path: root/rpki/rpkid_tasks.py

# $Id$
#
# Copyright (C) 2014  Dragon Research Labs ("DRL")
# Portions copyright (C) 2012--2013  Internet Systems Consortium ("ISC")
#
# Permission to use, copy, modify, and distribute this software for any
# purpose with or without fee is hereby granted, provided that the above
# copyright notices and this permission notice appear in all copies.
#
# THE SOFTWARE IS PROVIDED "AS IS" AND DRL AND ISC DISCLAIM ALL
# WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED
# WARRANTIES OF MERCHANTABILITY AND FITNESS.  IN NO EVENT SHALL DRL OR
# ISC BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL
# DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA
# OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER
# TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
# PERFORMANCE OF THIS SOFTWARE.

"""
rpkid task objects.  Split out from rpki.left_right and rpki.rpkid
because interactions with rpkid scheduler were getting too complicated.
"""

import logging
import random

import tornado.gen
import tornado.web
import tornado.locks
import tornado.ioloop
import tornado.httputil
import tornado.httpclient
import tornado.httpserver

import rpki.log
import rpki.rpkid
import rpki.up_down
import rpki.sundial
import rpki.publication
import rpki.exceptions

logger = logging.getLogger(__name__)

task_classes = ()

def queue_task(cls):
    """
    Class decorator to add a new task class to task_classes.
    """

    global task_classes                 # pylint: disable=W0603
    task_classes += (cls,)
    return cls


class PostponeTask(Exception):
    """
    Exit a task without finishing it.  We use this to signal that a
    long-running task wants to yield to the task loop but hasn't yet
    run to completion.
    """


class AbstractTask(object):
    """
    Abstract base class for rpkid scheduler task objects.
    """

    ## @var timeslice
    # How long before a task really should consider yielding the CPU
    # to let something else run.  Should this be something we can
    # configure from rpki.conf?

    #timeslice = rpki.sundial.timedelta(seconds = 15)
    timeslice = rpki.sundial.timedelta(seconds = 120)

    def __init__(self, rpkid, tenant, description = None):
        self.rpkid       = rpkid
        self.tenant      = tenant
        self.description = description
        self.done_this   = None
        self.done_next   = None
        self.due_date    = None
        self.started     = False
        self.postponed   = False
        self.clear()

    def __repr__(self):
        return rpki.log.log_repr(self, self.description)

    @tornado.gen.coroutine
    def start(self):
        try:
            logger.debug("%r: Starting", self)
            self.due_date = rpki.sundial.now() + self.timeslice
            self.clear()
            self.started = True
            self.postponed = False
            yield self.main()
        except PostponeTask:
            self.postponed = True
        except:
            logger.exception("%r: Unhandled exception", self)
        finally:
            self.due_date = None
            self.started  = False
            self.clear()
            if self.postponed:
                logger.debug("%r: Postponing", self)
                self.rpkid.task_add(self)
            else:
                logger.debug("%r: Exiting", self)
                if self.done_this is not None:
                    self.done_this.notify_all()
                self.done_this = self.done_next
                self.done_next = None

    def wait(self):
        done = "done_next" if self.started else "done_this"
        condition = getattr(self, done)
        if condition is None:
            condition = tornado.locks.Condition()
            setattr(self, done, condition)
        future = condition.wait()
        return future

    def waiting(self):
        return self.done_this is not None

    @tornado.gen.coroutine
    def overdue(self):
        yield tornado.gen.moment
        raise tornado.gen.Return(rpki.sundial.now() > self.due_date and
                                 any(not task.postponed for task in self.rpkid.task_ready))

    @tornado.gen.coroutine
    def main(self):
        raise NotImplementedError

    def clear(self):
        pass


@queue_task
class PollParentTask(AbstractTask):
    """
    Run the regular client poll cycle with each of this tenant's
    parents, in turn.
    """

    @tornado.gen.coroutine
    def main(self):
        logger.debug("%r: Polling parents", self)

        for parent in self.tenant.parents.all():
            try:
                logger.debug("%r: Executing list query", self)
                list_r_msg = yield parent.up_down_list_query(rpkid = self.rpkid)
            except:
                logger.exception("%r: Couldn't get resource class list from %r, skipping", self, parent)
                continue

            logger.debug("%r: Parsing list response", self)

            ca_map = dict((ca.parent_resource_class, ca) for ca in parent.cas.all())

            for rc in list_r_msg.getiterator(rpki.up_down.tag_class):
                try:
                    class_name = rc.get("class_name")
                    ca = ca_map.pop(class_name, None)
                    if ca is None:
                        yield self.create(parent = parent, rc = rc, class_name = class_name)
                    else:
                        yield self.update(parent = parent, rc = rc, class_name = class_name, ca = ca)
                except:
                    logger.exception("Couldn't update resource class %r, skipping", class_name)

            for class_name, ca in ca_map.iteritems():
                logger.debug("%r: Destroying orphaned %r for resource class %r", self, ca, class_name)
                yield ca.destroy(rpkid = self.rpkid, parent = parent)

    @tornado.gen.coroutine
    def create(self, parent, rc, class_name):
        logger.debug("%r: Creating new CA for resource class %r", self, class_name)
        ca = rpki.rpkidb.models.CA.objects.create(
            parent                = parent,
            parent_resource_class = class_name,
            sia_uri               = parent.construct_sia_uri(rc))
        ca_detail = ca.create_detail()
        r_msg = yield parent.up_down_issue_query(rpkid = self.rpkid, ca = ca, ca_detail = ca_detail)
        elt  = r_msg.find(rpki.up_down.tag_class).find(rpki.up_down.tag_certificate)
        uri  = elt.get("cert_url")
        cert = rpki.x509.X509(Base64 = elt.text)
        logger.debug("%r: %r received certificate %s", self, ca, uri)
        yield ca_detail.activate(rpkid = self.rpkid, ca = ca, cert = cert, uri = uri)

    @tornado.gen.coroutine
    def update(self, parent, rc, class_name, ca):

        # pylint: disable=C0330

        logger.debug("%r: Checking updates for %r", self, ca)

        sia_uri = parent.construct_sia_uri(rc)
        sia_uri_changed = ca.sia_uri != sia_uri

        if sia_uri_changed:
            logger.debug("SIA changed: was %s now %s", ca.sia_uri, sia_uri)
            ca.sia_uri = sia_uri

        rc_resources = rpki.resource_set.resource_bag(
            rc.get("resource_set_as"),
            rc.get("resource_set_ipv4"),
            rc.get("resource_set_ipv6"),
            rc.get("resource_set_notafter"))

        cert_map = {}

        for c in rc.getiterator(rpki.up_down.tag_certificate):
            x = rpki.x509.X509(Base64 = c.text)
            u = rpki.up_down.multi_uri(c.get("cert_url")).rsync()
            cert_map[x.gSKI()] = (x, u)

        ca_details = ca.ca_details.exclude(state = "revoked")

        if not ca_details:
            logger.warning("Existing resource class %s to %s from %s with no certificates, rekeying",
                           class_name, parent.tenant.tenant_handle, parent.parent_handle)
            yield ca.rekey(rpkid = self.rpkid)
            return

        for ca_detail in ca_details:

            rc_cert, rc_cert_uri = cert_map.pop(ca_detail.public_key.gSKI(), (None, None))

            if rc_cert is None:
                logger.warning("g(SKI) %s in resource class %s is in database but missing from list_response to %s from %s, "
                               "maybe parent certificate went away?",
                               ca_detail.public_key.gSKI(), class_name, parent.tenant.tenant_handle, parent.parent_handle)
                publisher = rpki.rpkid.publication_queue(rpkid = self.rpkid)
                ca_detail.destroy(publisher = publisher)
                yield publisher.call_pubd()
                continue

            if ca_detail.state == "active" and ca_detail.ca_cert_uri != rc_cert_uri:
                logger.debug("AIA changed: was %s now %s", ca_detail.ca_cert_uri, rc_cert_uri)
                ca_detail.ca_cert_uri = rc_cert_uri
                ca_detail.save()

            if ca_detail.state not in ("pending", "active"):
                continue

            if ca_detail.state == "pending":
                current_resources = rpki.resource_set.resource_bag()
            else:
                current_resources = ca_detail.latest_ca_cert.get_3779resources()

            if (ca_detail.state == "pending" or
                sia_uri_changed or
                ca_detail.latest_ca_cert != rc_cert or
                ca_detail.latest_ca_cert.getNotAfter() != rc_resources.valid_until or
                current_resources.undersized(rc_resources) or
                current_resources.oversized(rc_resources)):

                yield ca_detail.update(
                    rpkid            = self.rpkid,
                    parent           = parent,
                    ca               = ca,
                    rc               = rc,
                    sia_uri_changed  = sia_uri_changed,
                    old_resources    = current_resources)

        if cert_map:
            logger.warning("Unknown certificate g(SKI)%s %s in resource class %s in list_response to %s from %s, maybe you want to \"revoke_forgotten\"?",
                           "" if len(cert_map) == 1 else "s", ", ".join(cert_map), class_name, parent.tenant.tenant_handle, parent.parent_handle)


@queue_task
class UpdateChildrenTask(AbstractTask):
    """
    Check for updated IRDB data for all of this tenant's children and
    issue new certs as necessary.  Must handle changes both in
    resources and in expiration date.
    """

    @tornado.gen.coroutine
    def main(self):
        logger.debug("%r: Updating children", self)
        now = rpki.sundial.now()
        rsn = now + rpki.sundial.timedelta(seconds = self.tenant.regen_margin)
        publisher = rpki.rpkid.publication_queue(self.rpkid)
        postponing = False

        child_certs    = rpki.rpkidb.models.ChildCert.objects.filter(child__tenant = self.tenant, ca_detail__state = "active")
        child_handles  = sorted(set(child_cert.child.child_handle for child_cert in child_certs))
        irdb_resources = dict(zip(child_handles, (yield self.rpkid.irdb_query_children_resources(self.tenant.tenant_handle, child_handles))))

        for child_cert in child_certs:
            try:
                ca_detail = child_cert.ca_detail
                child_handle = child_cert.child.child_handle
                old_resources = child_cert.cert.get_3779resources()
                new_resources = old_resources & irdb_resources[child_handle] & ca_detail.latest_ca_cert.get_3779resources()
                old_aia = child_cert.cert.get_AIA()[0]
                new_aia = ca_detail.ca_cert_uri

                assert child_cert.gski == child_cert.cert.gSKI()

                if new_resources.empty():
                    logger.debug("Resources shrank to null set, revoking and withdrawing child %s g(SKI) %s",
                                 child_handle, child_cert.gski)
                    child_cert.revoke(publisher = publisher)
                    ca_detail.generate_crl_and_manifest(publisher = publisher)

                elif (old_resources != new_resources or old_aia != new_aia or
                      (old_resources.valid_until < rsn and
                       irdb_resources[child_handle].valid_until > now and
                       old_resources.valid_until != irdb_resources[child_handle].valid_until)):
                    logger.debug("Need to reissue child %s certificate g(SKI) %s", child_handle,
                                 child_cert.gski)
                    if old_resources != new_resources:
                        logger.debug("Child %s g(SKI) %s resources changed: old %s new %s",
                                     child_handle, child_cert.gski, old_resources, new_resources)
                    if old_resources.valid_until != irdb_resources[child_handle].valid_until:
                        logger.debug("Child %s g(SKI) %s validity changed: old %s new %s",
                                     child_handle, child_cert.gski, old_resources.valid_until,
                                     irdb_resources[child_handle].valid_until)

                    new_resources.valid_until = irdb_resources[child_handle].valid_until
                    child_cert.reissue(ca_detail = ca_detail, resources = new_resources, publisher = publisher)

                elif old_resources.valid_until < now:
                    logger.debug("Child %s certificate g(SKI) %s has expired: cert.valid_until %s, irdb.valid_until %s",
                                 child_handle, child_cert.gski, old_resources.valid_until,
                                 irdb_resources[child_handle].valid_until)
                    child_cert.delete()
                    publisher.queue(uri = child_cert.uri,
                                    old_obj = child_cert.cert,
                                    repository = ca_detail.ca.parent.repository)
                    ca_detail.generate_crl_and_manifest(publisher = publisher)

            except:
                logger.exception("%r: Couldn't update %r, skipping", self, child_cert)

            finally:
                if (yield self.overdue()):
                    postponing = True
                    break

        try:
            yield publisher.call_pubd()
        except:
            logger.exception("%r: Couldn't publish, skipping", self)

        if postponing:
            raise PostponeTask


@queue_task
class UpdateROAsTask(AbstractTask):
    """
    Generate or update ROAs for this tenant.
    """

    # XXX This might need rewriting to avoid race conditions.
    #
    # There's a theoretical race condition here if we're chugging away
    # and something else needs to update the manifest or CRL, or if
    # some back-end operation generates or destroys ROAs.  The risk is
    # fairly low given that we defer CRL and manifest generation until
    # we're ready to publish, but it's theoretically present.

    @tornado.gen.coroutine
    def main(self):
        logger.debug("%r: Updating ROAs", self)

        try:
            r_msg = yield self.rpkid.irdb_query_roa_requests(self.tenant.tenant_handle)
        except:
            logger.exception("Could not fetch ROA requests for %s, skipping", self.tenant.tenant_handle)
            return

        logger.debug("%r: Received response to query for ROA requests: %r", self, r_msg)

        roas = {}
        seen = set()
        orphans = []
        creates = []
        updates = []
        publisher = rpki.rpkid.publication_queue(self.rpkid)
        ca_details = set()
        
        for roa in self.tenant.roas.all():
            k = "{!s} {!s} {!s}".format(roa.asn, roa.ipv4, roa.ipv6)
            if k not in roas:
                roas[k] = roa
            elif roa.roa is not None and roa.cert is not None and roa.ca_detail is not None and roa.ca_detail.state == "active" and (roas[k].roa is None or roas[k].cert is None or roas[k].ca_detail is None or roas[k].ca_detail.state != "active"):
                orphans.append(roas[k])
                roas[k] = roa
            else:
                orphans.append(roa)

        for r_pdu in r_msg:
            k = "{!s} {!s} {!s}".format(r_pdu.get("asn"), r_pdu.get("ipv4"), r_pdu.get("ipv6"))
            if k in seen:
                logger.warning("%r: Skipping duplicate ROA request %r", self, r_pdu)
                continue
            seen.add(k)
            roa = roas.pop(k, None)
            if roa is None:
                roa = rpki.rpkidb.models.ROA(tenant = self.tenant, asn = long(r_pdu.get("asn")), ipv4 = r_pdu.get("ipv4"), ipv6 = r_pdu.get("ipv6"))
                logger.debug("%r: Try to create %r", self, roa)
                creates.append(roa)
            else:
                logger.debug("%r: Found existing %r", self, roa)
                updates.append(roa)

        orphans.extend(roas.itervalues())

        roas = creates + updates

        r_msg = seen = creates = updates = None

        postponing = False

        while roas and not postponing:
            if (yield self.overdue()):
                postponing = True
                break
            roa = roas.pop(0)
            try:
                roa.update(publisher = publisher)
                ca_details.add(roa.ca_detail)
            except rpki.exceptions.NoCoveringCertForROA:
                logger.warning("%r: No covering certificate for %r, skipping", self, roa)
            except:
                logger.exception("%r: Could not update %r, skipping", self, roa)

        if not postponing:
            for roa in orphans:
                try:
                    ca_details.add(roa.ca_detail)
                    roa.revoke(publisher = publisher)
                except:
                    logger.exception("%r: Could not revoke %r", self, roa)

        if not publisher.empty():
            for ca_detail in ca_details:
                logger.debug("%r: Generating new CRL and manifest for %r", self, ca_detail)
                ca_detail.generate_crl_and_manifest(publisher = publisher)
            yield publisher.call_pubd()

        if postponing:
            raise PostponeTask


@queue_task
class UpdateGhostbustersTask(AbstractTask):
    """
    Generate or update Ghostbuster records for this tenant.

    This was originally based on the ROA update code.  It's possible
    that both could benefit from refactoring, but at this point the
    potential scaling issues for ROAs completely dominate structure of
    the ROA code, and aren't relevant here unless someone is being
    exceptionally silly.
    """

    @tornado.gen.coroutine
    def main(self):
        logger.debug("%r: Updating Ghostbuster records", self)
        parent_handles = set(p.parent_handle for p in self.tenant.parents.all())

        try:
            r_msg = yield self.rpkid.irdb_query_ghostbuster_requests(self.tenant.tenant_handle, parent_handles)

            ghostbusters = {}
            orphans = []
            publisher = rpki.rpkid.publication_queue(self.rpkid)
            ca_details = set()
            seen = set()

            for ghostbuster in self.tenant.ghostbusters.all():
                k = (ghostbuster.ca_detail.pk, ghostbuster.vcard)
                if ghostbuster.ca_detail.state != "active" or k in ghostbusters:
                    orphans.append(ghostbuster)
                else:
                    ghostbusters[k] = ghostbuster

            for r_pdu in r_msg:
                try:
                    self.tenant.parents.get(parent_handle = r_pdu.get("parent_handle"))
                except rpki.rpkidb.models.Parent.DoesNotExist:
                    logger.warning("%r: Unknown parent_handle %r in Ghostbuster request, skipping", self, r_pdu.get("parent_handle"))
                    continue
                k = (r_pdu.get("parent_handle"), r_pdu.text)
                if k in seen:
                    logger.warning("%r: Skipping duplicate Ghostbuster request %r", self, r_pdu)
                    continue
                seen.add(k)
                for ca_detail in rpki.rpkidb.models.CADetail.objects.filter(ca__parent__parent_handle = r_pdu.get("parent_handle"), ca__parent__tenant = self.tenant, state = "active"):
                    ghostbuster = ghostbusters.pop((ca_detail.pk, r_pdu.text), None)
                    if ghostbuster is None:
                        ghostbuster = rpki.rpkidb.models.Ghostbuster(tenant = self.tenant, ca_detail = ca_detail, vcard = r_pdu.text)
                        logger.debug("%r: Created new %r for %r", self, ghostbuster, r_pdu.get("parent_handle"))
                    else:
                        logger.debug("%r: Found existing %r for %r", self, ghostbuster, r_pdu.get("parent_handle"))
                    ghostbuster.update(publisher = publisher)
                    ca_details.add(ca_detail)

            orphans.extend(ghostbusters.itervalues())
            for ghostbuster in orphans:
                ca_details.add(ghostbuster.ca_detail)
                ghostbuster.revoke(publisher = publisher)

            for ca_detail in ca_details:
                ca_detail.generate_crl_and_manifest(publisher = publisher)

            yield publisher.call_pubd()

        except:
            logger.exception("Could not update Ghostbuster records for %s, skipping", self.tenant.tenant_handle)


@queue_task
class UpdateEECertificatesTask(AbstractTask):
    """
    Generate or update EE certificates for this tenant.

    Not yet sure what kind of scaling constraints this task might have,
    so keeping it simple for initial version, we can optimize later.
    """

    @tornado.gen.coroutine
    def main(self):
        logger.debug("%r: Updating EE certificates", self)

        try:
            r_msg = yield self.rpkid.irdb_query_ee_certificate_requests(self.tenant.tenant_handle)

            publisher = rpki.rpkid.publication_queue(self.rpkid)

            logger.debug("%r: Examining EE certificate requests", self)

            existing = dict()
            for ee in self.tenant.ee_certificates.all():
                gski = ee.gski
                if gski not in existing:
                    existing[gski] = set()
                existing[gski].add(ee)

            ca_details = set()

            for r_pdu in r_msg:
                gski = r_pdu.get("gski")
                ees = existing.pop(gski, ())

                resources = rpki.resource_set.resource_bag(
                    asn         = rpki.resource_set.resource_set_as(r_pdu.get("asn")),
                    v4          = rpki.resource_set.resource_set_ipv4(r_pdu.get("ipv4")),
                    v6          = rpki.resource_set.resource_set_ipv6(r_pdu.get("ipv6")),
                    valid_until = rpki.sundial.datetime.fromXMLtime(r_pdu.get("valid_until")))
                covering = self.tenant.find_covering_ca_details(resources)
                ca_details.update(covering)

                for ee in ees:
                    if ee.ca_detail in covering:
                        logger.debug("%r: Updating %r for %s %s", self, ee, gski, resources)
                        ee.reissue(resources = resources, publisher = publisher)
                        covering.remove(ee.ca_detail)
                    else:
                        # This probably never happens, as the most likely cause would be a CA certificate
                        # being revoked, which should trigger automatic clean up of issued certificates.
                        logger.debug("%r: %r for %s %s is no longer covered", self, ee, gski, resources)
                        ca_details.add(ee.ca_detail)
                        ee.revoke(publisher = publisher)

                subject_name = rpki.x509.X501DN.from_cn(r_pdu.get("cn"), r_pdu.get("sn"))
                subject_key  = rpki.x509.PKCS10(Base64 = r_pdu[0].text).getPublicKey()

                for ca_detail in covering:
                    logger.debug("%r: No existing EE certificate for %s %s", self, gski, resources)
                    cn, sn = subject_name.extract_cn_and_sn()
                    cert = ca_detail.issue_ee(
                        ca          = ca_detail.ca,
                        subject_key = subject_key,
                        sia         = None,
                        resources   = resources,
                        notAfter    = resources.valid_until,
                        cn          = cn,
                        sn          = sn,
                        eku         = r_pdu.get("eku", "").split(",") or None)
                    ee = rpki.rpkidb.models.EECertificate.objects.create(
                        tenant      = ca_detail.ca.parent.tenant,
                        ca_detail   = ca_detail,
                        cert        = cert,
                        gski        = subject_key.gSKI())
                    publisher.queue(
                        uri        = ee.uri,
                        new_obj    = cert,
                        repository = ca_detail.ca.parent.repository,
                        handler    = ee.published_callback)

            # Anything left is an orphan
            for ees in existing.values():
                for ee in ees:
                    ca_details.add(ee.ca_detail)
                    ee.revoke(publisher = publisher)

            for ca_detail in ca_details:
                ca_detail.generate_crl_and_manifest(publisher = publisher)

            yield publisher.call_pubd()

        except:
            logger.exception("%r: Could not update EE certificates, skipping", self)


@queue_task
class RegenerateCRLsAndManifestsTask(AbstractTask):
    """
    Generate new CRLs and manifests as necessary for all of this tenant's
    CAs.  Extracting nextUpdate from a manifest is hard at the moment
    due to implementation silliness, so for now we generate a new
    manifest whenever we generate a new CRL

    This code also cleans up tombstones left behind by revoked ca_detail
    objects, since we're walking through the relevant portions of the
    database anyway.
    """

    @tornado.gen.coroutine
    def main(self):
        logger.debug("%r: Regenerating CRLs and manifests", self)

        try:
            publisher = rpki.rpkid.publication_queue(self.rpkid)
            now = rpki.sundial.now()

            ca_details = rpki.rpkidb.models.CADetail.objects.filter(ca__parent__tenant = self.tenant,
                                                                    next_crl_manifest_update__isnull = False)

            for ca_detail in ca_details.filter(next_crl_manifest_update__lt = now,
                                               state = "revoked"):
                ca_detail.destroy(publisher = publisher)

            for ca_detail in ca_details.filter(state__in = ("active", "deprecated"),
                                               next_crl_manifest_update__lt = now + max(
                                                   rpki.sundial.timedelta(seconds = self.tenant.crl_interval) / 4,
                                                   rpki.sundial.timedelta(seconds = self.rpkid.cron_period  ) * 2)):
                ca_detail.generate_crl_and_manifest(publisher = publisher)

            yield publisher.call_pubd()

        except:
            logger.exception("%r: Couldn't publish updated CRLs and manifests, skipping", self)


@queue_task
class CheckFailedPublication(AbstractTask):
    """
    Periodic check for objects we tried to publish but failed (eg, due
    to pubd being down or unreachable).
    """

    @tornado.gen.coroutine
    def main(self):
        logger.debug("%r: Checking for failed publication actions", self)

        try:
            publisher = rpki.rpkid.publication_queue(self.rpkid)
            for ca_detail in rpki.rpkidb.models.CADetail.objects.filter(ca__parent__tenant = self.tenant, state = "active"):
                ca_detail.check_failed_publication(publisher)
            yield publisher.call_pubd()

        except:
            logger.exception("%r: Couldn't run failed publications, skipping", self)