("warn", "ignoring invalid iq result from managing entity %s", stanza.attr.from)
		stanza_cache[stanza.attr.from][stanza.attr.id] = nil
		return true
	end

	local iq = forwarded.tags[1]
	if #forwarded ~= 1 or iq.name ~= "iq" or
		iq.attr.xmlns ~= 'jabber:client' or
		(iq.attr.type =='result' and #iq > 1) or
		(iq.attr.type == 'error' and #iq > 2) then
		module:log("warn", "ignoring invalid iq result from managing entity %s", stanza.attr.from)
		stanza_cache[stanza.attr.from][stanza.attr.id] = nil
		return true
	end

	clean_xmlns(iq)

	local original = stanza_cache[stanza.attr.from][stanza.attr.id]
	stanza_cache[stanza.attr.from][stanza.attr.id] = nil
	-- we get namespace from original and not iq
	-- because the namespace can be lacking in case of error
	local namespace = original.tags[1].attr.xmlns

	-- small hack for disco remaining feat
	if namespace == _DISCO_ITEMS_NS then
		namespace = _DISCO_ITEMS_REMAINING
	elseif namespace == _DISCO_INFO_NS then
		namespace = _DISCO_REMAINING
	end

	local ns_data = ns_delegations[namespace]

	if stanza.attr.from ~= ns_data.connected or (iq.attr.type ~= "result" and iq.attr.type ~= "error") or
		iq.attr.id ~= original.attr.id or iq.attr.to ~= original.attr.from then
		module:log("warn", "ignoring forbidden iq result from managing entity %s, please check that the component is no trying to do something bad (stanza: %s)", stanza.attr.from, tostring(stanza))
		module:send(st.error_reply(original, 'cancel', 'service-unavailable'))
		return true
	end

	-- at this point eveything is checked,
	-- and we (hopefully) can send the the result safely
	module:send(iq)
	return true
end

function managing_ent_error(event)
	local stanza = event.stanza
	if stanza.attr.to ~= module.host then
		module:log("warn", 'Stanza result has "to" attribute not addressed to current host, id conflict ?')
		return
	end
	module:unhook("iq-result/host/"..stanza.attr.id, managing_ent_result)
	module:unhook("iq-error/host/"..stanza.attr.id, managing_ent_error)
	local original = stanza_cache[stanza.attr.from][stanza.attr.id]
	stanza_cache[stanza.attr.from][stanza.attr.id] = nil
	module:log("warn", "Got an error after forwarding stanza to "..stanza.attr.from)
	module:send(st.error_reply(original, 'cancel', 'service-unavailable'))
	return true
end

local function forward_iq(stanza, ns_data)
	local to_jid = ns_data.connected
	stanza.attr.xmlns = 'jabber:client'
	local iq_id = new_id();
	local iq_stanza  = st.iq({ from=module.host, to=to_jid, type="set", id = iq_id })
		:tag("delegation", { xmlns=_DELEGATION_NS })
		:tag("forwarded", { xmlns=_FORWARDED_NS })
		:add_child(stanza)
	-- we save the original stanza to check the managing entity result
	if not stanza_cache[to_jid] then stanza_cache[to_jid] = {} end
	stanza_cache[to_jid][iq_id] = stanza
	module:hook("iq-result/host/"..iq_id, managing_ent_result)
	module:hook("iq-error/host/"..iq_id, managing_ent_error)
	module:log("debug", "stanza forwarded")
	module:send(iq_stanza)
end

local function iq_hook(event)
	-- general hook for all the iq which forward delegated ones
	-- and continue normal behaviour else. If a namespace is
	-- delegated but managing entity is offline, a service-unavailable
	-- error will be sent, as requested by the XEP
	local session, stanza = event.origin, event.stanza
	if #stanza == 1 and stanza.attr.type == 'get' or stanza.attr.type == 'set' then
		local namespace = stanza.tags[1].attr.xmlns
		local ns_data = ns_delegations[namespace]

		if ns_data then
			if stanza.attr.from == ns_data.connected then
				-- we don't forward stanzas from managing entity itself
				return
			end
			if ns_data.filtering then
				local first_child = stanza.tags[1]
				for _, attribute in pairs(ns_data.filtering) do
					-- if any filtered attribute if not present,
					-- we must continue the normal bahaviour
					if not first_child.attr[attribute] then
						-- Filtered attribute is not present, we do normal workflow
						return
					end
				end
			end
			if not ns_data.connected then
				module:log("warn", "No connected entity to manage "..namespace)
				session.send(st.error_reply(stanza, 'cancel', 'service-unavailable'))
			else
				forward_iq(stanza, ns_data)
			end
			return true
		else
			-- we have no delegation, we continue normal behaviour
			return
		end
	end
end

module:hook("iq/self", iq_hook, 2^32)
module:hook("iq/bare", iq_hook, 2^32)
module:hook("iq/host", iq_hook, 2^32)


--> discovery nesting <--

-- disabling internal features/identities

local function find_form_type(stanza)
	local form_type = nil
	for field in stanza:childtags('field', 'jabber:x:data') do
		if field.attr.var=='FORM_TYPE' and field.attr.type=='hidden' then
			local value = field:get_child('value')
			if not value then
				module:log("warn", "No value found in FORM_TYPE field: "..tostring(stanza))
			else
				form_type=value.get_text()
			end
		end
	end
	return form_type
end

-- modules whose features/identities are managed by delegation
local disabled_modules = set.new()
local disabled_identities = set.new()

local function identity_added(event)
	local source = event.source
	if disabled_modules:contains(source) then
		local item = event.item
		local category, type_, name = item.category, item.type, item.name
		module:log("debug", "Removing (%s/%s%s) identity because of delegation", category, type_, name and "/"..name or "")
		disabled_identities:add(item)
		source:remove_item("identity", item)
	end
end

local function feature_added(event)
	local source, item = event.source, event.item
	for namespace, _ in pairs(ns_delegations) do
		if source ~= nil and source ~= module and string.sub(item, 1, #namespace) == namespace then
			module:log("debug", "Removing %s feature which is delegated", item)
			source:remove_item("feature", item)
			disabled_modules:add(source)
			if source.items and source.items.identity then
				-- we remove all identities added by the source module
				-- that can cause issues if the module manages several features/identities
				-- but this case is probably rare (or doesn't happen at all)
				-- FIXME: any better way ?
				for _, identity in pairs(source.items.identity) do
					identity_added({source=source, item=identity})
				end
			end
		end
	end
end

local function extension_added(event)
	local source, stanza = event.source, event.item
	local form_type = find_form_type(stanza)
	if not form_type then return end

	for namespace, _ in pairs(ns_delegations) do
		if source ~= nil and source ~= module and string.sub(form_type, 1, #namespace) == namespace then
			module:log("debug", "Removing extension which is delegated: %s", tostring(stanza))
			source:remove_item("extension", stanza)
		end
	end
end

-- for disco nesting (see § 7.2) we need to remove internal features
-- we use handle_items as it allows to remove already added features
-- and catch the ones which can come later
module:handle_items("feature", feature_added, function(_) end)
module:handle_items("identity", identity_added, function(_) end, false)
module:handle_items("extension", extension_added, function(_) end)


-- managing entity features/identities collection

local disco_error
local bare_features = set.new()
local bare_identities = {}
local bare_extensions = {}

local function disco_result(event)
	-- parse result from disco nesting request
	-- and fill module features/identities and bare_features/bare_identities accordingly
	local session, stanza = event.origin, event.stanza
	if stanza.attr.to ~= module.host then
		module:log("warn", 'Stanza result has "to" attribute not addressed to current host, id conflict ?')
		return
	end
	module:unhook("iq-result/host/"..stanza.attr.id, disco_result)
	module:unhook("iq-error/host/"..stanza.attr.id, disco_error)
	local query = stanza:get_child("query", _DISCO_INFO_NS)
	if not query or not query.attr.node then
		session.send(st.error_reply(stanza, 'modify', 'not-acceptable'))
		return true
	end

	local node = query.attr.node
	local main

	if string.sub(node, 1, #_MAIN_PREFIX) == _MAIN_PREFIX then
		main=true
	elseif string.sub(node, 1, #_BARE_PREFIX) == _BARE_PREFIX then
		main=false
	else
		module:log("warn", "Unexpected node: "..node)
		session.send(st.error_reply(stanza, 'modify', 'not-acceptable'))
		return true
	end

	for feature in query:childtags("feature") do
		local namespace = feature.attr.var
		if main then
			module:add_feature(namespace)
		else
			bare_features:add(namespace)
		end
	end
	for identity in query:childtags("identity") do
		local category, type_, name = identity.attr.category, identity.attr.type, identity.attr.name
		if main then
			module:add_identity(category, type_, name)
		else
			table.insert(bare_identities, {category=category, type=type_, name=name})
		end
	end
	for extension in query:childtags("x", _DATA_NS) do
		if main then
			module:add_extension(extension)
		else
			table.insert(bare_extensions, extension)
		end
	end
end

function disco_error(event)
	local stanza = event.stanza
	if stanza.attr.to ~= module.host then
		module:log("warn", 'Stanza result has "to" attribute not addressed to current host, id conflict ?')
		return
	end
	module:unhook("iq-result/host/"..stanza.attr.id, disco_result)
	module:unhook("iq-error/host/"..stanza.attr.id, disco_error)
	module:log("warn", "Got an error while requesting disco for nesting to "..stanza.attr.from)
	module:log("warn", "Ignoring disco nesting")
end

function disco_nest(namespace, entity_jid)
	-- manage discovery nesting (see § 7.2)

	-- first we reset the current values
	if module.items then
		module.items['feature'] = nil
		module.items['identity'] = nil
		module.items['extension'] = nil
		bare_features = set.new()
		bare_identities = {}
		bare_extensions = {}
	end

	for _, prefix in ipairs(_PREFIXES) do
		local node = prefix..namespace

		local iq_id = new_id();
		local iq = st.iq({from=module.host, to=entity_jid, type='get', id = iq_id })
			:tag('query', {xmlns=_DISCO_INFO_NS, node=node})

		module:hook("iq-result/host/"..iq_id, disco_result)
		module:hook("iq-error/host/"..iq_id, disco_error)
		module:send(iq)
	end
end

-- disco to bare jids special cases

-- disco#info

local function disco_hook(event)
	-- this event is called when a disco info request is done on a bare jid
	-- we get the final reply and filter delegated features/identities/extensions
	local reply = event.reply
	reply.tags[1]:maptags(function(child)
		if child.name == 'feature' then
			local feature_ns = child.attr.var
			for namespace, _ in pairs(ns_delegations) do
				if string.sub(feature_ns, 1, #namespace) == namespace then
					module:log("debug", "Removing feature namespace %s which is delegated", feature_ns)
					return nil
				end
			end
		elseif child.name == 'identity' then
			for item in disabled_identities:items() do
				if item.category == child.attr.category
					and item.type == child.attr.type
					-- we don't check name, because mod_pep use a name for main disco, but not in account-disco-info hook
					-- and item.name == child.attr.name
				then
					module:log("debug", "Removing (%s/%s%s) identity because of delegation", item.category, item.type, item.name and "/"..item.name or "")
					return nil
				end
			end
		elseif child.name == 'x' and child.attr.xmlns == _DATA_NS then
			local form_type = find_form_type(child)
			if form_type then
				for namespace, _ in pairs(ns_delegations) do
					if string.sub(form_type, 1, #namespace) == namespace then
						module:log("debug", "Removing extension which is delegated: %s", tostring(child))
						return nil
					end
				end
			end

		end
		return child
	end)
	for feature in bare_features:items() do
		reply:tag('feature', {var=feature}):up()
	end
	for _, item in ipairs(bare_identities) do
		reply:tag('identity', {category=item.category, type=item.type, name=item.name}):up()
	end
	for _, stanza in ipairs(bare_extensions) do
		reply:add_child(stanza)
	end

end
module:hook("account-disco-info", disco_hook, -2^32)

local function disco_node_hook(event)
	-- we reach this hook if a disco node on account has not been found
	-- we then forward the request to managing entity
	if not event.exists then
		-- this node is not handled by the server
		local ns_data = ns_delegations[_DISCO_REMAINING]
		if ns_data ~= nil then
			-- remaining delegation is requested, we forward
			forward_iq(event.stanza, ns_data)
			-- and stop normal event handling
			return true
		end
	end
end
module:hook("account-disco-info-node", disco_node_hook, -2^32)

-- disco#items

local function disco_items_node_hook(event)
	-- check if node is not handled by server
	-- and forward the disco request to suitable entity
	if not event.exists then
		-- this node is not handled by the server
		local ns_data = ns_delegations[_DISCO_ITEMS_REMAINING]
		if ns_data ~= nil then
			-- remaining delegation is requested, we forward
			forward_iq(event.stanza, ns_data)
			-- and stop normal event handling
			return true
		end
	end
end
module:hook("account-disco-items-node", disco_items_node_hook, -2^32)

local function disco_items_hook(event)
	-- FIXME: we forward all bare-jid disco-items requests (without node) which will replace any Prosody reply
	--		for now it's OK because Prosody is not returning anything on request on bare jid
	--		but to be properly done, any Prosody reply should be kept and managing entities items should be added (merged) to it.
	--		account-disco-items can't be cancelled (return value of hooks are not checked in mod_disco), so coroutine needs
	--		to be used with util.async (to get the IQ result, merge items then return from the event)
	local origin, stanza = event.origin, event.stanza;
	local node = stanza.tags[1].attr.node;
	local username = jid_split(stanza.attr.to) or origin.username;
	if not stanza.attr.to or is_contact_subscribed(username, module.host, jid_bare(stanza.attr.from)) then
		if node == nil or node == "" then
			local ns_data = ns_delegations[_DISCO_ITEMS_REMAINING]
			if ns_data ~= nil then
				forward_iq(event.stanza, ns_data)
				return true
			end
		end
	end
end
module:hook("iq-get/bare/http://jabber.org/protocol/disco#items:query", disco_items_hook, 100)

local function disco_items_raw_hook(event)
	-- this method is called when account-disco-items-* event are not called
	-- notably when a disco-item is done by an unsubscibed entity
	-- (i.e. an entity doing a disco#item on an entity without having
	-- presence subscription)
	-- we forward the request to managing entity
	-- it's the responsability of the managing entity to filter the items
	local ns_data = ns_delegations[_DISCO_ITEMS_REMAINING]
	if ns_data ~= nil then
		forward_iq(event.stanza, ns_data)
		return true
	end
end
module:hook("iq-get/bare/http://jabber.org/protocol/disco#items:query", disco_items_raw_hook, -2^32)