GCOV - Varnish Cache


/*-
 * Copyright (c) 2015 Varnish Software AS
 * All rights reserved.
 *
 * Author: Poul-Henning Kamp <phk@phk.freebsd.dk>
 *
 * SPDX-License-Identifier: BSD-2-Clause
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 * (TCP|UDS) connection pools.
 *
 */

#include "config.h"

#include <stdlib.h>

#include "cache_varnishd.h"

#include "vsa.h"
#include "vsha256.h"
#include "vtcp.h"
#include "vus.h"
#include "vtim.h"
#include "waiter/waiter.h"

#include "cache_conn_pool.h"
#include "cache_pool.h"

struct conn_pool;
static inline int vcp_cmp(const struct conn_pool *a, const struct conn_pool *b);

/*--------------------------------------------------------------------
 */

struct pfd {
        unsigned                magic;
#define PFD_MAGIC               0x0c5e6593
        int                     fd;
        VTAILQ_ENTRY(pfd)       list;
        VCL_IP                  addr;
        uint8_t                 state;
        struct waited           waited[1];
        struct conn_pool        *conn_pool;

        pthread_cond_t          *cond;
};

/*--------------------------------------------------------------------
 */

typedef int cp_open_f(const struct conn_pool *, vtim_dur tmo, VCL_IP *ap);
typedef void cp_close_f(struct pfd *);
typedef void cp_name_f(const struct pfd *, char *, unsigned, char *, unsigned);

struct cp_methods {
        cp_open_f                               *open;
        cp_close_f                              *close;
        cp_name_f                               *local_name;
        cp_name_f                               *remote_name;
};

struct conn_pool {
        unsigned                                magic;
#define CONN_POOL_MAGIC                         0x85099bc3

        const struct cp_methods                 *methods;

        struct vrt_endpoint                     *endpoint;
        char                                    ident[VSHA256_DIGEST_LENGTH];

        VRBT_ENTRY(conn_pool)                   entry;
        int                                     refcnt;
        struct lock                             mtx;

        VTAILQ_HEAD(, pfd)                      connlist;
        int                                     n_conn;

        int                                     n_kill;

        int                                     n_used;

        vtim_mono                               holddown;
        int                                     holddown_errno;
};

static struct lock conn_pools_mtx;
static struct lock dead_pools_mtx;

VRBT_HEAD(vrb, conn_pool);
VRBT_GENERATE_REMOVE_COLOR(vrb, conn_pool, entry, static)
VRBT_GENERATE_REMOVE(vrb, conn_pool, entry, static)
VRBT_GENERATE_FIND(vrb, conn_pool, entry, vcp_cmp, static)
VRBT_GENERATE_INSERT_COLOR(vrb, conn_pool, entry, static)
VRBT_GENERATE_INSERT_FINISH(vrb, conn_pool, entry, static)
VRBT_GENERATE_INSERT(vrb, conn_pool, entry, vcp_cmp, static)
VRBT_GENERATE_NEXT(vrb, conn_pool, entry, static)
VRBT_GENERATE_MINMAX(vrb, conn_pool, entry, static)

static struct vrb conn_pools = VRBT_INITIALIZER(&conn_pools);
static struct vrb dead_pools = VRBT_INITIALIZER(&dying_cps);

/*--------------------------------------------------------------------
 */

unsigned
PFD_State(const struct pfd *p)
{
        CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
        return (p->state);
}

int *
PFD_Fd(struct pfd *p)
{
        CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
        return (&(p->fd));
}

void
PFD_LocalName(const struct pfd *p, char *abuf, unsigned alen, char *pbuf,
              unsigned plen)
{
        CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
        CHECK_OBJ_NOTNULL(p->conn_pool, CONN_POOL_MAGIC);
        p->conn_pool->methods->local_name(p, abuf, alen, pbuf, plen);
}

void
PFD_RemoteName(const struct pfd *p, char *abuf, unsigned alen, char *pbuf,
               unsigned plen)
{
        CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
        CHECK_OBJ_NOTNULL(p->conn_pool, CONN_POOL_MAGIC);
        p->conn_pool->methods->remote_name(p, abuf, alen, pbuf, plen);
}

/*--------------------------------------------------------------------
 */

static inline int
vcp_cmp(const struct conn_pool *a, const struct conn_pool *b)
{
        return (memcmp(a->ident, b->ident, sizeof b->ident));
}

/*--------------------------------------------------------------------
 * Waiter-handler
 */

static void  v_matchproto_(waiter_handle_f)
vcp_handle(struct waited *w, enum wait_event ev, vtim_real now)
{
        struct pfd *pfd;
        struct conn_pool *cp;

        CHECK_OBJ_NOTNULL(w, WAITED_MAGIC);
        CAST_OBJ_NOTNULL(pfd, w->priv1, PFD_MAGIC);
        (void)ev;
        (void)now;
        CHECK_OBJ_NOTNULL(pfd->conn_pool, CONN_POOL_MAGIC);
        cp = pfd->conn_pool;

        Lck_Lock(&cp->mtx);

        switch (pfd->state) {
        case PFD_STATE_STOLEN:
                pfd->state = PFD_STATE_USED;
                VTAILQ_REMOVE(&cp->connlist, pfd, list);
                AN(pfd->cond);
                PTOK(pthread_cond_signal(pfd->cond));
                break;
        case PFD_STATE_AVAIL:
                cp->methods->close(pfd);
                VTAILQ_REMOVE(&cp->connlist, pfd, list);
                cp->n_conn--;
                FREE_OBJ(pfd);
                break;
        case PFD_STATE_CLEANUP:
                cp->methods->close(pfd);
                cp->n_kill--;
                memset(pfd, 0x11, sizeof *pfd);
                free(pfd);
                break;
        default:
                WRONG("Wrong pfd state");
        }
        Lck_Unlock(&cp->mtx);
}


/*--------------------------------------------------------------------
 */

void
VCP_AddRef(struct conn_pool *cp)
{
        CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);

        Lck_Lock(&conn_pools_mtx);
        assert(cp->refcnt > 0);
        cp->refcnt++;
        Lck_Unlock(&conn_pools_mtx);
}

/*--------------------------------------------------------------------
 */

static void
vcp_destroy(struct conn_pool **cpp)
{
        struct conn_pool *cp;

        TAKE_OBJ_NOTNULL(cp, cpp, CONN_POOL_MAGIC);
        AZ(cp->n_conn);
        AZ(cp->n_kill);
        Lck_Delete(&cp->mtx);
        free(cp->endpoint);
        FREE_OBJ(cp);
}

/*--------------------------------------------------------------------
 * Release Conn pool, destroy if last reference.
 */

void
VCP_Rel(struct conn_pool **cpp)
{
        struct conn_pool *cp;
        struct pfd *pfd, *pfd2;
        int n_kill;

        TAKE_OBJ_NOTNULL(cp, cpp, CONN_POOL_MAGIC);

        Lck_Lock(&conn_pools_mtx);
        assert(cp->refcnt > 0);
        if (--cp->refcnt > 0) {
                Lck_Unlock(&conn_pools_mtx);
                return;
        }
        AZ(cp->n_used);
        VRBT_REMOVE(vrb, &conn_pools, cp);
        Lck_Unlock(&conn_pools_mtx);

        Lck_Lock(&cp->mtx);
        VTAILQ_FOREACH_SAFE(pfd, &cp->connlist, list, pfd2) {
                VTAILQ_REMOVE(&cp->connlist, pfd, list);
                cp->n_conn--;
                assert(pfd->state == PFD_STATE_AVAIL);
                pfd->state = PFD_STATE_CLEANUP;
                (void)shutdown(pfd->fd, SHUT_RDWR);
                cp->n_kill++;
        }
        n_kill = cp->n_kill;
        Lck_Unlock(&cp->mtx);
        if (n_kill == 0) {
                vcp_destroy(&cp);
                return;
        }
        Lck_Lock(&dead_pools_mtx);
        /*
         * Here we reuse cp's entry but it will probably not be correctly
         * indexed because of the hack in VCP_RelPoll
         */
        VRBT_INSERT(vrb, &dead_pools, cp);
        Lck_Unlock(&dead_pools_mtx);
}

void
VCP_RelPoll(void)
{
        struct vrb dead;
        struct conn_pool *cp, *cp2;
        int n_kill;

        ASSERT_CLI();

        Lck_Lock(&dead_pools_mtx);
        if (VRBT_EMPTY(&dead_pools)) {
                Lck_Unlock(&dead_pools_mtx);
                return;
        }
        dead = dead_pools;
        VRBT_INIT(&dead_pools);
        Lck_Unlock(&dead_pools_mtx);

        VRBT_FOREACH_SAFE(cp, vrb, &dead, cp2) {
                CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
                Lck_Lock(&cp->mtx);
                n_kill = cp->n_kill;
                Lck_Unlock(&cp->mtx);
                if (n_kill > 0)
                        continue;
                VRBT_REMOVE(vrb, &dead, cp);
                vcp_destroy(&cp);
        }

        if (VRBT_EMPTY(&dead))
                return;

        Lck_Lock(&dead_pools_mtx);
        /*
         * The following insertion will most likely result in an
         * unordered tree, but in this case it does not matter
         * as we just want to iterate over all the elements
         * in the tree in order to delete them.
         */
        VRBT_INSERT(vrb, &dead_pools, dead.rbh_root);
        Lck_Unlock(&dead_pools_mtx);
}

/*--------------------------------------------------------------------
 * Recycle a connection.
 */

void
VCP_Recycle(const struct worker *wrk, struct pfd **pfdp)
{
        struct pfd *pfd;
        struct conn_pool *cp;
        int i = 0;

        CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
        TAKE_OBJ_NOTNULL(pfd, pfdp, PFD_MAGIC);
        cp = pfd->conn_pool;
        CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);

        assert(pfd->state == PFD_STATE_USED);
        assert(pfd->fd > 0);

        Lck_Lock(&cp->mtx);
        cp->n_used--;

        pfd->waited->priv1 = pfd;
        pfd->waited->fd = pfd->fd;
        pfd->waited->idle = VTIM_real();
        pfd->state = PFD_STATE_AVAIL;
        pfd->waited->func = vcp_handle;
        pfd->waited->tmo = cache_param->backend_idle_timeout;
        if (Wait_Enter(wrk->pool->waiter, pfd->waited)) {
                cp->methods->close(pfd);
                memset(pfd, 0x33, sizeof *pfd);
                free(pfd);
                // XXX: stats
                pfd = NULL;
        } else {
                VTAILQ_INSERT_HEAD(&cp->connlist, pfd, list);
                i++;
        }

        if (pfd != NULL)
                cp->n_conn++;
        Lck_Unlock(&cp->mtx);

        if (i && DO_DEBUG(DBG_VTC_MODE)) {
                /*
                 * In varnishtest we do not have the luxury of using
                 * multiple backend connections, so whenever we end up
                 * in the "pending" case, take a short nap to let the
                 * waiter catch up and put the pfd back into circulations.
                 *
                 * In particular ESI:include related tests suffer random
                 * failures without this.
                 *
                 * In normal operation, the only effect is that we will
                 * have N+1 backend connections rather than N, which is
                 * entirely harmless.
                 */
                VTIM_sleep(0.01);
        }
}

/*--------------------------------------------------------------------
 * Open a new connection from pool.
 */

int
VCP_Open(struct conn_pool *cp, vtim_dur tmo, VCL_IP *ap, int *err)
{
        int r;
        vtim_mono h;

        CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
        AN(err);

        while (cp->holddown > 0) {
                Lck_Lock(&cp->mtx);
                if (cp->holddown == 0) {
                        Lck_Unlock(&cp->mtx);
                        break;
                }

                if (VTIM_mono() >= cp->holddown) {
                        cp->holddown = 0;
                        Lck_Unlock(&cp->mtx);
                        break;
                }

                *err = 0;
                errno = cp->holddown_errno;
                Lck_Unlock(&cp->mtx);
                return (-1);
        }

        *err = errno = 0;
        r = cp->methods->open(cp, tmo, ap);

        if (r >= 0 && errno == 0 && cp->endpoint->preamble != NULL &&
             cp->endpoint->preamble->len > 0) {
                if (write(r, cp->endpoint->preamble->blob,
                    cp->endpoint->preamble->len) !=
                    cp->endpoint->preamble->len) {
                        *err = errno;
                        closefd(&r);
                }
        } else {
                *err = errno;
        }

        if (r >= 0)
                return (r);

        h = 0;

        switch (errno) {
        case EACCES:
        case EPERM:
                h = cache_param->backend_local_error_holddown;
                break;
        case EADDRNOTAVAIL:
                h = cache_param->backend_local_error_holddown;
                break;
        case ECONNREFUSED:
                h = cache_param->backend_remote_error_holddown;
                break;
        case ENETUNREACH:
                h = cache_param->backend_remote_error_holddown;
                break;
        default:
                break;
        }

        if (h == 0)
                return (r);

        Lck_Lock(&cp->mtx);
        h += VTIM_mono();
        if (cp->holddown == 0 || h < cp->holddown) {
                cp->holddown = h;
                cp->holddown_errno = errno;
        }

        Lck_Unlock(&cp->mtx);

        return (r);
}

/*--------------------------------------------------------------------
 * Close a connection.
 */

void
VCP_Close(struct pfd **pfdp)
{
        struct pfd *pfd;
        struct conn_pool *cp;

        TAKE_OBJ_NOTNULL(pfd, pfdp, PFD_MAGIC);
        cp = pfd->conn_pool;
        CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);

        assert(pfd->fd > 0);

        Lck_Lock(&cp->mtx);
        assert(pfd->state == PFD_STATE_USED || pfd->state == PFD_STATE_STOLEN);
        cp->n_used--;
        if (pfd->state == PFD_STATE_STOLEN) {
                (void)shutdown(pfd->fd, SHUT_RDWR);
                VTAILQ_REMOVE(&cp->connlist, pfd, list);
                pfd->state = PFD_STATE_CLEANUP;
                cp->n_kill++;
        } else {
                assert(pfd->state == PFD_STATE_USED);
                cp->methods->close(pfd);
                memset(pfd, 0x44, sizeof *pfd);
                free(pfd);
        }
        Lck_Unlock(&cp->mtx);
}

/*--------------------------------------------------------------------
 * Get a connection, possibly recycled
 */

struct pfd *
VCP_Get(struct conn_pool *cp, vtim_dur tmo, struct worker *wrk,
    unsigned force_fresh, int *err)
{
        struct pfd *pfd;

        CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
        CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
        AN(err);

        *err = 0;
        Lck_Lock(&cp->mtx);
        pfd = VTAILQ_FIRST(&cp->connlist);
        CHECK_OBJ_ORNULL(pfd, PFD_MAGIC);
        if (force_fresh || pfd == NULL || pfd->state == PFD_STATE_STOLEN) {
                pfd = NULL;
        } else {
                assert(pfd->conn_pool == cp);
                assert(pfd->state == PFD_STATE_AVAIL);
                VTAILQ_REMOVE(&cp->connlist, pfd, list);
                VTAILQ_INSERT_TAIL(&cp->connlist, pfd, list);
                cp->n_conn--;
                VSC_C_main->backend_reuse++;
                pfd->state = PFD_STATE_STOLEN;
                pfd->cond = &wrk->cond;
        }
        cp->n_used++;                   // Opening mostly works
        Lck_Unlock(&cp->mtx);

        if (pfd != NULL)
                return (pfd);

        ALLOC_OBJ(pfd, PFD_MAGIC);
        AN(pfd);
        INIT_OBJ(pfd->waited, WAITED_MAGIC);
        pfd->state = PFD_STATE_USED;
        pfd->conn_pool = cp;
        pfd->fd = VCP_Open(cp, tmo, &pfd->addr, err);
        if (pfd->fd < 0) {
                FREE_OBJ(pfd);
                Lck_Lock(&cp->mtx);
                cp->n_used--;           // Nope, didn't work after all.
                Lck_Unlock(&cp->mtx);
        } else
                VSC_C_main->backend_conn++;

        return (pfd);
}

/*--------------------------------------------------------------------
 */

int
VCP_Wait(struct worker *wrk, struct pfd *pfd, vtim_real when)
{
        struct conn_pool *cp;
        int r;

        CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
        CHECK_OBJ_NOTNULL(pfd, PFD_MAGIC);
        cp = pfd->conn_pool;
        CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
        assert(pfd->cond == &wrk->cond);
        Lck_Lock(&cp->mtx);
        while (pfd->state == PFD_STATE_STOLEN) {
                r = Lck_CondWaitUntil(&wrk->cond, &cp->mtx, when);
                if (r != 0) {
                        if (r == EINTR)
                                continue;
                        assert(r == ETIMEDOUT);
                        Lck_Unlock(&cp->mtx);
                        return (1);
                }
        }
        assert(pfd->state == PFD_STATE_USED);
        pfd->cond = NULL;
        Lck_Unlock(&cp->mtx);

        return (0);
}

/*--------------------------------------------------------------------
 */

VCL_IP
VCP_GetIp(struct pfd *pfd)
{

        CHECK_OBJ_NOTNULL(pfd, PFD_MAGIC);
        return (pfd->addr);
}

/*--------------------------------------------------------------------*/

static void
vcp_panic_endpoint(struct vsb *vsb, const struct vrt_endpoint *vep)
{
        char h[VTCP_ADDRBUFSIZE];
        char p[VTCP_PORTBUFSIZE];

        if (PAN_dump_struct(vsb, vep, VRT_ENDPOINT_MAGIC, "vrt_endpoint"))
                return;
        if (vep->uds_path)
                VSB_printf(vsb, "uds_path = %s,\n", vep->uds_path);
        if (vep->ipv4 && VSA_Sane(vep->ipv4)) {
                VTCP_name(vep->ipv4, h, sizeof h, p, sizeof p);
                VSB_printf(vsb, "ipv4 = %s, ", h);
                VSB_printf(vsb, "port = %s,\n", p);
        }
        if (vep->ipv6 && VSA_Sane(vep->ipv6)) {
                VTCP_name(vep->ipv6, h, sizeof h, p, sizeof p);
                VSB_printf(vsb, "ipv6 = %s, ", h);
                VSB_printf(vsb, "port = %s,\n", p);
        }
        VSB_indent(vsb, -2);
        VSB_cat(vsb, "},\n");
}

void
VCP_Panic(struct vsb *vsb, struct conn_pool *cp)
{

        if (PAN_dump_struct(vsb, cp, CONN_POOL_MAGIC, "conn_pool"))
                return;
        VSB_cat(vsb, "ident = ");
        VSB_quote(vsb, cp->ident, VSHA256_DIGEST_LENGTH, VSB_QUOTE_HEX);
        VSB_cat(vsb, ",\n");
        vcp_panic_endpoint(vsb, cp->endpoint);
        VSB_indent(vsb, -2);
        VSB_cat(vsb, "},\n");
}

/*--------------------------------------------------------------------*/

void
VCP_Init(void)
{
        Lck_New(&conn_pools_mtx, lck_conn_pool);
        Lck_New(&dead_pools_mtx, lck_dead_pool);
}

/**********************************************************************/

static inline int
tmo2msec(vtim_dur tmo)
{
        return ((int)floor(tmo * 1000.0));
}

static int v_matchproto_(cp_open_f)
vtp_open(const struct conn_pool *cp, vtim_dur tmo, VCL_IP *ap)
{
        int s;
        int msec;

        CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);

        msec = tmo2msec(tmo);
        if (cache_param->prefer_ipv6) {
                *ap = cp->endpoint->ipv6;
                s = VTCP_connect(*ap, msec);
                if (s >= 0)
                        return (s);
        }
        *ap = cp->endpoint->ipv4;
        s = VTCP_connect(*ap, msec);
        if (s >= 0)
                return (s);
        if (!cache_param->prefer_ipv6) {
                *ap = cp->endpoint->ipv6;
                s = VTCP_connect(*ap, msec);
        }
        return (s);
}


/*--------------------------------------------------------------------*/

static void v_matchproto_(cp_close_f)
vtp_close(struct pfd *pfd)
{

        CHECK_OBJ_NOTNULL(pfd, PFD_MAGIC);
        VTCP_close(&pfd->fd);
}

static void v_matchproto_(cp_name_f)
vtp_local_name(const struct pfd *pfd, char *addr, unsigned alen, char *pbuf,
               unsigned plen)
{
        CHECK_OBJ_NOTNULL(pfd, PFD_MAGIC);
        VTCP_myname(pfd->fd, addr, alen, pbuf, plen);
}

static void v_matchproto_(cp_name_f)
vtp_remote_name(const struct pfd *pfd, char *addr, unsigned alen, char *pbuf,
                unsigned plen)
{
        CHECK_OBJ_NOTNULL(pfd, PFD_MAGIC);
        VTCP_hisname(pfd->fd, addr, alen, pbuf, plen);
}

static const struct cp_methods vtp_methods = {
        .open = vtp_open,
        .close = vtp_close,
        .local_name = vtp_local_name,
        .remote_name = vtp_remote_name,
};

/*--------------------------------------------------------------------
 */

static int v_matchproto_(cp_open_f)
vus_open(const struct conn_pool *cp, vtim_dur tmo, VCL_IP *ap)
{
        int s;
        int msec;

        CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
        AN(cp->endpoint->uds_path);

        msec = tmo2msec(tmo);
        *ap = bogo_ip;
        s = VUS_connect(cp->endpoint->uds_path, msec);
        return (s);
}

static void v_matchproto_(cp_name_f)
vus_name(const struct pfd *pfd, char *addr, unsigned alen, char *pbuf,
         unsigned plen)
{
        (void) pfd;
        assert(alen > strlen("0.0.0.0"));
        assert(plen > 1);
        strcpy(addr, "0.0.0.0");
        strcpy(pbuf, "0");
}

static const struct cp_methods vus_methods = {
        .open = vus_open,
        .close = vtp_close,
        .local_name = vus_name,
        .remote_name = vus_name,
};

/*--------------------------------------------------------------------
 * Reference a TCP pool given by {ip4, ip6} pair or a UDS.  Create if
 * it doesn't exist already.
 */

struct conn_pool *
VCP_Ref(const struct vrt_endpoint *vep, const char *ident)
{
        struct conn_pool *cp, *cp2;
        struct VSHA256Context cx[1];
        unsigned char digest[VSHA256_DIGEST_LENGTH];

        CHECK_OBJ_NOTNULL(vep, VRT_ENDPOINT_MAGIC);
        AN(ident);
        VSHA256_Init(cx);
        VSHA256_Update(cx, ident, strlen(ident) + 1); // include \0
        if (vep->uds_path != NULL) {
                AZ(vep->ipv4);
                AZ(vep->ipv6);
                VSHA256_Update(cx, "UDS", 4); // include \0
                VSHA256_Update(cx, vep->uds_path, strlen(vep->uds_path));
        } else {
                assert(vep->ipv4 != NULL || vep->ipv6 != NULL);
                if (vep->ipv4 != NULL) {
                        assert(VSA_Sane(vep->ipv4));
                        VSHA256_Update(cx, "IP4", 4); // include \0
                        VSHA256_Update(cx, vep->ipv4, vsa_suckaddr_len);
                }
                if (vep->ipv6 != NULL) {
                        assert(VSA_Sane(vep->ipv6));
                        VSHA256_Update(cx, "IP6", 4); // include \0
                        VSHA256_Update(cx, vep->ipv6, vsa_suckaddr_len);
                }
        }
        if (vep->preamble != NULL && vep->preamble->len > 0) {
                VSHA256_Update(cx, "PRE", 4); // include \0
                VSHA256_Update(cx, vep->preamble->blob, vep->preamble->len);
        }
        VSHA256_Final(digest, cx);

        ALLOC_OBJ(cp, CONN_POOL_MAGIC);
        AN(cp);
        cp->refcnt = 1;
        cp->holddown = 0;
        cp->endpoint = VRT_Endpoint_Clone(vep);
        CHECK_OBJ_NOTNULL(cp->endpoint, VRT_ENDPOINT_MAGIC);
        memcpy(cp->ident, digest, sizeof cp->ident);
        if (vep->uds_path != NULL)
                cp->methods = &vus_methods;
        else
                cp->methods = &vtp_methods;
        Lck_New(&cp->mtx, lck_conn_pool);
        VTAILQ_INIT(&cp->connlist);

        CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
        Lck_Lock(&conn_pools_mtx);
        cp2 = VRBT_FIND(vrb, &conn_pools, cp);
        if (cp2 == NULL)
                AZ(VRBT_INSERT(vrb, &conn_pools, cp));
        else {
                CHECK_OBJ(cp2, CONN_POOL_MAGIC);
                assert(cp2->refcnt > 0);
                cp2->refcnt++;
        }
        Lck_Unlock(&conn_pools_mtx);

        if (cp2 == NULL) {
                CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
                return (cp);
        }

        Lck_Delete(&cp->mtx);
        AZ(cp->n_conn);
        AZ(cp->n_kill);
        FREE_OBJ(cp->endpoint);
        FREE_OBJ(cp);
        CHECK_OBJ_NOTNULL(cp2, CONN_POOL_MAGIC);
        return (cp2);
}

		varnish-cache/bin/varnishd/cache/cache_conn_pool.c
0		/*-
1		* Copyright (c) 2015 Varnish Software AS
2		* All rights reserved.
3		*
4		* Author: Poul-Henning Kamp <phk@phk.freebsd.dk>
5		*
6		* SPDX-License-Identifier: BSD-2-Clause
7		*
8		* Redistribution and use in source and binary forms, with or without
9		* modification, are permitted provided that the following conditions
10		* are met:
11		* 1. Redistributions of source code must retain the above copyright
12		* notice, this list of conditions and the following disclaimer.
13		* 2. Redistributions in binary form must reproduce the above copyright
14		* notice, this list of conditions and the following disclaimer in the
15		* documentation and/or other materials provided with the distribution.
16		*
17		* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18		* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19		* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20		* ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
21		* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22		* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23		* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24		* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25		* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26		* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27		* SUCH DAMAGE.
28		*
29		* (TCP\|UDS) connection pools.
30		*
31		*/
32
33		#include "config.h"
34
35		#include <stdlib.h>
36
37		#include "cache_varnishd.h"
38
39		#include "vsa.h"
40		#include "vsha256.h"
41		#include "vtcp.h"
42		#include "vus.h"
43		#include "vtim.h"
44		#include "waiter/waiter.h"
45
46		#include "cache_conn_pool.h"
47		#include "cache_pool.h"
48
49		struct conn_pool;
50		static inline int vcp_cmp(const struct conn_pool a, const struct conn_pool b);
51
52		/*--------------------------------------------------------------------
53		*/
54
55		struct pfd {
56		unsigned magic;
57		#define PFD_MAGIC 0x0c5e6593
58		int fd;
59		VTAILQ_ENTRY(pfd) list;
60		VCL_IP addr;
61		uint8_t state;
62		struct waited waited[1];
63		struct conn_pool *conn_pool;
64
65		pthread_cond_t *cond;
66		};
67
68		/*--------------------------------------------------------------------
69		*/
70
71		typedef int cp_open_f(const struct conn_pool , vtim_dur tmo, VCL_IP ap);
72		typedef void cp_close_f(struct pfd *);
73		typedef void cp_name_f(const struct pfd , char , unsigned, char *, unsigned);
74
75		struct cp_methods {
76		cp_open_f *open;
77		cp_close_f *close;
78		cp_name_f *local_name;
79		cp_name_f *remote_name;
80		};
81
82		struct conn_pool {
83		unsigned magic;
84		#define CONN_POOL_MAGIC 0x85099bc3
85
86		const struct cp_methods *methods;
87
88		struct vrt_endpoint *endpoint;
89		char ident[VSHA256_DIGEST_LENGTH];
90
91		VRBT_ENTRY(conn_pool) entry;
92		int refcnt;
93		struct lock mtx;
94
95		VTAILQ_HEAD(, pfd) connlist;
96		int n_conn;
97
98		int n_kill;
99
100		int n_used;
101
102		vtim_mono holddown;
103		int holddown_errno;
104		};
105
106		static struct lock conn_pools_mtx;
107		static struct lock dead_pools_mtx;
108
109		VRBT_HEAD(vrb, conn_pool);
110	66	VRBT_GENERATE_REMOVE_COLOR(vrb, conn_pool, entry, static)
111	161	VRBT_GENERATE_REMOVE(vrb, conn_pool, entry, static)
112	6184	VRBT_GENERATE_FIND(vrb, conn_pool, entry, vcp_cmp, static)
113	867	VRBT_GENERATE_INSERT_COLOR(vrb, conn_pool, entry, static)
114	4080	VRBT_GENERATE_INSERT_FINISH(vrb, conn_pool, entry, static)
115	4957	VRBT_GENERATE_INSERT(vrb, conn_pool, entry, vcp_cmp, static)
116	8	VRBT_GENERATE_NEXT(vrb, conn_pool, entry, static)
117	16	VRBT_GENERATE_MINMAX(vrb, conn_pool, entry, static)
118
119		static struct vrb conn_pools = VRBT_INITIALIZER(&conn_pools);
120		static struct vrb dead_pools = VRBT_INITIALIZER(&dying_cps);
121
122		/*--------------------------------------------------------------------
123		*/
124
125		unsigned
126	40353	PFD_State(const struct pfd *p)
127		{
128	40353	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
129	40353	return (p->state);
130		}
131
132		int *
133	17052	PFD_Fd(struct pfd *p)
134		{
135	17052	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
136	17052	return (&(p->fd));
137		}
138
139		void
140	8480	PFD_LocalName(const struct pfd p, char abuf, unsigned alen, char *pbuf,
141		unsigned plen)
142		{
143	8480	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
144	8480	CHECK_OBJ_NOTNULL(p->conn_pool, CONN_POOL_MAGIC);
145	8480	p->conn_pool->methods->local_name(p, abuf, alen, pbuf, plen);
146	8480	}
147
148		void
149	8480	PFD_RemoteName(const struct pfd p, char abuf, unsigned alen, char *pbuf,
150		unsigned plen)
151		{
152	8480	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
153	8480	CHECK_OBJ_NOTNULL(p->conn_pool, CONN_POOL_MAGIC);
154	8480	p->conn_pool->methods->remote_name(p, abuf, alen, pbuf, plen);
155	8480	}
156
157		/*--------------------------------------------------------------------
158		*/
159
160		static inline int
161	2989	vcp_cmp(const struct conn_pool a, const struct conn_pool b)
162		{
163	2989	return (memcmp(a->ident, b->ident, sizeof b->ident));
164		}
165
166		/*--------------------------------------------------------------------
167		* Waiter-handler
168		*/
169
170		static void v_matchproto_(waiter_handle_f)
171	6877	vcp_handle(struct waited *w, enum wait_event ev, vtim_real now)
172		{
173		struct pfd *pfd;
174		struct conn_pool *cp;
175
176	6877	CHECK_OBJ_NOTNULL(w, WAITED_MAGIC);
177	6877	CAST_OBJ_NOTNULL(pfd, w->priv1, PFD_MAGIC);
178	6877	(void)ev;
179	6877	(void)now;
180	6877	CHECK_OBJ_NOTNULL(pfd->conn_pool, CONN_POOL_MAGIC);
181	6877	cp = pfd->conn_pool;
182
183	6877	Lck_Lock(&cp->mtx);
184
185	6877	switch (pfd->state) {
186		case PFD_STATE_STOLEN:
187	3152	pfd->state = PFD_STATE_USED;
188	3152	VTAILQ_REMOVE(&cp->connlist, pfd, list);
189	3152	AN(pfd->cond);
190	3152	PTOK(pthread_cond_signal(pfd->cond));
191	3152	break;
192		case PFD_STATE_AVAIL:
193	3709	cp->methods->close(pfd);
194	3709	VTAILQ_REMOVE(&cp->connlist, pfd, list);
195	3709	cp->n_conn--;
196	3709	FREE_OBJ(pfd);
197	3709	break;
198		case PFD_STATE_CLEANUP:
199	16	cp->methods->close(pfd);
200	16	cp->n_kill--;
201	16	memset(pfd, 0x11, sizeof *pfd);
202	16	free(pfd);
203	16	break;
204		default:
205	0	WRONG("Wrong pfd state");
206	0	}
207	6877	Lck_Unlock(&cp->mtx);
208	6877	}
209
210
211		/*--------------------------------------------------------------------
212		*/
213
214		void
215	148	VCP_AddRef(struct conn_pool *cp)
216		{
217	148	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
218
219	148	Lck_Lock(&conn_pools_mtx);
220	148	assert(cp->refcnt > 0);
221	148	cp->refcnt++;
222	148	Lck_Unlock(&conn_pools_mtx);
223	148	}
224
225		/*--------------------------------------------------------------------
226		*/
227
228		static void
229	108	vcp_destroy(struct conn_pool **cpp)
230		{
231		struct conn_pool *cp;
232
233	108	TAKE_OBJ_NOTNULL(cp, cpp, CONN_POOL_MAGIC);
234	108	AZ(cp->n_conn);
235	108	AZ(cp->n_kill);
236	108	Lck_Delete(&cp->mtx);
237	108	free(cp->endpoint);
238	108	FREE_OBJ(cp);
239	108	}
240
241		/*--------------------------------------------------------------------
242		* Release Conn pool, destroy if last reference.
243		*/
244
245		void
246	347	VCP_Rel(struct conn_pool **cpp)
247		{
248		struct conn_pool *cp;
249		struct pfd pfd, pfd2;
250		int n_kill;
251
252	347	TAKE_OBJ_NOTNULL(cp, cpp, CONN_POOL_MAGIC);
253
254	347	Lck_Lock(&conn_pools_mtx);
255	347	assert(cp->refcnt > 0);
256	347	if (--cp->refcnt > 0) {
257	239	Lck_Unlock(&conn_pools_mtx);
258	239	return;
259		}
260	108	AZ(cp->n_used);
261	108	VRBT_REMOVE(vrb, &conn_pools, cp);
262	108	Lck_Unlock(&conn_pools_mtx);
263
264	108	Lck_Lock(&cp->mtx);
265	116	VTAILQ_FOREACH_SAFE(pfd, &cp->connlist, list, pfd2) {
266	8	VTAILQ_REMOVE(&cp->connlist, pfd, list);
267	8	cp->n_conn--;
268	8	assert(pfd->state == PFD_STATE_AVAIL);
269	8	pfd->state = PFD_STATE_CLEANUP;
270	8	(void)shutdown(pfd->fd, SHUT_RDWR);
271	8	cp->n_kill++;
272	8	}
273	108	n_kill = cp->n_kill;
274	108	Lck_Unlock(&cp->mtx);
275	108	if (n_kill == 0) {
276	100	vcp_destroy(&cp);
277	100	return;
278		}
279	8	Lck_Lock(&dead_pools_mtx);
280		/*
281		* Here we reuse cp's entry but it will probably not be correctly
282		* indexed because of the hack in VCP_RelPoll
283		*/
284	8	VRBT_INSERT(vrb, &dead_pools, cp);
285	8	Lck_Unlock(&dead_pools_mtx);
286	347	}
287
288		void
289	30265	VCP_RelPoll(void)
290		{
291		struct vrb dead;
292		struct conn_pool cp, cp2;
293		int n_kill;
294
295	30265	ASSERT_CLI();
296
297	30265	Lck_Lock(&dead_pools_mtx);
298	30265	if (VRBT_EMPTY(&dead_pools)) {
299	30257	Lck_Unlock(&dead_pools_mtx);
300	30257	return;
301		}
302	8	dead = dead_pools;
303	8	VRBT_INIT(&dead_pools);
304	8	Lck_Unlock(&dead_pools_mtx);
305
306	16	VRBT_FOREACH_SAFE(cp, vrb, &dead, cp2) {
307	8	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
308	8	Lck_Lock(&cp->mtx);
309	8	n_kill = cp->n_kill;
310	8	Lck_Unlock(&cp->mtx);
311	8	if (n_kill > 0)
312	0	continue;
313	8	VRBT_REMOVE(vrb, &dead, cp);
314	8	vcp_destroy(&cp);
315	8	}
316
317	8	if (VRBT_EMPTY(&dead))
318	8	return;
319
320	0	Lck_Lock(&dead_pools_mtx);
321		/*
322		* The following insertion will most likely result in an
323		* unordered tree, but in this case it does not matter
324		* as we just want to iterate over all the elements
325		* in the tree in order to delete them.
326		*/
327	0	VRBT_INSERT(vrb, &dead_pools, dead.rbh_root);
328	0	Lck_Unlock(&dead_pools_mtx);
329	30265	}
330
331		/*--------------------------------------------------------------------
332		* Recycle a connection.
333		*/
334
335		void
336	6957	VCP_Recycle(const struct worker wrk, struct pfd *pfdp)
337		{
338		struct pfd *pfd;
339		struct conn_pool *cp;
340	6957	int i = 0;
341
342	6957	CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
343	6957	TAKE_OBJ_NOTNULL(pfd, pfdp, PFD_MAGIC);
344	6957	cp = pfd->conn_pool;
345	6957	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
346
347	6957	assert(pfd->state == PFD_STATE_USED);
348	6957	assert(pfd->fd > 0);
349
350	6957	Lck_Lock(&cp->mtx);
351	6957	cp->n_used--;
352
353	6957	pfd->waited->priv1 = pfd;
354	6957	pfd->waited->fd = pfd->fd;
355	6957	pfd->waited->idle = VTIM_real();
356	6957	pfd->state = PFD_STATE_AVAIL;
357	6957	pfd->waited->func = vcp_handle;
358	6957	pfd->waited->tmo = cache_param->backend_idle_timeout;
359	6957	if (Wait_Enter(wrk->pool->waiter, pfd->waited)) {
360	0	cp->methods->close(pfd);
361	0	memset(pfd, 0x33, sizeof *pfd);
362	0	free(pfd);
363		// XXX: stats
364	0	pfd = NULL;
365	0	} else {
366	6957	VTAILQ_INSERT_HEAD(&cp->connlist, pfd, list);
367	6957	i++;
368		}
369
370	6957	if (pfd != NULL)
371	6957	cp->n_conn++;
372	6957	Lck_Unlock(&cp->mtx);
373
374	6957	if (i && DO_DEBUG(DBG_VTC_MODE)) {
375		/*
376		* In varnishtest we do not have the luxury of using
377		* multiple backend connections, so whenever we end up
378		* in the "pending" case, take a short nap to let the
379		* waiter catch up and put the pfd back into circulations.
380		*
381		* In particular ESI:include related tests suffer random
382		* failures without this.
383		*
384		* In normal operation, the only effect is that we will
385		* have N+1 backend connections rather than N, which is
386		* entirely harmless.
387		*/
388	6957	VTIM_sleep(0.01);
389	6957	}
390	6957	}
391
392		/*--------------------------------------------------------------------
393		* Open a new connection from pool.
394		*/
395
396		int
397	6222	VCP_Open(struct conn_pool cp, vtim_dur tmo, VCL_IP ap, int *err)
398		{
399		int r;
400		vtim_mono h;
401
402	6222	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
403	6222	AN(err);
404
405	6222	while (cp->holddown > 0) {
406	88	Lck_Lock(&cp->mtx);
407	88	if (cp->holddown == 0) {
408	0	Lck_Unlock(&cp->mtx);
409	0	break;
410		}
411
412	88	if (VTIM_mono() >= cp->holddown) {
413	11	cp->holddown = 0;
414	11	Lck_Unlock(&cp->mtx);
415	11	break;
416		}
417
418	77	*err = 0;
419	77	errno = cp->holddown_errno;
420	77	Lck_Unlock(&cp->mtx);
421	77	return (-1);
422		}
423
424	6145	*err = errno = 0;
425	6145	r = cp->methods->open(cp, tmo, ap);
426
427	6145	if (r >= 0 && errno == 0 && cp->endpoint->preamble != NULL &&
428	36	cp->endpoint->preamble->len > 0) {
429	108	if (write(r, cp->endpoint->preamble->blob,
430	72	cp->endpoint->preamble->len) !=
431	36	cp->endpoint->preamble->len) {
432	0	*err = errno;
433	0	closefd(&r);
434	0	}
435	36	} else {
436	6109	*err = errno;
437		}
438
439	6145	if (r >= 0)
440	6045	return (r);
441
442	100	h = 0;
443
444	100	switch (errno) {
445		case EACCES:
446		case EPERM:
447	0	h = cache_param->backend_local_error_holddown;
448	0	break;
449		case EADDRNOTAVAIL:
450	0	h = cache_param->backend_local_error_holddown;
451	0	break;
452		case ECONNREFUSED:
453	100	h = cache_param->backend_remote_error_holddown;
454	100	break;
455		case ENETUNREACH:
456	0	h = cache_param->backend_remote_error_holddown;
457	0	break;
458		default:
459	0	break;
460		}
461
462	100	if (h == 0)
463	0	return (r);
464
465	100	Lck_Lock(&cp->mtx);
466	100	h += VTIM_mono();
467	100	if (cp->holddown == 0 \|\| h < cp->holddown) {
468	89	cp->holddown = h;
469	89	cp->holddown_errno = errno;
470	89	}
471
472	100	Lck_Unlock(&cp->mtx);
473
474	100	return (r);
475	6222	}
476
477		/*--------------------------------------------------------------------
478		* Close a connection.
479		*/
480
481		void
482	1519	VCP_Close(struct pfd **pfdp)
483		{
484		struct pfd *pfd;
485		struct conn_pool *cp;
486
487	1519	TAKE_OBJ_NOTNULL(pfd, pfdp, PFD_MAGIC);
488	1519	cp = pfd->conn_pool;
489	1519	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
490
491	1519	assert(pfd->fd > 0);
492
493	1519	Lck_Lock(&cp->mtx);
494	1519	assert(pfd->state == PFD_STATE_USED \|\| pfd->state == PFD_STATE_STOLEN);
495	1519	cp->n_used--;
496	1519	if (pfd->state == PFD_STATE_STOLEN) {
497	8	(void)shutdown(pfd->fd, SHUT_RDWR);
498	8	VTAILQ_REMOVE(&cp->connlist, pfd, list);
499	8	pfd->state = PFD_STATE_CLEANUP;