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;

        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++;
        }
        Lck_Unlock(&cp->mtx);
        if (cp->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;

        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);
                if (cp->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.
                 */
                (void)usleep(10000);
        }
}

/*--------------------------------------------------------------------
 * 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	82	VRBT_GENERATE_REMOVE_COLOR(vrb, conn_pool, entry, static)
111	215	VRBT_GENERATE_REMOVE(vrb, conn_pool, entry, static)
112	7708	VRBT_GENERATE_FIND(vrb, conn_pool, entry, vcp_cmp, static)
113	1089	VRBT_GENERATE_INSERT_COLOR(vrb, conn_pool, entry, static)
114	5077	VRBT_GENERATE_INSERT_FINISH(vrb, conn_pool, entry, static)
115	6172	VRBT_GENERATE_INSERT(vrb, conn_pool, entry, vcp_cmp, static)
116	12	VRBT_GENERATE_NEXT(vrb, conn_pool, entry, static)
117	24	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	49957	PFD_State(const struct pfd *p)
127		{
128	49957	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
129	49957	return (p->state);
130		}
131
132		int *
133	21125	PFD_Fd(struct pfd *p)
134		{
135	21125	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
136	21125	return (&(p->fd));
137		}
138
139		void
140	10505	PFD_LocalName(const struct pfd p, char abuf, unsigned alen, char *pbuf,
141		unsigned plen)
142		{
143	10505	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
144	10505	CHECK_OBJ_NOTNULL(p->conn_pool, CONN_POOL_MAGIC);
145	10505	p->conn_pool->methods->local_name(p, abuf, alen, pbuf, plen);
146	10505	}
147
148		void
149	10505	PFD_RemoteName(const struct pfd p, char abuf, unsigned alen, char *pbuf,
150		unsigned plen)
151		{
152	10505	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
153	10505	CHECK_OBJ_NOTNULL(p->conn_pool, CONN_POOL_MAGIC);
154	10505	p->conn_pool->methods->remote_name(p, abuf, alen, pbuf, plen);
155	10505	}
156
157		/*--------------------------------------------------------------------
158		*/
159
160		static inline int
161	3738	vcp_cmp(const struct conn_pool a, const struct conn_pool b)
162		{
163	3738	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	8494	vcp_handle(struct waited *w, enum wait_event ev, vtim_real now)
172		{
173		struct pfd *pfd;
174		struct conn_pool *cp;
175
176	8494	CHECK_OBJ_NOTNULL(w, WAITED_MAGIC);
177	8494	CAST_OBJ_NOTNULL(pfd, w->priv1, PFD_MAGIC);
178	8494	(void)ev;
179	8494	(void)now;
180	8494	CHECK_OBJ_NOTNULL(pfd->conn_pool, CONN_POOL_MAGIC);
181	8494	cp = pfd->conn_pool;
182
183	8494	Lck_Lock(&cp->mtx);
184
185	8494	switch (pfd->state) {
186		case PFD_STATE_STOLEN:
187	3874	pfd->state = PFD_STATE_USED;
188	3874	VTAILQ_REMOVE(&cp->connlist, pfd, list);
189	3874	AN(pfd->cond);
190	3874	PTOK(pthread_cond_signal(pfd->cond));
191	3874	break;
192		case PFD_STATE_AVAIL:
193	4600	cp->methods->close(pfd);
194	4600	VTAILQ_REMOVE(&cp->connlist, pfd, list);
195	4600	cp->n_conn--;
196	4600	FREE_OBJ(pfd);
197	4600	break;
198		case PFD_STATE_CLEANUP:
199	20	cp->methods->close(pfd);
200	20	cp->n_kill--;
201	20	memset(pfd, 0x11, sizeof *pfd);
202	20	free(pfd);
203	20	break;
204		default:
205	0	WRONG("Wrong pfd state");
206	0	}
207	8494	Lck_Unlock(&cp->mtx);
208	8494	}
209
210
211		/*--------------------------------------------------------------------
212		*/
213
214		void
215	185	VCP_AddRef(struct conn_pool *cp)
216		{
217	185	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
218
219	185	Lck_Lock(&conn_pools_mtx);
220	185	assert(cp->refcnt > 0);
221	185	cp->refcnt++;
222	185	Lck_Unlock(&conn_pools_mtx);
223	185	}
224
225		/*--------------------------------------------------------------------
226		*/
227
228		static void
229	135	vcp_destroy(struct conn_pool **cpp)
230		{
231		struct conn_pool *cp;
232
233	135	TAKE_OBJ_NOTNULL(cp, cpp, CONN_POOL_MAGIC);
234	135	AZ(cp->n_conn);
235	135	AZ(cp->n_kill);
236	135	Lck_Delete(&cp->mtx);
237	135	free(cp->endpoint);
238	135	FREE_OBJ(cp);
239	135	}
240
241		/*--------------------------------------------------------------------
242		* Release Conn pool, destroy if last reference.
243		*/
244
245		void
246	429	VCP_Rel(struct conn_pool **cpp)
247		{
248		struct conn_pool *cp;
249		struct pfd pfd, pfd2;
250
251	429	TAKE_OBJ_NOTNULL(cp, cpp, CONN_POOL_MAGIC);
252
253	429	Lck_Lock(&conn_pools_mtx);
254	429	assert(cp->refcnt > 0);
255	429	if (--cp->refcnt > 0) {
256	294	Lck_Unlock(&conn_pools_mtx);
257	294	return;
258		}
259	135	AZ(cp->n_used);
260	135	VRBT_REMOVE(vrb, &conn_pools, cp);
261	135	Lck_Unlock(&conn_pools_mtx);
262
263	135	Lck_Lock(&cp->mtx);
264	145	VTAILQ_FOREACH_SAFE(pfd, &cp->connlist, list, pfd2) {
265	10	VTAILQ_REMOVE(&cp->connlist, pfd, list);
266	10	cp->n_conn--;
267	10	assert(pfd->state == PFD_STATE_AVAIL);
268	10	pfd->state = PFD_STATE_CLEANUP;
269	10	(void)shutdown(pfd->fd, SHUT_RDWR);
270	10	cp->n_kill++;
271	10	}
272	135	Lck_Unlock(&cp->mtx);
273	135	if (cp->n_kill == 0) {
274	125	vcp_destroy(&cp);
275	125	return;
276		}
277	10	Lck_Lock(&dead_pools_mtx);
278		/*
279		* Here we reuse cp's entry but it will probably not be correctly
280		* indexed because of the hack in VCP_RelPoll
281		*/
282	10	VRBT_INSERT(vrb, &dead_pools, cp);
283	10	Lck_Unlock(&dead_pools_mtx);
284	429	}
285
286		void
287	37604	VCP_RelPoll(void)
288		{
289		struct vrb dead;
290		struct conn_pool cp, cp2;
291
292	37604	ASSERT_CLI();
293
294	37604	Lck_Lock(&dead_pools_mtx);
295	37604	if (VRBT_EMPTY(&dead_pools)) {
296	37592	Lck_Unlock(&dead_pools_mtx);
297	37592	return;
298		}
299	12	dead = dead_pools;
300	12	VRBT_INIT(&dead_pools);
301	12	Lck_Unlock(&dead_pools_mtx);
302
303	24	VRBT_FOREACH_SAFE(cp, vrb, &dead, cp2) {
304	12	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
305	12	if (cp->n_kill > 0)
306	2	continue;
307	10	VRBT_REMOVE(vrb, &dead, cp);
308	10	vcp_destroy(&cp);
309	10	}
310
311	12	if (VRBT_EMPTY(&dead))
312	10	return;
313
314	2	Lck_Lock(&dead_pools_mtx);
315		/*
316		* The following insertion will most likely result in an
317		* unordered tree, but in this case it does not matter
318		* as we just want to iterate over all the elements
319		* in the tree in order to delete them.
320		*/
321	2	VRBT_INSERT(vrb, &dead_pools, dead.rbh_root);
322	2	Lck_Unlock(&dead_pools_mtx);
323	37604	}
324
325		/*--------------------------------------------------------------------
326		* Recycle a connection.
327		*/
328
329		void
330	8597	VCP_Recycle(const struct worker wrk, struct pfd *pfdp)
331		{
332		struct pfd *pfd;
333		struct conn_pool *cp;
334	8597	int i = 0;
335
336	8597	CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
337	8597	TAKE_OBJ_NOTNULL(pfd, pfdp, PFD_MAGIC);
338	8597	cp = pfd->conn_pool;
339	8597	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
340
341	8597	assert(pfd->state == PFD_STATE_USED);
342	8597	assert(pfd->fd > 0);
343
344	8597	Lck_Lock(&cp->mtx);
345	8597	cp->n_used--;
346
347	8597	pfd->waited->priv1 = pfd;
348	8597	pfd->waited->fd = pfd->fd;
349	8597	pfd->waited->idle = VTIM_real();
350	8597	pfd->state = PFD_STATE_AVAIL;
351	8597	pfd->waited->func = vcp_handle;
352	8597	pfd->waited->tmo = cache_param->backend_idle_timeout;
353	8597	if (Wait_Enter(wrk->pool->waiter, pfd->waited)) {
354	0	cp->methods->close(pfd);
355	0	memset(pfd, 0x33, sizeof *pfd);
356	0	free(pfd);
357		// XXX: stats
358	0	pfd = NULL;
359	0	} else {
360	8597	VTAILQ_INSERT_HEAD(&cp->connlist, pfd, list);
361	8597	i++;
362		}
363
364	8597	if (pfd != NULL)
365	8597	cp->n_conn++;
366	8597	Lck_Unlock(&cp->mtx);
367
368	8597	if (i && DO_DEBUG(DBG_VTC_MODE)) {
369		/*
370		* In varnishtest we do not have the luxury of using
371		* multiple backend connections, so whenever we end up
372		* in the "pending" case, take a short nap to let the
373		* waiter catch up and put the pfd back into circulations.
374		*
375		* In particular ESI:include related tests suffer random
376		* failures without this.
377		*
378		* In normal operation, the only effect is that we will
379		* have N+1 backend connections rather than N, which is
380		* entirely harmless.
381		*/
382	8597	(void)usleep(10000);
383	8597	}
384	8597	}
385
386		/*--------------------------------------------------------------------
387		* Open a new connection from pool.
388		*/
389
390		int
391	7763	VCP_Open(struct conn_pool cp, vtim_dur tmo, VCL_IP ap, int *err)
392		{
393		int r;
394		vtim_mono h;
395
396	7763	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
397	7763	AN(err);
398
399	7763	while (cp->holddown > 0) {
400	115	Lck_Lock(&cp->mtx);
401	115	if (cp->holddown == 0) {
402	0	Lck_Unlock(&cp->mtx);
403	0	break;
404		}
405
406	115	if (VTIM_mono() >= cp->holddown) {
407	16	cp->holddown = 0;
408	16	Lck_Unlock(&cp->mtx);
409	16	break;
410		}
411
412	99	*err = 0;
413	99	errno = cp->holddown_errno;
414	99	Lck_Unlock(&cp->mtx);
415	99	return (-1);
416		}
417
418	7664	*err = errno = 0;
419	7664	r = cp->methods->open(cp, tmo, ap);
420
421	7664	if (r >= 0 && errno == 0 && cp->endpoint->preamble != NULL &&
422	45	cp->endpoint->preamble->len > 0) {
423	135	if (write(r, cp->endpoint->preamble->blob,
424	90	cp->endpoint->preamble->len) !=
425	45	cp->endpoint->preamble->len) {
426	0	*err = errno;
427	0	closefd(&r);
428	0	}
429	45	} else {
430	7619	*err = errno;
431		}
432
433	7664	if (r >= 0)
434	7535	return (r);
435
436	129	h = 0;
437
438	129	switch (errno) {
439		case EACCES:
440		case EPERM:
441	0	h = cache_param->backend_local_error_holddown;
442	0	break;
443		case EADDRNOTAVAIL:
444	0	h = cache_param->backend_local_error_holddown;
445	0	break;
446		case ECONNREFUSED:
447	129	h = cache_param->backend_remote_error_holddown;
448	129	break;
449		case ENETUNREACH:
450	0	h = cache_param->backend_remote_error_holddown;
451	0	break;
452		default:
453	0	break;
454		}
455
456	129	if (h == 0)
457	0	return (r);
458
459	129	Lck_Lock(&cp->mtx);
460	129	h += VTIM_mono();
461	129	if (cp->holddown == 0 \|\| h < cp->holddown) {
462	116	cp->holddown = h;
463	116	cp->holddown_errno = errno;
464	116	}
465
466	129	Lck_Unlock(&cp->mtx);
467
468	129	return (r);
469	7763	}
470
471		/*--------------------------------------------------------------------
472		* Close a connection.
473		*/
474
475		void
476	1903	VCP_Close(struct pfd **pfdp)
477		{
478		struct pfd *pfd;
479		struct conn_pool *cp;
480
481	1903	TAKE_OBJ_NOTNULL(pfd, pfdp, PFD_MAGIC);
482	1903	cp = pfd->conn_pool;
483	1903	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
484
485	1903	assert(pfd->fd > 0);
486
487	1903	Lck_Lock(&cp->mtx);
488	1903	assert(pfd->state == PFD_STATE_USED \|\| pfd->state == PFD_STATE_STOLEN);
489	1903	cp->n_used--;
490	1903	if (pfd->state == PFD_STATE_STOLEN) {
491	10	(void)shutdown(pfd->fd, SHUT_RDWR);
492	10	VTAILQ_REMOVE(&cp->connlist, pfd, list);
493	10	pfd->state = PFD_STATE_CLEANUP;
494	10	cp->n_kill++;
495	10	} else {
496	1893	assert(pfd->state == PFD_STATE_USED);
497	1893	cp->methods->close(pfd);
498	1893	memset(pfd, 0x44, sizeof *pfd);
499	1893	free(pfd);