lib/dns/acache.c

	acache.c revision 70e5a7403f0e0a3bd292b8287c5fed5772c15270
/*
 * Copyright (C) 2004-2007  Internet Systems Consortium, Inc. ("ISC")
 *
 * Permission to use, copy, modify, and/or 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 ISC DISCLAIMS ALL WARRANTIES WITH
 * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
 * AND FITNESS.  IN NO EVENT SHALL 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.
 */

/* $Id: acache.c,v 1.20 2007/06/19 23:47:16 tbox Exp $ */

#include <config.h>

#include <isc/atomic.h>
#include <isc/event.h>
#include <isc/hash.h>
#include <isc/magic.h>
#include <isc/mem.h>
#include <isc/mutex.h>
#include <isc/random.h>
#include <isc/refcount.h>
#include <isc/rwlock.h>
#include <isc/task.h>
#include <isc/time.h>
#include <isc/timer.h>

#include <dns/acache.h>
#include <dns/db.h>
#include <dns/events.h>
#include <dns/log.h>
#include <dns/message.h>
#include <dns/name.h>
#include <dns/rdataset.h>
#include <dns/result.h>
#include <dns/zone.h>

#define ACACHE_MAGIC            ISC_MAGIC('A', 'C', 'H', 'E')
#define DNS_ACACHE_VALID(acache)    ISC_MAGIC_VALID(acache, ACACHE_MAGIC)

#define ACACHEENTRY_MAGIC       ISC_MAGIC('A', 'C', 'E', 'T')
#define DNS_ACACHEENTRY_VALID(entry)    ISC_MAGIC_VALID(entry, ACACHEENTRY_MAGIC)

#define DBBUCKETS   67

#if 0
#define ATRACE(m)       isc_log_write(dns_lctx, \
                      DNS_LOGCATEGORY_DATABASE, \
                      DNS_LOGMODULE_ACACHE, \
                      ISC_LOG_DEBUG(3), \
                      "acache %p: %s", acache, (m))
#define AATRACE(a,m)    isc_log_write(dns_lctx, \
                      DNS_LOGCATEGORY_DATABASE, \
                      DNS_LOGMODULE_ACACHE, \
                      ISC_LOG_DEBUG(3), \
                      "acache %p: %s", (a), (m))
#else
#define ATRACE(m)
#define AATRACE(a, m)
#endif

/*
 * The following variables control incremental cleaning.
 * MINSIZE is how many bytes is the floor for dns_acache_setcachesize().
 * CLEANERINCREMENT is how many entries are examined in one pass.
 * (XXX simply derived from definitions in cache.c  There may be better
 *  constants here.)
 */
#define DNS_ACACHE_MINSIZE      2097152 /* Bytes.  2097152 = 2 MB */
#define DNS_ACACHE_CLEANERINCREMENT 1000    /* Number of entries. */

#define DEFAULT_ACACHE_ENTRY_LOCK_COUNT 1009    /*%< Should be prime. */

#if defined(ISC_RWLOCK_USEATOMIC) && defined(ISC_PLATFORM_HAVEATOMICSTORE)
#define ACACHE_USE_RWLOCK 1
#endif

#ifdef ACACHE_USE_RWLOCK
#define ACACHE_INITLOCK(l)  isc_rwlock_init((l), 0, 0)
#define ACACHE_DESTROYLOCK(l)   isc_rwlock_destroy(l)
#define ACACHE_LOCK(l, t)   RWLOCK((l), (t))
#define ACACHE_UNLOCK(l, t) RWUNLOCK((l), (t))

#define acache_storetime(entry, t) \
    (isc_atomic_store((isc_int32_t *)&(entry)->lastused, (t)))
#else
#define ACACHE_INITLOCK(l)  isc_mutex_init(l)
#define ACACHE_DESTROYLOCK(l)   DESTROYLOCK(l)
#define ACACHE_LOCK(l, t)   LOCK(l)
#define ACACHE_UNLOCK(l, t) UNLOCK(l)

#define acache_storetime(entry, t) ((entry)->lastused = (t))
#endif

/* Locked by acache lock */
typedef struct dbentry {
    ISC_LINK(struct dbentry)    link;

    dns_db_t            *db;
    ISC_LIST(dns_acacheentry_t) originlist;
    ISC_LIST(dns_acacheentry_t) referlist;
} dbentry_t;

typedef ISC_LIST(dbentry_t) dbentrylist_t;

typedef struct acache_cleaner acache_cleaner_t;

typedef enum {
    cleaner_s_idle, /* Waiting for cleaning-interval to expire. */
    cleaner_s_busy, /* Currently cleaning. */
    cleaner_s_done  /* Freed enough memory after being overmem. */
} cleaner_state_t;

/*
 * Convenience macros for comprehensive assertion checking.
 */
#define CLEANER_IDLE(c) ((c)->state == cleaner_s_idle && \
             (c)->resched_event != NULL)
#define CLEANER_BUSY(c) ((c)->state == cleaner_s_busy && \
             (c)->resched_event == NULL)

struct acache_cleaner {
    isc_mutex_t     lock;
    /*
     * Locks overmem_event, overmem.  (See cache.c)
     */

    dns_acache_t        *acache;
    unsigned int        cleaning_interval; /* The cleaning-interval
                              from named.conf,
                              in seconds. */

    isc_stdtime_t       last_cleanup_time; /* The time when the last
                                  cleanup task completed */

    isc_timer_t         *cleaning_timer;
    isc_event_t     *resched_event; /* Sent by cleaner task to
                           itself to reschedule */
    isc_event_t     *overmem_event;

    dns_acacheentry_t   *current_entry; /* The bookmark entry to
                           restart the cleaning.
                           Locked by acache lock. */
    int             increment;  /* Number of entries to
                           clean in one increment */

    unsigned long       ncleaned;   /* Number of entries cleaned
                           up (for logging purposes) */
    cleaner_state_t     state;      /* Idle/Busy/Done. */
    isc_boolean_t       overmem;    /* The acache is in an overmem
                           state. */
};

struct dns_acachestats {
    unsigned int            hits;
    unsigned int            queries;
    unsigned int            misses;
    unsigned int            adds;
    unsigned int            deleted;
    unsigned int            cleaned;
    unsigned int            cleaner_runs;
    unsigned int            overmem;
    unsigned int            overmem_nocreates;
    unsigned int            nomem;
};

/*
 * The actual acache object.
 */

struct dns_acache {
    unsigned int            magic;

    isc_mem_t           *mctx;
    isc_refcount_t          refs;

#ifdef ACACHE_USE_RWLOCK
    isc_rwlock_t            *entrylocks;
#else
    isc_mutex_t             *entrylocks;
#endif

    isc_mutex_t         lock;

    int             live_cleaners;
    acache_cleaner_t        cleaner;
    ISC_LIST(dns_acacheentry_t) entries;
    unsigned int            dbentries;
    dbentrylist_t           dbbucket[DBBUCKETS];

    isc_boolean_t           shutting_down;

    isc_task_t          *task;
    isc_event_t         cevent;
    isc_boolean_t           cevent_sent;

    dns_acachestats_t       stats;
};

struct dns_acacheentry {
    unsigned int        magic;

    unsigned int        locknum;
    isc_refcount_t      references;

    dns_acache_t        *acache;

    /* Data for Management of cache entries */
    ISC_LINK(dns_acacheentry_t) link;
    ISC_LINK(dns_acacheentry_t) olink;
    ISC_LINK(dns_acacheentry_t) rlink;

    dns_db_t        *origdb; /* reference to the DB
                        holding this entry */

    /* Cache data */
    dns_zone_t      *zone;      /* zone this entry
                           belongs to */
    dns_db_t        *db;        /* DB this entry belongs to */
    dns_dbversion_t     *version;   /* the version of the DB */
    dns_dbnode_t        *node;      /* node this entry
                           belongs to */
    dns_name_t      *foundname; /* corresponding DNS name
                           and rdataset */

    /* Callback function and its argument */
    void            (*callback)(dns_acacheentry_t *, void **);
    void            *cbarg;

    /* Timestamp of the last time this entry is referred to */
    isc_stdtime32_t     lastused;
};

/*
 *  Internal functions (and prototypes).
 */
static inline isc_boolean_t check_noentry(dns_acache_t *acache);
static void destroy(dns_acache_t *acache);
static void shutdown_entries(dns_acache_t *acache);
static void shutdown_buckets(dns_acache_t *acache);
static void destroy_entry(dns_acacheentry_t *ent);
static inline void unlink_dbentries(dns_acache_t *acache,
                    dns_acacheentry_t *ent);
static inline isc_result_t finddbent(dns_acache_t *acache,
                     dns_db_t *db, dbentry_t **dbentryp);
static inline void clear_entry(dns_acache_t *acache, dns_acacheentry_t *entry);
static isc_result_t acache_cleaner_init(dns_acache_t *acache,
                    isc_timermgr_t *timermgr,
                    acache_cleaner_t *cleaner);
static void acache_cleaning_timer_action(isc_task_t *task, isc_event_t *event);
static void acache_incremental_cleaning_action(isc_task_t *task,
                           isc_event_t *event);
static void acache_overmem_cleaning_action(isc_task_t *task,
                       isc_event_t *event);
static void acache_cleaner_shutdown_action(isc_task_t *task,
                       isc_event_t *event);

/*
 * acache should be locked.  If it is not, the stats can get out of whack,
 * which is not a big deal for us since this is for debugging / stats
 */
static void
reset_stats(dns_acache_t *acache) {
    acache->stats.hits = 0;
    acache->stats.queries = 0;
    acache->stats.misses = 0;
    acache->stats.adds = 0;
    acache->stats.deleted = 0;
    acache->stats.cleaned = 0;
    acache->stats.overmem = 0;
    acache->stats.overmem_nocreates = 0;
    acache->stats.nomem = 0;
}

/*
 * The acache must be locked before calling.
 */
static inline isc_boolean_t
check_noentry(dns_acache_t *acache) {
    if (ISC_LIST_EMPTY(acache->entries) && acache->dbentries == 0) {
        return (ISC_TRUE);
    }

    return (ISC_FALSE);
}

/*
 * The acache must be locked before calling.
 */
static void
shutdown_entries(dns_acache_t *acache) {
    dns_acacheentry_t *entry, *entry_next;

    REQUIRE(DNS_ACACHE_VALID(acache));
    INSIST(acache->shutting_down);

    /*
     * Release the dependency of all entries, and detach them.
     */
    for (entry = ISC_LIST_HEAD(acache->entries);
         entry != NULL;
         entry = entry_next) {
        entry_next = ISC_LIST_NEXT(entry, link);

        ACACHE_LOCK(&acache->entrylocks[entry->locknum],
                isc_rwlocktype_write);

        /*
         * If the cleaner holds this entry, it will be unlinked and
         * freed in the cleaner later.
         */
        if (acache->cleaner.current_entry != entry)
            ISC_LIST_UNLINK(acache->entries, entry, link);
        unlink_dbentries(acache, entry);
        if (entry->callback != NULL) {
            (entry->callback)(entry, &entry->cbarg);
            entry->callback = NULL;
        }

        ACACHE_UNLOCK(&acache->entrylocks[entry->locknum],
                  isc_rwlocktype_write);

        if (acache->cleaner.current_entry != entry)
            dns_acache_detachentry(&entry);
    }
}

/*
 * The acache must be locked before calling.
 */
static void
shutdown_buckets(dns_acache_t *acache) {
    int i;
    dbentry_t *dbent;

    REQUIRE(DNS_ACACHE_VALID(acache));
    INSIST(acache->shutting_down);

    for (i = 0; i < DBBUCKETS; i++) {
        while ((dbent = ISC_LIST_HEAD(acache->dbbucket[i])) != NULL) {
            INSIST(ISC_LIST_EMPTY(dbent->originlist) &&
                   ISC_LIST_EMPTY(dbent->referlist));
            ISC_LIST_UNLINK(acache->dbbucket[i], dbent, link);

            dns_db_detach(&dbent->db);

            isc_mem_put(acache->mctx, dbent, sizeof(*dbent));

            acache->dbentries--;
        }
    }

    INSIST(acache->dbentries == 0);
}

static void
shutdown_task(isc_task_t *task, isc_event_t *ev) {
    dns_acache_t *acache;

    UNUSED(task);

    acache = ev->ev_arg;
    INSIST(DNS_ACACHE_VALID(acache));

    isc_event_free(&ev);

    LOCK(&acache->lock);

    shutdown_entries(acache);
    shutdown_buckets(acache);

    UNLOCK(&acache->lock);

    dns_acache_detach(&acache);
}

/* The acache and the entry must be locked before calling. */
static inline void
unlink_dbentries(dns_acache_t *acache, dns_acacheentry_t *ent) {
    isc_result_t result;
    dbentry_t *dbent;

    if (ISC_LINK_LINKED(ent, olink)) {
        INSIST(ent->origdb != NULL);
        dbent = NULL;
        result = finddbent(acache, ent->origdb, &dbent);
        INSIST(result == ISC_R_SUCCESS);

        ISC_LIST_UNLINK(dbent->originlist, ent, olink);
    }
    if (ISC_LINK_LINKED(ent, rlink)) {
        INSIST(ent->db != NULL);
        dbent = NULL;
        result = finddbent(acache, ent->db, &dbent);
        INSIST(result == ISC_R_SUCCESS);

        ISC_LIST_UNLINK(dbent->referlist, ent, rlink);
    }
}

/* There must not be a reference to this entry. */
static void
destroy_entry(dns_acacheentry_t *entry) {
    dns_acache_t *acache;

    REQUIRE(DNS_ACACHEENTRY_VALID(entry));

    acache = entry->acache;
    REQUIRE(DNS_ACACHE_VALID(acache));

    /*
     * Since there is no reference to this entry, it is safe to call
     * clear_entry() here.
     */
    clear_entry(acache, entry);

    isc_mem_put(acache->mctx, entry, sizeof(*entry));

    dns_acache_detach(&acache);
}

static void
destroy(dns_acache_t *acache) {
    int i;

    REQUIRE(DNS_ACACHE_VALID(acache));

    ATRACE("destroy");

    isc_mem_setwater(acache->mctx, NULL, NULL, 0, 0);

    if (acache->cleaner.overmem_event != NULL)
        isc_event_free(&acache->cleaner.overmem_event);

    if (acache->cleaner.resched_event != NULL)
        isc_event_free(&acache->cleaner.resched_event);

    if (acache->task != NULL)
        isc_task_detach(&acache->task);

    for (i = 0; i < DEFAULT_ACACHE_ENTRY_LOCK_COUNT; i++)
        ACACHE_DESTROYLOCK(&acache->entrylocks[i]);
    isc_mem_put(acache->mctx, acache->entrylocks,
            sizeof(*acache->entrylocks) *
            DEFAULT_ACACHE_ENTRY_LOCK_COUNT);

    DESTROYLOCK(&acache->cleaner.lock);

    DESTROYLOCK(&acache->lock);
    acache->magic = 0;

    isc_mem_putanddetach(&acache->mctx, acache, sizeof(*acache));
}

static inline isc_result_t
finddbent(dns_acache_t *acache, dns_db_t *db, dbentry_t **dbentryp) {
    int bucket;
    dbentry_t *dbentry;

    REQUIRE(DNS_ACACHE_VALID(acache));
    REQUIRE(db != NULL);
    REQUIRE(dbentryp != NULL && *dbentryp == NULL);

    /*
     * The caller must be holding the acache lock.
     */

    bucket = isc_hash_calc((const unsigned char *)&db,
                   sizeof(db), ISC_TRUE) % DBBUCKETS;

    for (dbentry = ISC_LIST_HEAD(acache->dbbucket[bucket]);
         dbentry != NULL;
         dbentry = ISC_LIST_NEXT(dbentry, link)) {
        if (dbentry->db == db)
            break;
    }

    *dbentryp = dbentry;

    if (dbentry == NULL)
        return (ISC_R_NOTFOUND);
    else
        return (ISC_R_SUCCESS);
}

static inline void
clear_entry(dns_acache_t *acache, dns_acacheentry_t *entry) {
    REQUIRE(DNS_ACACHE_VALID(acache));
    REQUIRE(DNS_ACACHEENTRY_VALID(entry));

    /*
     * The caller must be holing the entry lock.
     */

    if (entry->foundname) {
        dns_rdataset_t *rdataset, *rdataset_next;

        for (rdataset = ISC_LIST_HEAD(entry->foundname->list);
             rdataset != NULL;
             rdataset = rdataset_next) {
            rdataset_next = ISC_LIST_NEXT(rdataset, link);
            ISC_LIST_UNLINK(entry->foundname->list,
                    rdataset, link);
            dns_rdataset_disassociate(rdataset);
            isc_mem_put(acache->mctx, rdataset, sizeof(*rdataset));
        }
        if (dns_name_dynamic(entry->foundname))
            dns_name_free(entry->foundname, acache->mctx);
        isc_mem_put(acache->mctx, entry->foundname,
                sizeof(*entry->foundname));
        entry->foundname = NULL;
    }

    if (entry->node != NULL) {
        INSIST(entry->db != NULL);
        dns_db_detachnode(entry->db, &entry->node);
    }
    if (entry->version != NULL) {
        INSIST(entry->db != NULL);
        dns_db_closeversion(entry->db, &entry->version, ISC_FALSE);
    }
    if (entry->db != NULL)
        dns_db_detach(&entry->db);
    if (entry->zone != NULL)
        dns_zone_detach(&entry->zone);

    if (entry->origdb != NULL)
        dns_db_detach(&entry->origdb);
}

static isc_result_t
acache_cleaner_init(dns_acache_t *acache, isc_timermgr_t *timermgr,
            acache_cleaner_t *cleaner)
{
    int result;

    ATRACE("acache cleaner init");

    result = isc_mutex_init(&cleaner->lock);
    if (result != ISC_R_SUCCESS)
        goto fail;

    cleaner->increment = DNS_ACACHE_CLEANERINCREMENT;
    cleaner->state = cleaner_s_idle;
    cleaner->acache = acache;
    cleaner->overmem = ISC_FALSE;

    cleaner->cleaning_timer = NULL;
    cleaner->resched_event = NULL;
    cleaner->overmem_event = NULL;
    cleaner->current_entry = NULL;

    if (timermgr != NULL) {
        cleaner->acache->live_cleaners++;

        result = isc_task_onshutdown(acache->task,
                         acache_cleaner_shutdown_action,
                         acache);
        if (result != ISC_R_SUCCESS) {
            UNEXPECTED_ERROR(__FILE__, __LINE__,
                     "acache cleaner: "
                     "isc_task_onshutdown() failed: %s",
                     dns_result_totext(result));
            goto cleanup;
        }

        cleaner->cleaning_interval = 0; /* Initially turned off. */
        isc_stdtime_get(&cleaner->last_cleanup_time);
        result = isc_timer_create(timermgr, isc_timertype_inactive,
                      NULL, NULL,
                      acache->task,
                      acache_cleaning_timer_action,
                      cleaner, &cleaner->cleaning_timer);
        if (result != ISC_R_SUCCESS) {
            UNEXPECTED_ERROR(__FILE__, __LINE__,
                     "isc_timer_create() failed: %s",
                     dns_result_totext(result));
            result = ISC_R_UNEXPECTED;
            goto cleanup;
        }

        cleaner->resched_event =
            isc_event_allocate(acache->mctx, cleaner,
                       DNS_EVENT_ACACHECLEAN,
                       acache_incremental_cleaning_action,
                       cleaner, sizeof(isc_event_t));
        if (cleaner->resched_event == NULL) {
            result = ISC_R_NOMEMORY;
            goto cleanup;
        }

        cleaner->overmem_event =
            isc_event_allocate(acache->mctx, cleaner,
                       DNS_EVENT_ACACHEOVERMEM,
                       acache_overmem_cleaning_action,
                       cleaner, sizeof(isc_event_t));
        if (cleaner->overmem_event == NULL) {
            result = ISC_R_NOMEMORY;
            goto cleanup;
        }
    }

    return (ISC_R_SUCCESS);

 cleanup:
    if (cleaner->overmem_event != NULL)
        isc_event_free(&cleaner->overmem_event);
    if (cleaner->resched_event != NULL)
        isc_event_free(&cleaner->resched_event);
    if (cleaner->cleaning_timer != NULL)
        isc_timer_detach(&cleaner->cleaning_timer);
    cleaner->acache->live_cleaners--;
    DESTROYLOCK(&cleaner->lock);
 fail:
    return (result);
}

static void
begin_cleaning(acache_cleaner_t *cleaner) {
    dns_acacheentry_t *head;
    dns_acache_t *acache = cleaner->acache;

    /*
     * This function does not have to lock the cleaner, since critical
     * parameters (except current_entry, which is locked by acache lock,)
     * are only used in a single task context.
     */

    REQUIRE(CLEANER_IDLE(cleaner));
    INSIST(DNS_ACACHE_VALID(acache));
    INSIST(cleaner->current_entry == NULL);

    isc_log_write(dns_lctx, DNS_LOGCATEGORY_DATABASE,
              DNS_LOGMODULE_ACACHE, ISC_LOG_DEBUG(1),
              "begin acache cleaning, mem inuse %lu",
              (unsigned long)isc_mem_inuse(cleaner->acache->mctx));

    LOCK(&acache->lock);

    head = ISC_LIST_HEAD(acache->entries);
    if (head != NULL)
        dns_acache_attachentry(head, &cleaner->current_entry);

    UNLOCK(&acache->lock);

    if (cleaner->current_entry != NULL) {
        cleaner->ncleaned = 0;
        cleaner->state = cleaner_s_busy;
        isc_task_send(acache->task, &cleaner->resched_event);
    }

    return;
}

static void
end_cleaning(acache_cleaner_t *cleaner, isc_event_t *event) {
    dns_acache_t *acache = cleaner->acache;

    REQUIRE(CLEANER_BUSY(cleaner));
    REQUIRE(event != NULL);
    REQUIRE(DNS_ACACHEENTRY_VALID(cleaner->current_entry));

    /* No need to lock the cleaner (see begin_cleaning()). */

    LOCK(&acache->lock);

    /*
     * Even if the cleaner has the last reference to the entry, which means
     * the entry has been unused, it may still be linked if unlinking the
     * entry has been delayed due to the reference.
     */
    if (isc_refcount_current(&cleaner->current_entry->references) == 1) {
        INSIST(cleaner->current_entry->callback == NULL);

        if (ISC_LINK_LINKED(cleaner->current_entry, link)) {
            ISC_LIST_UNLINK(acache->entries,
                    cleaner->current_entry, link);
        }
    }
    dns_acache_detachentry(&cleaner->current_entry);

    if (cleaner->overmem)
        acache->stats.overmem++;
    acache->stats.cleaned += cleaner->ncleaned;
    acache->stats.cleaner_runs++;

    isc_log_write(dns_lctx, DNS_LOGCATEGORY_DATABASE, DNS_LOGMODULE_ACACHE,
              ISC_LOG_NOTICE,
              "acache %p stats: hits=%d misses=%d queries=%d "
              "adds=%d deleted=%d "
              "cleaned=%d cleaner_runs=%d overmem=%d "
              "overmem_nocreates=%d nomem=%d",
              acache,
              acache->stats.hits, acache->stats.misses,
              acache->stats.queries,
              acache->stats.adds, acache->stats.deleted,
              acache->stats.cleaned, acache->stats.cleaner_runs,
              acache->stats.overmem, acache->stats.overmem_nocreates,
              acache->stats.nomem);
    reset_stats(acache);

    isc_stdtime_get(&cleaner->last_cleanup_time);

    UNLOCK(&acache->lock);

    dns_acache_setcleaninginterval(cleaner->acache,
                       cleaner->cleaning_interval);

    isc_log_write(dns_lctx, DNS_LOGCATEGORY_DATABASE, DNS_LOGMODULE_ACACHE,
              ISC_LOG_DEBUG(1), "end acache cleaning, "
              "%lu entries cleaned, mem inuse %lu",
              cleaner->ncleaned,
              (unsigned long)isc_mem_inuse(cleaner->acache->mctx));

    if (cleaner->overmem) {
        isc_log_write(dns_lctx, DNS_LOGCATEGORY_DATABASE,
                  DNS_LOGMODULE_ACACHE, ISC_LOG_NOTICE,
                  "acache is still in overmem state "
                  "after cleaning");
    }

    cleaner->ncleaned = 0;
    cleaner->state = cleaner_s_idle;
    cleaner->resched_event = event;
}

/*
 * This is run once for every acache-cleaning-interval as defined
 * in named.conf.
 */
static void
acache_cleaning_timer_action(isc_task_t *task, isc_event_t *event) {
    acache_cleaner_t *cleaner = event->ev_arg;

    UNUSED(task);

    INSIST(event->ev_type == ISC_TIMEREVENT_TICK);

    isc_log_write(dns_lctx, DNS_LOGCATEGORY_DATABASE, DNS_LOGMODULE_ACACHE,
              ISC_LOG_DEBUG(1), "acache cleaning timer fired, "
              "cleaner state = %d", cleaner->state);

    if (cleaner->state == cleaner_s_idle)
        begin_cleaning(cleaner);

    isc_event_free(&event);
}

/* The caller must hold entry lock. */
static inline isc_boolean_t
entry_stale(acache_cleaner_t *cleaner, dns_acacheentry_t *entry,
        isc_stdtime32_t now32, unsigned int interval)
{
    /*
     * If the callback has been canceled, we definitely do not need the
     * entry.
     */
    if (entry->callback == NULL)
        return (ISC_TRUE);

    if (interval > cleaner->cleaning_interval)
        interval = cleaner->cleaning_interval;

    if (entry->lastused + interval < now32)
        return (ISC_TRUE);

    /*
     * If the acache is in the overmem state, probabilistically decide if
     * the entry should be purged, based on the time passed from its last
     * use and the cleaning interval.
     */
    if (cleaner->overmem) {
        unsigned int passed =
            now32 - entry->lastused; /* <= interval */
        isc_uint32_t val;

        if (passed > interval / 2)
            return (ISC_TRUE);
        isc_random_get(&val);
        if (passed > interval / 4)
            return (ISC_TF(val % 4 == 0));
        return (ISC_TF(val % 8 == 0));
    }

    return (ISC_FALSE);
}

/*
 * Do incremental cleaning.
 */
static void
acache_incremental_cleaning_action(isc_task_t *task, isc_event_t *event) {
    acache_cleaner_t *cleaner = event->ev_arg;
    dns_acache_t *acache = cleaner->acache;
    dns_acacheentry_t *entry, *next = NULL;
    int n_entries;
    isc_stdtime32_t now32, last32;
    isc_stdtime_t now;
    unsigned int interval;

    INSIST(DNS_ACACHE_VALID(acache));
    INSIST(task == acache->task);
    INSIST(event->ev_type == DNS_EVENT_ACACHECLEAN);

    if (cleaner->state == cleaner_s_done) {
        cleaner->state = cleaner_s_busy;
        end_cleaning(cleaner, event);
        return;
    }

    INSIST(CLEANER_BUSY(cleaner));

    n_entries = cleaner->increment;

    isc_stdtime_get(&now);
    isc_stdtime_convert32(now, &now32);

    LOCK(&acache->lock);

    entry = cleaner->current_entry;
    isc_stdtime_convert32(cleaner->last_cleanup_time, &last32);
    INSIST(now32 > last32);
    interval = now32 - last32;

    while (n_entries-- > 0) {
        isc_boolean_t is_stale = ISC_FALSE;

        INSIST(entry != NULL);

        next = ISC_LIST_NEXT(entry, link);

        ACACHE_LOCK(&acache->entrylocks[entry->locknum],
                isc_rwlocktype_write);

        is_stale = entry_stale(cleaner, entry, now32, interval);
        if (is_stale) {
            ISC_LIST_UNLINK(acache->entries, entry, link);
            unlink_dbentries(acache, entry);
            if (entry->callback != NULL)
                (entry->callback)(entry, &entry->cbarg);
            entry->callback = NULL;

            cleaner->ncleaned++;
        }

        ACACHE_UNLOCK(&acache->entrylocks[entry->locknum],
                  isc_rwlocktype_write);

        if (is_stale)
            dns_acache_detachentry(&entry);

        if (next == NULL) {
            if (cleaner->overmem) {
                entry = ISC_LIST_HEAD(acache->entries);
                if (entry != NULL) {
                    /*
                     * If we are still in the overmem
                     * state, keep cleaning.
                     */
                    isc_log_write(dns_lctx,
                              DNS_LOGCATEGORY_DATABASE,
                              DNS_LOGMODULE_ACACHE,
                              ISC_LOG_DEBUG(1),
                              "acache cleaner: "
                              "still overmem, "
                              "reset and try again");
                    continue;
                }
            }

            UNLOCK(&acache->lock);
            end_cleaning(cleaner, event);
            return;
        }

        entry = next;
    }

    /*
     * We have successfully performed a cleaning increment but have
     * not gone through the entire cache.  Remember the entry that will
     * be the starting point in the next clean-up, and reschedule another
     * batch.  If it fails, just try to continue anyway.
     */
    INSIST(next != NULL && next != cleaner->current_entry);
    dns_acache_detachentry(&cleaner->current_entry);
    dns_acache_attachentry(next, &cleaner->current_entry);

    UNLOCK(&acache->lock);

    isc_log_write(dns_lctx, DNS_LOGCATEGORY_DATABASE, DNS_LOGMODULE_ACACHE,
              ISC_LOG_DEBUG(1), "acache cleaner: checked %d entries, "
              "mem inuse %lu, sleeping", cleaner->increment,
              (unsigned long)isc_mem_inuse(cleaner->acache->mctx));

    isc_task_send(task, &event);
    INSIST(CLEANER_BUSY(cleaner));

    return;
}

/*
 * This is called when the acache either surpasses its upper limit
 * or shrinks beyond its lower limit.
 */
static void
acache_overmem_cleaning_action(isc_task_t *task, isc_event_t *event) {
    acache_cleaner_t *cleaner = event->ev_arg;
    isc_boolean_t want_cleaning = ISC_FALSE;

    UNUSED(task);

    INSIST(event->ev_type == DNS_EVENT_ACACHEOVERMEM);
    INSIST(cleaner->overmem_event == NULL);

    isc_log_write(dns_lctx, DNS_LOGCATEGORY_DATABASE, DNS_LOGMODULE_ACACHE,
              ISC_LOG_DEBUG(1), "overmem_cleaning_action called, "
              "overmem = %d, state = %d", cleaner->overmem,
              cleaner->state);

    LOCK(&cleaner->lock);

    if (cleaner->overmem) {
        if (cleaner->state == cleaner_s_idle)
            want_cleaning = ISC_TRUE;
    } else {
        if (cleaner->state == cleaner_s_busy)
            /*
             * end_cleaning() can't be called here because
             * then both cleaner->overmem_event and
             * cleaner->resched_event will point to this
             * event.  Set the state to done, and then
             * when the acache_incremental_cleaning_action() event
             * is posted, it will handle the end_cleaning.
             */
            cleaner->state = cleaner_s_done;
    }

    cleaner->overmem_event = event;

    UNLOCK(&cleaner->lock);

    if (want_cleaning)
        begin_cleaning(cleaner);
}

static void
water(void *arg, int mark) {
    dns_acache_t *acache = arg;
    isc_boolean_t overmem = ISC_TF(mark == ISC_MEM_HIWATER);

    REQUIRE(DNS_ACACHE_VALID(acache));

    isc_log_write(dns_lctx, DNS_LOGCATEGORY_DATABASE,
              DNS_LOGMODULE_ACACHE, ISC_LOG_DEBUG(1),
              "acache memory reaches %s watermark, mem inuse %lu",
              overmem ? "high" : "low",
              (unsigned long)isc_mem_inuse(acache->mctx));

    LOCK(&acache->cleaner.lock);

    acache->cleaner.overmem = overmem;

    if (acache->cleaner.overmem_event != NULL)
        isc_task_send(acache->task, &acache->cleaner.overmem_event);

    UNLOCK(&acache->cleaner.lock);
}

/*
 * The cleaner task is shutting down; do the necessary cleanup.
 */
static void
acache_cleaner_shutdown_action(isc_task_t *task, isc_event_t *event) {
    dns_acache_t *acache = event->ev_arg;
    isc_boolean_t should_free = ISC_FALSE;

    INSIST(task == acache->task);
    INSIST(event->ev_type == ISC_TASKEVENT_SHUTDOWN);
    INSIST(DNS_ACACHE_VALID(acache));

    ATRACE("acache cleaner shutdown");

    if (CLEANER_BUSY(&acache->cleaner))
        end_cleaning(&acache->cleaner, event);
    else
        isc_event_free(&event);

    LOCK(&acache->lock);

    acache->live_cleaners--;
    INSIST(acache->live_cleaners == 0);

    if (isc_refcount_current(&acache->refs) == 0) {
        INSIST(check_noentry(acache) == ISC_TRUE);
        should_free = ISC_TRUE;
    }

    /*
     * By detaching the timer in the context of its task,
     * we are guaranteed that there will be no further timer
     * events.
     */
    if (acache->cleaner.cleaning_timer != NULL)
        isc_timer_detach(&acache->cleaner.cleaning_timer);

    /* Make sure we don't reschedule anymore. */
    (void)isc_task_purge(task, NULL, DNS_EVENT_ACACHECLEAN, NULL);

    UNLOCK(&acache->lock);

    if (should_free)
        destroy(acache);
}

/*
 *  Public functions.
 */

isc_result_t
dns_acache_create(dns_acache_t **acachep, isc_mem_t *mctx,
          isc_taskmgr_t *taskmgr, isc_timermgr_t *timermgr)
{
    int i;
    isc_result_t result;
    dns_acache_t *acache;

    REQUIRE(acachep != NULL && *acachep == NULL);
    REQUIRE(mctx != NULL);
    REQUIRE(taskmgr != NULL);

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

    ATRACE("create");

    result = isc_refcount_init(&acache->refs, 1);
    if (result != ISC_R_SUCCESS) {
        isc_mem_put(mctx, acache, sizeof(*acache));
        return (result);
    }

    result = isc_mutex_init(&acache->lock);
    if (result != ISC_R_SUCCESS) {
        isc_refcount_decrement(&acache->refs, NULL);
        isc_refcount_destroy(&acache->refs);
        isc_mem_put(mctx, acache, sizeof(*acache));
        return (result);
    }

    acache->mctx = NULL;
    isc_mem_attach(mctx, &acache->mctx);
    ISC_LIST_INIT(acache->entries);

    acache->shutting_down = ISC_FALSE;

    acache->task = NULL;
    acache->entrylocks = NULL;

    result = isc_task_create(taskmgr, 1, &acache->task);
    if (result != ISC_R_SUCCESS) {
        UNEXPECTED_ERROR(__FILE__, __LINE__,
                 "isc_task_create() failed(): %s",
                 dns_result_totext(result));
        result = ISC_R_UNEXPECTED;
        goto cleanup;
    }
    isc_task_setname(acache->task, "acachetask", acache);
    ISC_EVENT_INIT(&acache->cevent, sizeof(acache->cevent), 0, NULL,
               DNS_EVENT_ACACHECONTROL, shutdown_task, NULL,
               NULL, NULL, NULL);
    acache->cevent_sent = ISC_FALSE;

    acache->dbentries = 0;
    for (i = 0; i < DBBUCKETS; i++)
        ISC_LIST_INIT(acache->dbbucket[i]);

    acache->entrylocks = isc_mem_get(mctx, sizeof(*acache->entrylocks) *
                     DEFAULT_ACACHE_ENTRY_LOCK_COUNT);
    if (acache->entrylocks == NULL) {
        result = ISC_R_NOMEMORY;
        goto cleanup;
    }
    for (i = 0; i < DEFAULT_ACACHE_ENTRY_LOCK_COUNT; i++) {
        result = ACACHE_INITLOCK(&acache->entrylocks[i]);
        if (result != ISC_R_SUCCESS) {
            while (i-- > 0)
                ACACHE_DESTROYLOCK(&acache->entrylocks[i]);
            isc_mem_put(mctx, acache->entrylocks,
                    sizeof(*acache->entrylocks) *
                    DEFAULT_ACACHE_ENTRY_LOCK_COUNT);
            acache->entrylocks = NULL;
            goto cleanup;
        }
    }

    acache->live_cleaners = 0;
    result = acache_cleaner_init(acache, timermgr, &acache->cleaner);
    if (result != ISC_R_SUCCESS)
        goto cleanup;

    acache->stats.cleaner_runs = 0;
    reset_stats(acache);

    acache->magic = ACACHE_MAGIC;

    *acachep = acache;
    return (ISC_R_SUCCESS);

 cleanup:
    if (acache->task != NULL)
        isc_task_detach(&acache->task);
    DESTROYLOCK(&acache->lock);
    isc_refcount_decrement(&acache->refs, NULL);
    isc_refcount_destroy(&acache->refs);
    if (acache->entrylocks != NULL) {
        for (i = 0; i < DEFAULT_ACACHE_ENTRY_LOCK_COUNT; i++)
            ACACHE_DESTROYLOCK(&acache->entrylocks[i]);
        isc_mem_put(mctx, acache->entrylocks,
                sizeof(*acache->entrylocks) *
                DEFAULT_ACACHE_ENTRY_LOCK_COUNT);
    }
    isc_mem_put(mctx, acache, sizeof(*acache));
    isc_mem_detach(&mctx);

    return (result);
}

void
dns_acache_attach(dns_acache_t *source, dns_acache_t **targetp) {
    REQUIRE(DNS_ACACHE_VALID(source));
    REQUIRE(targetp != NULL && *targetp == NULL);

    AATRACE(source, "attach");

    isc_refcount_increment(&source->refs, NULL);

    *targetp = source;
}

void
dns_acache_countquerymiss(dns_acache_t *acache) {
    acache->stats.misses++;     /* XXXSK danger: unlocked! */
    acache->stats.queries++;    /* XXXSK danger: unlocked! */
}

void
dns_acache_detach(dns_acache_t **acachep) {
    dns_acache_t *acache;
    unsigned int refs;
    isc_boolean_t should_free = ISC_FALSE;

    REQUIRE(acachep != NULL && DNS_ACACHE_VALID(*acachep));
    acache = *acachep;

    ATRACE("detach");

    isc_refcount_decrement(&acache->refs, &refs);
    if (refs == 0) {
        INSIST(check_noentry(acache) == ISC_TRUE);
        should_free = ISC_TRUE;
    }

    *acachep = NULL;

    /*
     * If we're exiting and the cleaner task exists, let it free the cache.
     */
    if (should_free && acache->live_cleaners > 0) {
        isc_task_shutdown(acache->task);
        should_free = ISC_FALSE;
    }

    if (should_free)
        destroy(acache);
}

void
dns_acache_shutdown(dns_acache_t *acache) {
    REQUIRE(DNS_ACACHE_VALID(acache));

    LOCK(&acache->lock);

    ATRACE("shutdown");

    if (!acache->shutting_down) {
        isc_event_t *event;
        dns_acache_t *acache_evarg = NULL;

        INSIST(!acache->cevent_sent);

        acache->shutting_down = ISC_TRUE;

        isc_mem_setwater(acache->mctx, NULL, NULL, 0, 0);

        /*
         * Self attach the object in order to prevent it from being
         * destroyed while waiting for the event.
         */
        dns_acache_attach(acache, &acache_evarg);
        event = &acache->cevent;
        event->ev_arg = acache_evarg;
        isc_task_send(acache->task, &event);
        acache->cevent_sent = ISC_TRUE;
    }

    UNLOCK(&acache->lock);
}

isc_result_t
dns_acache_setdb(dns_acache_t *acache, dns_db_t *db) {
    int bucket;
    dbentry_t *dbentry;
    isc_result_t result = ISC_R_SUCCESS;

    REQUIRE(DNS_ACACHE_VALID(acache));
    REQUIRE(db != NULL);

    ATRACE("setdb");

    LOCK(&acache->lock);

    dbentry = NULL;
    result = finddbent(acache, db, &dbentry);
    if (result == ISC_R_SUCCESS) {
        result = ISC_R_EXISTS;
        goto end;
    }
    result = ISC_R_SUCCESS;

    dbentry = isc_mem_get(acache->mctx, sizeof(*dbentry));
    if (dbentry == NULL) {
        result = ISC_R_NOMEMORY;
        goto end;
    }

    ISC_LINK_INIT(dbentry, link);
    ISC_LIST_INIT(dbentry->originlist);
    ISC_LIST_INIT(dbentry->referlist);

    dbentry->db = NULL;
    dns_db_attach(db, &dbentry->db);

    bucket = isc_hash_calc((const unsigned char *)&db,
                   sizeof(db), ISC_TRUE) % DBBUCKETS;

    ISC_LIST_APPEND(acache->dbbucket[bucket], dbentry, link);

    acache->dbentries++;

 end:
    UNLOCK(&acache->lock);

    return (result);
}

isc_result_t
dns_acache_putdb(dns_acache_t *acache, dns_db_t *db) {
    int bucket;
    isc_result_t result;
    dbentry_t *dbentry;
    dns_acacheentry_t *entry;

    REQUIRE(DNS_ACACHE_VALID(acache));
    REQUIRE(db != NULL);

    ATRACE("putdb");

    LOCK(&acache->lock);

    dbentry = NULL;
    result = finddbent(acache, db, &dbentry);
    if (result != ISC_R_SUCCESS) {
        /*
         * The entry may have not been created due to memory shortage.
         */
        UNLOCK(&acache->lock);
        return (ISC_R_NOTFOUND);
    }

    /*
     * Release corresponding cache entries: for each entry, release all
     * links the entry has, and then callback to the entry holder (if any).
     * If no other external references exist (this can happen if the
     * original holder has canceled callback,) destroy it here.
     */
    while ((entry = ISC_LIST_HEAD(dbentry->originlist)) != NULL) {
        ACACHE_LOCK(&acache->entrylocks[entry->locknum],
                isc_rwlocktype_write);

        /*
         * Releasing olink first would avoid finddbent() in
         * unlink_dbentries().
         */
        ISC_LIST_UNLINK(dbentry->originlist, entry, olink);
        if (acache->cleaner.current_entry != entry)
            ISC_LIST_UNLINK(acache->entries, entry, link);
        unlink_dbentries(acache, entry);

        if (entry->callback != NULL)
            (entry->callback)(entry, &entry->cbarg);
        entry->callback = NULL;

        ACACHE_UNLOCK(&acache->entrylocks[entry->locknum],
                  isc_rwlocktype_write);

        if (acache->cleaner.current_entry != entry)
            dns_acache_detachentry(&entry);
    }
    while ((entry = ISC_LIST_HEAD(dbentry->referlist)) != NULL) {
        ACACHE_LOCK(&acache->entrylocks[entry->locknum],
                isc_rwlocktype_write);

        ISC_LIST_UNLINK(dbentry->referlist, entry, rlink);
        if (acache->cleaner.current_entry != entry)
            ISC_LIST_UNLINK(acache->entries, entry, link);
        unlink_dbentries(acache, entry);

        if (entry->callback != NULL)
            (entry->callback)(entry, &entry->cbarg);
        entry->callback = NULL;

        ACACHE_UNLOCK(&acache->entrylocks[entry->locknum],
                  isc_rwlocktype_write);

        if (acache->cleaner.current_entry != entry)
            dns_acache_detachentry(&entry);
    }

    INSIST(ISC_LIST_EMPTY(dbentry->originlist) &&
           ISC_LIST_EMPTY(dbentry->referlist));

    bucket = isc_hash_calc((const unsigned char *)&db,
                   sizeof(db), ISC_TRUE) % DBBUCKETS;
    ISC_LIST_UNLINK(acache->dbbucket[bucket], dbentry, link);
    dns_db_detach(&dbentry->db);

    isc_mem_put(acache->mctx, dbentry, sizeof(*dbentry));

    acache->dbentries--;

    acache->stats.deleted++;

    UNLOCK(&acache->lock);

    return (ISC_R_SUCCESS);
}

isc_result_t
dns_acache_createentry(dns_acache_t *acache, dns_db_t *origdb,
               void (*callback)(dns_acacheentry_t *, void **),
               void *cbarg, dns_acacheentry_t **entryp)
{
    dns_acacheentry_t *newentry;
    isc_result_t result;
    isc_uint32_t r;

    REQUIRE(DNS_ACACHE_VALID(acache));
    REQUIRE(entryp != NULL && *entryp == NULL);
    REQUIRE(origdb != NULL);

    /*
     * Should we exceed our memory limit for some reason (for
     * example, if the cleaner does not run aggressively enough),
     * then we will not create additional entries.
     *
     * XXXSK: It might be better to lock the acache->cleaner->lock,
     * but locking may be an expensive bottleneck. If we misread
     * the value, we will occasionally refuse to create a few
     * cache entries, or create a few that we should not. I do not
     * expect this to happen often, and it will not have very bad
     * effects when it does. So no lock for now.
     */
    if (acache->cleaner.overmem) {
        acache->stats.overmem_nocreates++; /* XXXSK danger: unlocked! */
        return (ISC_R_NORESOURCES);
    }

    newentry = isc_mem_get(acache->mctx, sizeof(*newentry));
    if (newentry == NULL) {
        acache->stats.nomem++;  /* XXXMLG danger: unlocked! */
        return (ISC_R_NOMEMORY);
    }

    isc_random_get(&r);
    newentry->locknum = r % DEFAULT_ACACHE_ENTRY_LOCK_COUNT;

    result = isc_refcount_init(&newentry->references, 1);
    if (result != ISC_R_SUCCESS) {
        isc_mem_put(acache->mctx, newentry, sizeof(*newentry));
        return (result);
    };

    ISC_LINK_INIT(newentry, link);
    ISC_LINK_INIT(newentry, olink);
    ISC_LINK_INIT(newentry, rlink);

    newentry->acache = NULL;
    dns_acache_attach(acache, &newentry->acache);

    newentry->zone = NULL;
    newentry->db = NULL;
    newentry->version = NULL;
    newentry->node = NULL;
    newentry->foundname = NULL;

    newentry->callback = callback;
    newentry->cbarg = cbarg;
    newentry->origdb = NULL;
    dns_db_attach(origdb, &newentry->origdb);

    isc_stdtime_get(&newentry->lastused);

    newentry->magic = ACACHEENTRY_MAGIC;

    *entryp = newentry;

    return (ISC_R_SUCCESS);
}

isc_result_t
dns_acache_getentry(dns_acacheentry_t *entry, dns_zone_t **zonep,
            dns_db_t **dbp, dns_dbversion_t **versionp,
            dns_dbnode_t **nodep, dns_name_t *fname,
            dns_message_t *msg, isc_stdtime_t now)
{
    isc_result_t result = ISC_R_SUCCESS;
    dns_rdataset_t *erdataset;
    isc_stdtime32_t now32;
    dns_acache_t *acache;
    int locknum;

    REQUIRE(DNS_ACACHEENTRY_VALID(entry));
    REQUIRE(zonep == NULL || *zonep == NULL);
    REQUIRE(dbp != NULL && *dbp == NULL);
    REQUIRE(versionp != NULL && *versionp == NULL);
    REQUIRE(nodep != NULL && *nodep == NULL);
    REQUIRE(fname != NULL);
    REQUIRE(msg != NULL);
    acache = entry->acache;
    REQUIRE(DNS_ACACHE_VALID(acache));

    locknum = entry->locknum;
    ACACHE_LOCK(&acache->entrylocks[locknum], isc_rwlocktype_read);

    isc_stdtime_convert32(now, &now32);
    acache_storetime(entry, now32);

    if (entry->zone != NULL && zonep != NULL)
        dns_zone_attach(entry->zone, zonep);

    if (entry->db == NULL) {
        *dbp = NULL;
        *versionp = NULL;
    } else {
        dns_db_attach(entry->db, dbp);
        dns_db_attachversion(entry->db, entry->version, versionp);
    }
    if (entry->node == NULL)
        *nodep = NULL;
    else {
        dns_db_attachnode(entry->db, entry->node, nodep);

        INSIST(entry->foundname != NULL);
        dns_name_copy(entry->foundname, fname, NULL);
        for (erdataset = ISC_LIST_HEAD(entry->foundname->list);
             erdataset != NULL;
             erdataset = ISC_LIST_NEXT(erdataset, link)) {
            dns_rdataset_t *ardataset;

            ardataset = NULL;
            result = dns_message_gettemprdataset(msg, &ardataset);
            if (result != ISC_R_SUCCESS) {
                ACACHE_UNLOCK(&acache->entrylocks[locknum],
                          isc_rwlocktype_read);
                goto fail;
            }

            /*
             * XXXJT: if we simply clone the rdataset, we'll get
             * lost wrt cyclic ordering.  We'll need an additional
             * trick to get the latest counter from the original
             * header.
             */
            dns_rdataset_init(ardataset);
            dns_rdataset_clone(erdataset, ardataset);
            ISC_LIST_APPEND(fname->list, ardataset, link);
        }
    }

    entry->acache->stats.hits++; /* XXXMLG danger: unlocked! */
    entry->acache->stats.queries++;

    ACACHE_UNLOCK(&acache->entrylocks[locknum], isc_rwlocktype_read);

    return (result);

  fail:
    while ((erdataset = ISC_LIST_HEAD(fname->list)) != NULL) {
        ISC_LIST_UNLINK(fname->list, erdataset, link);
        dns_rdataset_disassociate(erdataset);
        dns_message_puttemprdataset(msg, &erdataset);
    }
    if (*nodep != NULL)
        dns_db_detachnode(*dbp, nodep);
    if (*versionp != NULL)
        dns_db_closeversion(*dbp, versionp, ISC_FALSE);
    if (*dbp != NULL)
        dns_db_detach(dbp);
    if (zonep != NULL && *zonep != NULL)
        dns_zone_detach(zonep);

    return (result);
}

isc_result_t
dns_acache_setentry(dns_acache_t *acache, dns_acacheentry_t *entry,
            dns_zone_t *zone, dns_db_t *db, dns_dbversion_t *version,
            dns_dbnode_t *node, dns_name_t *fname)
{
    isc_result_t result;
    dbentry_t *odbent;
    dbentry_t *rdbent = NULL;
    isc_boolean_t close_version = ISC_FALSE;
    dns_acacheentry_t *dummy_entry = NULL;

    REQUIRE(DNS_ACACHE_VALID(acache));
    REQUIRE(DNS_ACACHEENTRY_VALID(entry));

    LOCK(&acache->lock);    /* XXX: need to lock it here for ordering */
    ACACHE_LOCK(&acache->entrylocks[entry->locknum], isc_rwlocktype_write);

    /* Set zone */
    if (zone != NULL)
        dns_zone_attach(zone, &entry->zone);
    /* Set DB */
    if (db != NULL)
        dns_db_attach(db, &entry->db);
    /*
     * Set DB version.  If the version is not given by the caller,
     * which is the case for glue or cache DBs, use the current version.
     */
    if (version == NULL) {
        if (db != NULL) {
            dns_db_currentversion(db, &version);
            close_version = ISC_TRUE;
        }
    }
    if (version != NULL) {
        INSIST(db != NULL);
        dns_db_attachversion(db, version, &entry->version);
    }
    if (close_version)
        dns_db_closeversion(db, &version, ISC_FALSE);
    /* Set DB node. */
    if (node != NULL) {
        INSIST(db != NULL);
        dns_db_attachnode(db, node, &entry->node);
    }

    /*
     * Set list of the corresponding rdatasets, if given.
     * To minimize the overhead and memory consumption, we'll do this for
     * positive cache only, in which case the DB node is non NULL.
     * We do not want to cache incomplete information, so give up the
     * entire entry when a memory shortage happen during the process.
     */
    if (node != NULL) {
        dns_rdataset_t *ardataset, *crdataset;

        entry->foundname = isc_mem_get(acache->mctx,
                           sizeof(*entry->foundname));

        if (entry->foundname == NULL) {
            result = ISC_R_NOMEMORY;
            goto fail;
        }
        dns_name_init(entry->foundname, NULL);
        result = dns_name_dup(fname, acache->mctx,
                      entry->foundname);
        if (result != ISC_R_SUCCESS)
            goto fail;

        for (ardataset = ISC_LIST_HEAD(fname->list);
             ardataset != NULL;
             ardataset = ISC_LIST_NEXT(ardataset, link)) {
            crdataset = isc_mem_get(acache->mctx,
                        sizeof(*crdataset));
            if (crdataset == NULL) {
                result = ISC_R_NOMEMORY;
                goto fail;
            }

            dns_rdataset_init(crdataset);
            dns_rdataset_clone(ardataset, crdataset);
            ISC_LIST_APPEND(entry->foundname->list, crdataset,
                    link);
        }
    }

    odbent = NULL;
    result = finddbent(acache, entry->origdb, &odbent);
    if (result != ISC_R_SUCCESS)
        goto fail;
    if (db != NULL) {
        rdbent = NULL;
        result = finddbent(acache, db, &rdbent);
        if (result != ISC_R_SUCCESS)
            goto fail;
    }

    ISC_LIST_APPEND(acache->entries, entry, link);
    ISC_LIST_APPEND(odbent->originlist, entry, olink);
    if (rdbent != NULL)
        ISC_LIST_APPEND(rdbent->referlist, entry, rlink);

    /*
     * The additional cache needs an implicit reference to entries in its
     * link.
     */
    dns_acache_attachentry(entry, &dummy_entry);

    ACACHE_UNLOCK(&acache->entrylocks[entry->locknum],
              isc_rwlocktype_write);

    acache->stats.adds++;
    UNLOCK(&acache->lock);

    return (ISC_R_SUCCESS);

 fail:
    clear_entry(acache, entry);

    ACACHE_UNLOCK(&acache->entrylocks[entry->locknum],
              isc_rwlocktype_write);
    UNLOCK(&acache->lock);

    return (result);
}

void
dns_acache_cancelentry(dns_acacheentry_t *entry) {
    dns_acache_t *acache = entry->acache;

    REQUIRE(DNS_ACACHEENTRY_VALID(entry));
    INSIST(DNS_ACACHE_VALID(acache));

    LOCK(&acache->lock);
    ACACHE_LOCK(&acache->entrylocks[entry->locknum], isc_rwlocktype_write);

    /*
     * Release dependencies stored in this entry as much as possible.
     * The main link cannot be released, since the acache object has
     * a reference to this entry; the empty entry will be released in
     * the next cleaning action.
     */
    unlink_dbentries(acache, entry);
    clear_entry(entry->acache, entry);

    entry->callback = NULL;
    entry->cbarg = NULL;

    ACACHE_UNLOCK(&acache->entrylocks[entry->locknum],
              isc_rwlocktype_write);
    UNLOCK(&acache->lock);
}

void
dns_acache_attachentry(dns_acacheentry_t *source,
               dns_acacheentry_t **targetp)
{
    REQUIRE(DNS_ACACHEENTRY_VALID(source));
    REQUIRE(targetp != NULL && *targetp == NULL);

    isc_refcount_increment(&source->references, NULL);

    *targetp = source;
}

void
dns_acache_detachentry(dns_acacheentry_t **entryp) {
    dns_acacheentry_t *entry;
    unsigned int refs;

    REQUIRE(entryp != NULL && DNS_ACACHEENTRY_VALID(*entryp));
    entry = *entryp;

    isc_refcount_decrement(&entry->references, &refs);

    /*
     * If there are no references to the entry, the entry must have been
     * unlinked and can be destroyed safely.
     */
    if (refs == 0) {
        INSIST(!ISC_LINK_LINKED(entry, link));
        (*entryp)->acache->stats.deleted++;
        destroy_entry(entry);
    }

    *entryp = NULL;
}

void
dns_acache_setcleaninginterval(dns_acache_t *acache, unsigned int t) {
    isc_interval_t interval;
    isc_result_t result;

    REQUIRE(DNS_ACACHE_VALID(acache));

    ATRACE("dns_acache_setcleaninginterval");

    LOCK(&acache->lock);

    /*
     * It may be the case that the acache has already shut down.
     * If so, it has no timer.  (Not sure if this can really happen.)
     */
    if (acache->cleaner.cleaning_timer == NULL)
        goto unlock;

    acache->cleaner.cleaning_interval = t;

    if (t == 0) {
        result = isc_timer_reset(acache->cleaner.cleaning_timer,
                     isc_timertype_inactive,
                     NULL, NULL, ISC_TRUE);
    } else {
        isc_interval_set(&interval, acache->cleaner.cleaning_interval,
                 0);
        result = isc_timer_reset(acache->cleaner.cleaning_timer,
                     isc_timertype_ticker,
                     NULL, &interval, ISC_FALSE);
    }
    if (result != ISC_R_SUCCESS)
        isc_log_write(dns_lctx, DNS_LOGCATEGORY_DATABASE,
                  DNS_LOGMODULE_ACACHE, ISC_LOG_WARNING,
                  "could not set acache cleaning interval: %s",
                  isc_result_totext(result));
    else
        isc_log_write(dns_lctx, DNS_LOGCATEGORY_DATABASE,
                  DNS_LOGMODULE_ACACHE, ISC_LOG_NOTICE,
                  "acache %p cleaning interval set to %d.",
                  acache, t);

 unlock:
    UNLOCK(&acache->lock);
}

/*
 * This function was derived from cache.c:dns_cache_setcachesize().  See the
 * function for more details about the logic.
 */
void
dns_acache_setcachesize(dns_acache_t *acache, isc_uint32_t size) {
    isc_uint32_t lowater;
    isc_uint32_t hiwater;

    REQUIRE(DNS_ACACHE_VALID(acache));

    if (size != 0 && size < DNS_ACACHE_MINSIZE)
        size = DNS_ACACHE_MINSIZE;

    hiwater = size - (size >> 3);
    lowater = size - (size >> 2);

    if (size == 0 || hiwater == 0 || lowater == 0)
        isc_mem_setwater(acache->mctx, water, acache, 0, 0);
    else
        isc_mem_setwater(acache->mctx, water, acache,
                 hiwater, lowater);
}