bin/named/client.c

	client.c revision d468db05b3aea6d745ee188258e59da11510d637
/*
 * Copyright (C) 1999  Internet Software Consortium.
 *
 * Permission to use, copy, modify, and distribute this software for any
 * purpose with or without fee is hereby granted, provided that the above
 * copyright notice and this permission notice appear in all copies.
 *
 * THE SOFTWARE IS PROVIDED "AS IS" AND INTERNET SOFTWARE CONSORTIUM DISCLAIMS
 * ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL INTERNET SOFTWARE
 * CONSORTIUM 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.
 */

#include <config.h>

#include <isc/assertions.h>
#include <isc/mem.h>
#include <isc/mutex.h>
#include <isc/result.h>
#include <isc/task.h>
#include <isc/timer.h>
#include <isc/util.h>

#include <dns/acl.h>
#include <dns/dispatch.h>
#include <dns/events.h>
#include <dns/message.h>
#include <dns/rdata.h>
#include <dns/rdatalist.h>
#include <dns/rdataset.h>
#include <dns/view.h>
#include <dns/xfrin.h>

#include <named/globals.h>
#include <named/client.h>
#include <named/log.h>
#include <named/query.h>
#include <named/server.h>
#include <named/update.h>
#include <named/notify.h>
#include <named/interfacemgr.h>

#define NS_CLIENT_TRACE
#ifdef NS_CLIENT_TRACE
#define CTRACE(m)   isc_log_write(ns_g_lctx, \
                      NS_LOGCATEGORY_CLIENT, \
                      NS_LOGMODULE_CLIENT, \
                      ISC_LOG_DEBUG(3), \
                      "client %p: %s", client, (m))
#define MTRACE(m)   isc_log_write(ns_g_lctx, \
                      NS_LOGCATEGORY_GENERAL, \
                      NS_LOGMODULE_CLIENT, \
                      ISC_LOG_DEBUG(3), \
                      "clientmgr %p: %s", manager, (m))
#endif

#define TCP_CLIENT(c)   (((c)->attributes & NS_CLIENTATTR_TCP) != 0)

#define SEND_BUFFER_SIZE        2048

struct ns_clientmgr {
    /* Unlocked. */
    unsigned int            magic;
    isc_mem_t *         mctx;
    isc_taskmgr_t *         taskmgr;
    isc_timermgr_t *        timermgr;
    isc_mutex_t         lock;
    /* Locked by lock. */
    isc_boolean_t           exiting;
    unsigned int            nclients;
    ISC_LIST(ns_client_t)       clients;
};

#define MANAGER_MAGIC           0x4E53436DU /* NSCm */
#define VALID_MANAGER(m)        ((m) != NULL && \
                     (m)->magic == MANAGER_MAGIC)


static void clientmgr_destroy(ns_clientmgr_t *manager);


/***
 *** Client
 ***/

/*
 * Important note!
 *
 * All client state changes, other than that from idle to listening, occur
 * as a result of events.  This guarantees serialization and avoids the
 * need for locking.
 *
 * If a routine is ever created that allows someone other than the client's
 * task to change the client, then the client will have to be locked.
 */

static void
release_quotas(ns_client_t *client) {
    if (client->tcpquota != NULL)
        isc_quota_detach(&client->tcpquota);
    if (client->recursionquota != NULL)
        isc_quota_detach(&client->recursionquota);
}

/*
 * Free a client immediately if possible, otherwise start
 * shutting it down and postpone freeing to later.
 */
static void
maybe_free(ns_client_t *client) {
    isc_boolean_t need_clientmgr_destroy = ISC_FALSE;
    ns_clientmgr_t *manager;

    REQUIRE(NS_CLIENT_VALID(client));

    /*
     * When "shuttingdown" is true, either the task has received
     * its shutdown event or no shutdown event has ever been
     * set up.  Thus, we have no outstanding shutdown
     * event at this point.
     */
    REQUIRE(client->shuttingdown == ISC_TRUE);

    if (client->naccepts > 0)
        isc_socket_cancel(client->tcplistener, client->task,
                  ISC_SOCKCANCEL_ACCEPT);
    if (client->nreads > 0)
        dns_tcpmsg_cancelread(&client->tcpmsg);
    if (client->nsends > 0) {
        isc_socket_t *socket;
        if (TCP_CLIENT(client))
            socket = client->tcpsocket;
        else
            socket = dns_dispatch_getsocket(client->dispatch);
        isc_socket_cancel(client->tcpsocket, client->task,
                  ISC_SOCKCANCEL_SEND);
    }

    if (!(client->nreads == 0 && client->naccepts == 0 &&
          client->nsends == 0 && client->nwaiting == 0)) {
        /* Still waiting for events. */
        return;
    }

    /* We have received our last event. */

    ns_interface_detach(&client->interface);
    ns_query_free(client);
    isc_mempool_destroy(&client->sendbufs);
    isc_timer_detach(&client->timer);

    if (client->dispentry != NULL) {
        dns_dispatchevent_t **deventp;
        if (client->dispevent != NULL)
            deventp = &client->dispevent;
        else
            deventp = NULL;
        dns_dispatch_removerequest(client->dispatch,
                       &client->dispentry,
                       deventp);
    }
    if (client->view != NULL)
        dns_view_detach(&client->view);
    if (client->opt != NULL) {
        INSIST(dns_rdataset_isassociated(client->opt));
        dns_rdataset_disassociate(client->opt);
        dns_message_puttemprdataset(client->message, &client->opt);
    }
    dns_message_destroy(&client->message);
    if (client->tcpmsg_valid)
        dns_tcpmsg_invalidate(&client->tcpmsg);
    if (client->dispatch != NULL)
        dns_dispatch_detach(&client->dispatch);
    if (client->tcpsocket != NULL)
        isc_socket_detach(&client->tcpsocket);
    if (client->tcplistener != NULL)
        isc_socket_detach(&client->tcplistener);
    if (client->task != NULL)
        isc_task_detach(&client->task);
    if (client->manager != NULL) {
        manager = client->manager;
        LOCK(&manager->lock);

        INSIST(manager->nclients > 0);
        manager->nclients--;
        if (manager->nclients == 0 && manager->exiting)
                need_clientmgr_destroy = ISC_TRUE;
        ISC_LIST_UNLINK(manager->clients, client, link);

        UNLOCK(&manager->lock);
    }

    release_quotas(client);

    CTRACE("free");
    client->magic = 0;
    isc_mem_put(client->mctx, client, sizeof *client);

    if (need_clientmgr_destroy)
        clientmgr_destroy(manager);
}

/*
 * The client's task has received a shutdown event.
 */
static void
client_shutdown(isc_task_t *task, isc_event_t *event) {
    ns_client_t *client;

    REQUIRE(event != NULL);
    REQUIRE(event->type == ISC_TASKEVENT_SHUTDOWN);
    client = event->arg;
    REQUIRE(NS_CLIENT_VALID(client));
    REQUIRE(task == client->task);

    CTRACE("shutdown");

    client->shuttingdown = ISC_TRUE;

    if (client->shutdown != NULL)
        (client->shutdown)(client->shutdown_arg);

    maybe_free(client);

    isc_event_free(&event);
}

static void client_read(ns_client_t *client);
static void client_accept(ns_client_t *client);

void
ns_client_next(ns_client_t *client, isc_result_t result) {

    REQUIRE(NS_CLIENT_VALID(client));

    CTRACE("next");

    if (client->next != NULL) {
        (client->next)(client, result);
        client->next = NULL;
    }

    /*
     * XXXRTH  If result != ISC_R_SUCCESS:
     *      Log result if there is interest in doing so.
     */

    if (client->view != NULL)
        dns_view_detach(&client->view);
    if (client->opt != NULL) {
        INSIST(dns_rdataset_isassociated(client->opt));
        dns_rdataset_disassociate(client->opt);
        dns_message_puttemprdataset(client->message, &client->opt);
    }

    client->udpsize = 512;
    dns_message_reset(client->message, DNS_MESSAGE_INTENTPARSE);

    release_quotas(client);

    if (client->mortal) {
        /*
         * This client object is supposed to die now, but if we
         * have fewer client objects than planned due to
         * quota exhaustion, don't.
         */
        isc_boolean_t need_another_client = ISC_FALSE;
        if (TCP_CLIENT(client)) {
            LOCK(&client->interface->lock);
            if (client->interface->ntcpcurrent <
                client->interface->ntcptarget)
                need_another_client = ISC_TRUE;
            UNLOCK(&client->interface->lock);
        } else {
            /*
             * The UDP client quota is enforced by making
             * requests fail rather than by not listening
             * for new ones.  Therefore, there is always a
             * full set of UDP clients listening.
             */
        }
        if (! need_another_client) {
            /* XXX should put in "idle client pool" instead. */
            isc_task_shutdown(client->task);
            return;
        }
        client->mortal = ISC_FALSE;
    }

    if (client->dispevent != NULL) {
        /*
         * Give the processed dispatch event back to the dispatch.
         * This tells the dispatch that we are ready to receive
         * the next event.
         */
        dns_dispatch_freeevent(client->dispatch, client->dispentry,
                       &client->dispevent);
    } else if (TCP_CLIENT(client)) {
        if (result == ISC_R_SUCCESS) {
            client_read(client);
        } else {
            /*
             * There was an error processing a TCP request.
             * It may have have left the connection out of
             * sync.  Close the connection and listen for a
             * new one.
             */
            if (client->tcpsocket != NULL) {
                /*
                 * There should be no outstanding read
                 * request on the TCP socket at this point,
                 * therefore invalidating the tcpmsg is safe.
                 */
                INSIST(client->nreads == 0);
                INSIST(client->tcpmsg_valid == ISC_TRUE);
                dns_tcpmsg_invalidate(&client->tcpmsg);
                client->tcpmsg_valid = ISC_FALSE;
                isc_socket_detach(&client->tcpsocket);
            }
            client_accept(client);
        }
    }
}

static void
client_senddone(isc_task_t *task, isc_event_t *event) {
    ns_client_t *client;
    isc_socketevent_t *sevent = (isc_socketevent_t *)event;

    REQUIRE(sevent != NULL);
    REQUIRE(sevent->type == ISC_SOCKEVENT_SENDDONE);
    client = sevent->arg;
    REQUIRE(NS_CLIENT_VALID(client));
    REQUIRE(task == client->task);

    CTRACE("senddone");

    INSIST(client->nsends > 0);
    client->nsends--;
    isc_mempool_put(client->sendbufs, sevent->region.base);

    isc_event_free(&event);

    /*
     * If all of its sendbufs buffers were busy, the client might be
     * waiting for one to become available.
     */
    if (client->waiting_for_bufs == ISC_TRUE) {
        client->waiting_for_bufs = ISC_FALSE;
        ns_client_send(client);
        return;
    }
    /* XXXRTH need to add exit draining mode. */
}

void
ns_client_send(ns_client_t *client) {
    isc_result_t result;
    unsigned char *data;
    isc_buffer_t buffer;
    isc_buffer_t tcpbuffer;
    isc_region_t r;
    isc_socket_t *socket;
    isc_sockaddr_t *address;
    unsigned int bufsize = 512;

    REQUIRE(NS_CLIENT_VALID(client));

    CTRACE("send");

    if ((client->attributes & NS_CLIENTATTR_RA) != 0)
        client->message->flags |= DNS_MESSAGEFLAG_RA;

    data = isc_mempool_get(client->sendbufs);
    if (data == NULL) {
        CTRACE("no buffers available");
        if (client->nsends > 0) {
            /*
             * We couldn't get memory, but there is at least one
             * send outstanding.  We arrange to be restarted when a
             * send completes.
             */
            CTRACE("waiting");
            client->waiting_for_bufs = ISC_TRUE;
        } else
            ns_client_next(client, ISC_R_NOMEMORY);
        return;
    }

    /*
     * XXXRTH  The following doesn't deal with TSIGs, TCP buffer resizing,
     *         or ENDS1 more data packets.
     */
    if (TCP_CLIENT(client)) {
        /*
         * XXXRTH  "tcpbuffer" is a hack to get things working.
         */
        isc_buffer_init(&tcpbuffer, data, SEND_BUFFER_SIZE,
                ISC_BUFFERTYPE_BINARY);
        isc_buffer_init(&buffer, data + 2, SEND_BUFFER_SIZE - 2,
                ISC_BUFFERTYPE_BINARY);
    } else {
        if (client->udpsize < SEND_BUFFER_SIZE)
            bufsize = client->udpsize;
        else
            bufsize = SEND_BUFFER_SIZE;
        isc_buffer_init(&buffer, data, bufsize, ISC_BUFFERTYPE_BINARY);
    }

    result = dns_message_renderbegin(client->message, &buffer);
    if (result != ISC_R_SUCCESS)
        goto done;
    if (client->opt != NULL) {
        result = dns_message_setopt(client->message, client->opt);
        if (result != ISC_R_SUCCESS)
            goto done;
        /*
         * XXXRTH dns_message_setopt() should probably do this...
         */
        client->opt = NULL;
    }
    result = dns_message_rendersection(client->message,
                       DNS_SECTION_QUESTION, 0);
    if (result != ISC_R_SUCCESS)
        goto done;
    result = dns_message_rendersection(client->message,
                       DNS_SECTION_ANSWER, 0);
    if (result == ISC_R_NOSPACE) {
        client->message->flags |= DNS_MESSAGEFLAG_TC;
        goto renderend;
    }
    if (result != ISC_R_SUCCESS)
        goto done;
    result = dns_message_rendersection(client->message,
                       DNS_SECTION_AUTHORITY, 0);
    if (result == ISC_R_NOSPACE) {
        client->message->flags |= DNS_MESSAGEFLAG_TC;
        goto renderend;
    }
    if (result != ISC_R_SUCCESS)
        goto done;
    result = dns_message_rendersection(client->message,
                       DNS_SECTION_ADDITIONAL, 0);
    if (result != ISC_R_SUCCESS && result != ISC_R_NOSPACE)
        goto done;
 renderend:
    result = dns_message_renderend(client->message);
    if (result != ISC_R_SUCCESS)
        goto done;

    if (TCP_CLIENT(client)) {
        socket = client->tcpsocket;
        address = NULL;
        isc_buffer_used(&buffer, &r);
        isc_buffer_putuint16(&tcpbuffer, (isc_uint16_t)r.length);
        isc_buffer_add(&tcpbuffer, r.length);
        isc_buffer_used(&tcpbuffer, &r);
    } else {
        socket = dns_dispatch_getsocket(client->dispatch);
        address = &client->dispevent->addr;
        isc_buffer_used(&buffer, &r);
    }
    CTRACE("sendto");
    result = isc_socket_sendto(socket, &r, client->task, client_senddone,
                   client, address, NULL);
    if (result == ISC_R_SUCCESS)
        client->nsends++;

 done:
    if (result != ISC_R_SUCCESS)
        isc_mempool_put(client->sendbufs, data);

    ns_client_next(client, result);
}

void
ns_client_error(ns_client_t *client, isc_result_t result) {
    dns_rcode_t rcode;
    dns_message_t *message;

    REQUIRE(NS_CLIENT_VALID(client));

    CTRACE("error");

    message = client->message;
    rcode = dns_result_torcode(result);

    /*
     * message may be an in-progress reply that we had trouble
     * with, in which case QR will be set.  We need to clear QR before
     * calling dns_message_reply() to avoid triggering an assertion.
     */
    message->flags &= ~DNS_MESSAGEFLAG_QR;
    /*
     * AA and AD shouldn't be set.
     */
    message->flags &= ~(DNS_MESSAGEFLAG_AA | DNS_MESSAGEFLAG_AD);
    result = dns_message_reply(message, ISC_TRUE);
    if (result != ISC_R_SUCCESS) {
        /*
         * It could be that we've got a query with a good header,
         * but a bad question section, so we try again with
         * want_question_section set to ISC_FALSE.
         */
        result = dns_message_reply(message, ISC_FALSE);
        if (result != ISC_R_SUCCESS) {
            /*
             * There's no hope of replying to this request.
             *
             * XXXRTH  Mark this client to that if it is a
             * TCP session, the session will be closed.
             */
            ns_client_next(client, result);
            return;
        }
    }
    message->rcode = rcode;
    ns_client_send(client);
}

static inline isc_result_t
client_addopt(ns_client_t *client) {
    dns_rdataset_t *rdataset;
    dns_rdatalist_t *rdatalist;
    dns_rdata_t *rdata;
    isc_result_t result;

    REQUIRE(client->opt == NULL);   /* XXXRTH free old. */

    rdatalist = NULL;
    result = dns_message_gettemprdatalist(client->message, &rdatalist);
    if (result != ISC_R_SUCCESS)
        return (result);
    rdata = NULL;
    result = dns_message_gettemprdata(client->message, &rdata);
    if (result != ISC_R_SUCCESS)
        return (result);
    rdataset = NULL;
    result = dns_message_gettemprdataset(client->message, &rdataset);
    if (result != ISC_R_SUCCESS)
        return (result);
    dns_rdataset_init(rdataset);

    rdatalist->type = dns_rdatatype_opt;
    rdatalist->covers = 0;

    /*
     * Set Maximum UDP buffer size.
     */
    rdatalist->rdclass = SEND_BUFFER_SIZE;

    /*
     * Set EXTENDED-RCODE, VERSION, and Z to 0.
     */
    rdatalist->ttl = 0;

    /*
     * No ENDS options.
     */
    rdata->data = NULL;
    rdata->length = 0;

    ISC_LIST_INIT(rdatalist->rdata);
    ISC_LIST_APPEND(rdatalist->rdata, rdata, link);
    dns_rdatalist_tordataset(rdatalist, rdataset);

    client->opt = rdataset;

    return (ISC_R_SUCCESS);
}

/*
 * Handle an incoming request event from the dispatch (UDP case)
 * or tcpmsg (TCP case).
 */
static void
client_request(isc_task_t *task, isc_event_t *event) {
    ns_client_t *client;
    dns_dispatchevent_t *devent;
    isc_result_t result;
    isc_buffer_t *buffer;
    dns_view_t *view;
    dns_rdataset_t *opt;
    isc_boolean_t ra;   /* Recursion available. */

    REQUIRE(event != NULL);
    client = event->arg;
    REQUIRE(NS_CLIENT_VALID(client));
    REQUIRE(task == client->task);

    INSIST(client->recursionquota == NULL);

    if (event->type == DNS_EVENT_DISPATCH) {
        devent = (dns_dispatchevent_t *)event;
        REQUIRE(client->dispentry != NULL);
        client->dispevent = devent;
        buffer = &devent->buffer;
        result = devent->result;
    } else {
        REQUIRE(event->type == DNS_EVENT_TCPMSG);
        REQUIRE(event->sender == &client->tcpmsg);
        buffer = &client->tcpmsg.buffer;
        result = client->tcpmsg.result;
        INSIST(client->nreads == 1);
        client->nreads--;
    }

    CTRACE("request");

    if (client->shuttingdown) {
        maybe_free(client);
        return;
    }

    isc_stdtime_get(&client->requesttime);
    client->now = client->requesttime;

    if (result != ISC_R_SUCCESS) {
        if (TCP_CLIENT(client))
            ns_client_next(client, result);
        else
            isc_task_shutdown(client->task);
        return;
    }

    result = dns_message_parse(client->message, buffer, ISC_FALSE);
    if (result != ISC_R_SUCCESS) {
        ns_client_error(client, result);
        return;
    }

    /*
     * We expect a query, not a response.  Unexpected UDP responses
     * are discarded early by the dispatcher, but TCP responses
     * bypass the dispatcher and must be discarded here.
     */
    if ((client->message->flags & DNS_MESSAGEFLAG_QR) != 0) {
        CTRACE("unexpected response");
        ns_client_next(client, DNS_R_FORMERR);
        return;
    }

    /*
     * Deal with EDNS.
     */
    opt = dns_message_getopt(client->message);
    if (opt != NULL) {
        unsigned int version;

        /*
         * Set the client's UDP buffer size.
         */
        client->udpsize = opt->rdclass;

        /*
         * Create an OPT for our reply.
         */
        result = client_addopt(client);
        if (result != ISC_R_SUCCESS) {
            ns_client_error(client, result);
            return;
        }

        /*
         * Do we understand this version of ENDS?
         *
         * XXXRTH need library support for this!
         */
        version = (opt->ttl & 0x00FF0000) >> 16;
        if (version != 0) {
            ns_client_error(client, DNS_R_BADVERS);
            return;
        }
    }

    /*
     * Check for a signature.  We log bad signatures regardless of
     * whether they ultimately cause the request to be rejected or
     * not.  We do not log the lack of a signature unless we are
     * debugging.
     */
    client->signer = NULL;
    result = dns_message_signer(client->message, &client->signername);
    if (result == DNS_R_SUCCESS) {
        isc_log_write(dns_lctx, DNS_LOGCATEGORY_SECURITY,
                  NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
                  "request has valid signature");
        client->signer = &client->signername;
    } else if (result == DNS_R_NOTFOUND) {
        isc_log_write(dns_lctx, DNS_LOGCATEGORY_SECURITY,
                  NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
                  "request is not signed");
    } else {
        /* There is a signature, but it is bad. */
        isc_log_write(dns_lctx, DNS_LOGCATEGORY_SECURITY,
                  NS_LOGMODULE_CLIENT, ISC_LOG_ERROR,
                  "request has invalid signature: %s",
                  isc_result_totext(result));
    }

    /*
     * XXXRTH  View list management code will be moving to its own module
     *         soon.
     */
    RWLOCK(&ns_g_server->viewlock, isc_rwlocktype_read);
    for (view = ISC_LIST_HEAD(ns_g_server->viewlist);
         view != NULL;
         view = ISC_LIST_NEXT(view, link)) {
        /*
         * XXXRTH  View matching will become more powerful later.
         */
        if (client->message->rdclass == view->rdclass) {
            dns_view_attach(view, &client->view);
            break;
        }
    }
    RWUNLOCK(&ns_g_server->viewlock, isc_rwlocktype_read);

    if (view == NULL) {
        CTRACE("no view");
        ns_client_error(client, DNS_R_REFUSED);
        return;
    }

    /*
     * Decide whether recursive service is available to this client.
     * We do this here rather than in the query code so that we can
     * set the RA bit correctly on all kinds of responses, not just
     * responses to ordinary queries.
     */
    if (client->view->resolver == NULL) {
        ra = ISC_FALSE;
    } else {
        ra = ISC_TRUE;
        if (ns_g_server->recursion == ISC_TRUE) {
            /* XXX ACL should be view specific. */
            /* XXX this will log too much too early */
            result = dns_acl_checkrequest(client->signer,
                          ns_client_getsockaddr(client),
                          "recursion",
                          ns_g_server->recursionacl,
                          NULL, ISC_TRUE);
            if (result != DNS_R_SUCCESS)
                ra = ISC_FALSE;
        }
    }
    if (ra == ISC_TRUE)
        client->attributes |= NS_CLIENTATTR_RA;

    /*
     * Dispatch the request.
     */
    switch (client->message->opcode) {
    case dns_opcode_query:
        CTRACE("query");
        ns_query_start(client);
        break;
    case dns_opcode_update:
        CTRACE("update");
        ns_update_start(client);
        break;
    case dns_opcode_notify:
        CTRACE("notify");
        ns_notify_start(client);
        break;
    case dns_opcode_iquery:
        CTRACE("iquery");
        ns_client_error(client, DNS_R_NOTIMP);
    default:
        CTRACE("unknown opcode");
        ns_client_error(client, DNS_R_NOTIMP);
    }
}

static void
client_timeout(isc_task_t *task, isc_event_t *event) {
    ns_client_t *client;

    REQUIRE(event != NULL);
    REQUIRE(event->type == ISC_TIMEREVENT_LIFE ||
        event->type == ISC_TIMEREVENT_IDLE);
    client = event->arg;
    REQUIRE(NS_CLIENT_VALID(client));
    REQUIRE(task == client->task);
    REQUIRE(client->timer != NULL);

    CTRACE("timeout");

    isc_event_free(&event);

    ns_client_next(client, ISC_R_TIMEDOUT);
}

static isc_result_t
client_create(ns_clientmgr_t *manager,
          ns_interface_t *ifp, ns_client_t **clientp)
{
    ns_client_t *client;
    isc_result_t result;

    /*
     * Caller must be holding the manager lock.
     *
     * Note: creating a client does not add the client to the
     * manager's client list or set the client's manager pointer.
     * The caller is responsible for that.
     */

    REQUIRE(clientp != NULL && *clientp == NULL);

    client = isc_mem_get(manager->mctx, sizeof *client);
    if (client == NULL)
        return (ISC_R_NOMEMORY);

    client->task = NULL;
    result = isc_task_create(manager->taskmgr, manager->mctx, 0,
                 &client->task);
    if (result != ISC_R_SUCCESS)
        goto cleanup_client;
    result = isc_task_onshutdown(client->task, client_shutdown, client);
    if (result != ISC_R_SUCCESS)
        goto cleanup_task;

    client->timer = NULL;
    result = isc_timer_create(manager->timermgr, isc_timertype_inactive,
                  NULL, NULL, client->task, client_timeout,
                  client, &client->timer);
    if (result != ISC_R_SUCCESS)
        goto cleanup_task;

    client->message = NULL;
    result = dns_message_create(manager->mctx, DNS_MESSAGE_INTENTPARSE,
                    &client->message);
    if (result != ISC_R_SUCCESS)
        goto cleanup_timer;

    /* XXXRTH  Hardwired constants */
    client->sendbufs = NULL;
    result = isc_mempool_create(manager->mctx, SEND_BUFFER_SIZE,
                    &client->sendbufs);
    if (result != ISC_R_SUCCESS)
        goto cleanup_message;
    isc_mempool_setfreemax(client->sendbufs, 3);
    isc_mempool_setmaxalloc(client->sendbufs, 3);

    client->magic = NS_CLIENT_MAGIC;
    client->mctx = manager->mctx;
    client->manager = NULL;
    client->shuttingdown = ISC_FALSE;
    client->waiting_for_bufs = ISC_FALSE;
    client->naccepts = 0;
    client->nreads = 0;
    client->nsends = 0;
    client->nwaiting = 0;
    client->attributes = 0;
    client->view = NULL;
    client->dispatch = NULL;
    client->dispentry = NULL;
    client->dispevent = NULL;
    client->tcplistener = NULL;
    client->tcpsocket = NULL;
    client->tcpmsg_valid = ISC_FALSE;
    client->opt = NULL;
    client->udpsize = 512;
    client->next = NULL;
    client->shutdown = NULL;
    client->shutdown_arg = NULL;
    dns_name_init(&client->signername, NULL);
    client->mortal = ISC_FALSE;
    client->tcpquota = NULL;
    client->recursionquota = NULL;
    client->interface = NULL;
    ISC_LINK_INIT(client, link);

    /*
     * We call the init routines for the various kinds of client here,
     * after we have created an otherwise valid client, because some
     * of them call routines that REQUIRE(NS_CLIENT_VALID(client)).
     */
    result = ns_query_init(client);
    if (result != ISC_R_SUCCESS)
        goto cleanup_sendbufs;

    ns_interface_attach(ifp, &client->interface);

    CTRACE("create");

    *clientp = client;

    return (ISC_R_SUCCESS);

 cleanup_sendbufs:
    isc_mempool_destroy(&client->sendbufs);

    client->magic = 0;

 cleanup_message:
    dns_message_destroy(&client->message);

 cleanup_timer:
    isc_timer_detach(&client->timer);

 cleanup_task:
    isc_task_detach(&client->task);

 cleanup_client:
    isc_mem_put(manager->mctx, client, sizeof *client);

    return (result);
}

static void
client_read(ns_client_t *client) {
    isc_result_t result;

    CTRACE("read");

    result = dns_tcpmsg_readmessage(&client->tcpmsg, client->task,
                    client_request, client);
    if (result != ISC_R_SUCCESS)
        ns_client_next(client, result);
    INSIST(client->nreads == 0);
    client->nreads++;
}

static void
client_newconn(isc_task_t *task, isc_event_t *event) {
    ns_client_t *client = event->arg;
    isc_socket_newconnev_t *nevent = (isc_socket_newconnev_t *)event;
    isc_result_t result;

    REQUIRE(event->type == ISC_SOCKEVENT_NEWCONN);
    REQUIRE(NS_CLIENT_VALID(client));
    REQUIRE(client->task == task);

    CTRACE("newconn");

    INSIST(client->naccepts == 1);
    client->naccepts--;

    LOCK(&client->interface->lock);
    INSIST(client->interface->ntcpcurrent > 0);
    client->interface->ntcpcurrent--;
    UNLOCK(&client->interface->lock);

    if (client->shuttingdown) {
        maybe_free(client);
    } else if (nevent->result == ISC_R_SUCCESS) {
        client->tcpsocket = nevent->newsocket;
        INSIST(client->tcpmsg_valid == ISC_FALSE);
        dns_tcpmsg_init(client->mctx, client->tcpsocket,
                &client->tcpmsg);
        client->tcpmsg_valid = ISC_TRUE;

        /*
         * Let a new client take our place immediately, before
         * we wait for a request packet.  If we don't,
         * telnetting to port 35 (once per CPU) will
         * deny service to legititmate TCP clients.
         */
        result = isc_quota_attach(&ns_g_server->tcpquota,
                      &client->tcpquota);
        if (result == ISC_R_SUCCESS)
            result = ns_client_replace(client);
        if (result != ISC_R_SUCCESS) {
            isc_log_write(ns_g_lctx, NS_LOGCATEGORY_CLIENT,
                      NS_LOGMODULE_CLIENT, ISC_LOG_WARNING,
                      "no more TCP clients: %s",
                      isc_result_totext(result));
        }
        client_read(client);
    } else {
        /*
         * XXXRTH  What should we do?  We're trying to accept but
         *         it didn't work.  If we just give up, then TCP
         *     service may eventually stop.
         *
         *     For now, we just go idle.
         *
         *     Going idle is probably the right thing if the
         *     I/O was canceled.
         */
    }
    isc_event_free(&event);
}

static void
client_accept(ns_client_t *client) {
    isc_result_t result;

    CTRACE("accept");

    result = isc_socket_accept(client->tcplistener, client->task,
                   client_newconn, client);
    if (result != ISC_R_SUCCESS) {
        UNEXPECTED_ERROR(__FILE__, __LINE__,
                 "isc_socket_accept() failed: %s",
                 isc_result_totext(result));
        /*
         * XXXRTH  What should we do?  We're trying to accept but
         *         it didn't work.  If we just give up, then TCP
         *     service may eventually stop.
         *
         *     For now, we just go idle.
         */
        return;
    }
    INSIST(client->naccepts == 0);
    client->naccepts++;
    LOCK(&client->interface->lock);
    client->interface->ntcpcurrent++;
    UNLOCK(&client->interface->lock);
}

void
ns_client_wait(ns_client_t *client) {
    client->nwaiting++;
}

isc_boolean_t
ns_client_shuttingdown(ns_client_t *client) {
    return (client->shuttingdown);
}

void
ns_client_unwait(ns_client_t *client) {
    client->nwaiting--;
    INSIST(client->nwaiting >= 0);
    if (client->shuttingdown)
        maybe_free(client);
}

isc_result_t
ns_client_replace(ns_client_t *client) {
    isc_result_t result;
    CTRACE("replace");

    result = ns_clientmgr_createclients(client->manager,
                        1, client->interface,
                        (TCP_CLIENT(client) ?
                         ISC_TRUE : ISC_FALSE));
    if (result != ISC_R_SUCCESS)
        return (result);

    /*
     * The responsibility for listening for new requests is hereby
     * transferred to the new client.  Therefore, the old client
     * should refrain from listening for any more requests.
     */
    client->mortal = ISC_TRUE;

    return (ISC_R_SUCCESS);
}

/***
 *** Client Manager
 ***/

static void
clientmgr_destroy(ns_clientmgr_t *manager) {
    REQUIRE(manager->nclients == 0);
    REQUIRE(ISC_LIST_EMPTY(manager->clients));

    MTRACE("clientmgr_destroy");

    manager->magic = 0;
    isc_mem_put(manager->mctx, manager, sizeof *manager);
}

isc_result_t
ns_clientmgr_create(isc_mem_t *mctx, isc_taskmgr_t *taskmgr,
            isc_timermgr_t *timermgr, ns_clientmgr_t **managerp)
{
    ns_clientmgr_t *manager;
    isc_result_t result;

    manager = isc_mem_get(mctx, sizeof *manager);
    if (manager == NULL)
        return (ISC_R_NOMEMORY);

    result = isc_mutex_init(&manager->lock);
    if (result != ISC_R_SUCCESS)
        goto cleanup_manager;

    manager->mctx = mctx;
    manager->taskmgr = taskmgr;
    manager->timermgr = timermgr;
    manager->exiting = ISC_FALSE;
    manager->nclients = 0;
    ISC_LIST_INIT(manager->clients);
    manager->magic = MANAGER_MAGIC;

    MTRACE("create");

    *managerp = manager;

    return (ISC_R_SUCCESS);

 cleanup_manager:
    isc_mem_put(manager->mctx, manager, sizeof *manager);

    return (result);
}

void
ns_clientmgr_destroy(ns_clientmgr_t **managerp) {
    ns_clientmgr_t *manager;
    ns_client_t *client;
    isc_boolean_t need_destroy = ISC_FALSE;

    REQUIRE(managerp != NULL);
    manager = *managerp;
    REQUIRE(VALID_MANAGER(manager));

    MTRACE("destroy");

    LOCK(&manager->lock);

    manager->exiting = ISC_TRUE;

    for (client = ISC_LIST_HEAD(manager->clients);
         client != NULL;
         client = ISC_LIST_NEXT(client, link))
        isc_task_shutdown(client->task);

    if (ISC_LIST_EMPTY(manager->clients))
        need_destroy = ISC_TRUE;

    UNLOCK(&manager->lock);

    if (need_destroy)
        clientmgr_destroy(manager);

    *managerp = NULL;
}

isc_result_t
ns_clientmgr_createclients(ns_clientmgr_t *manager, unsigned int n,
               ns_interface_t *ifp, isc_boolean_t tcp)
{
    isc_result_t result = ISC_R_SUCCESS;
    unsigned int i;
    ns_client_t *client;

    REQUIRE(VALID_MANAGER(manager));
    REQUIRE(n > 0);

    MTRACE("createclients");

    /*
     * We MUST lock the manager lock for the entire client creation
     * process.  If we didn't do this, then a client could get a
     * shutdown event and disappear out from under us.
     */

    LOCK(&manager->lock);

    for (i = 0; i < n; i++) {
        client = NULL;
        result = client_create(manager, ifp, &client);
        if (result != ISC_R_SUCCESS)
            break;
        if (tcp) {
            client->attributes |= NS_CLIENTATTR_TCP;
            isc_socket_attach(ifp->tcpsocket, &client->tcplistener);
            client_accept(client);
        } else {
            dns_dispatch_attach(ifp->udpdispatch, &client->dispatch);
            result = dns_dispatch_addrequest(client->dispatch,
                             client->task,
                             client_request,
                             client, &client->dispentry);
            if (result != ISC_R_SUCCESS) {
                isc_log_write(dns_lctx, DNS_LOGCATEGORY_SECURITY,
                          NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
                          "dns_dispatch_addrequest() failed: %s",
                          isc_result_totext(result));
                isc_task_shutdown(client->task);
                break;
            }
        }
        client->manager = manager;
        ISC_LIST_APPEND(manager->clients, client, link);
        manager->nclients++;
    }
    if (i != 0) {
        /*
         * We managed to create at least one client, so we
         * declare victory.
         */
        result = ISC_R_SUCCESS;
    }

    UNLOCK(&manager->lock);

    return (result);
}

isc_sockaddr_t *
ns_client_getsockaddr(ns_client_t *client) {
    if (TCP_CLIENT(client))
        return (&client->tcpmsg.address);
    else
        return (&client->dispevent->addr);
}