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;

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

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];

        VTAILQ_ENTRY(conn_pool)                 list;
        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 VTAILQ_HEAD(, conn_pool) conn_pools =
    VTAILQ_HEAD_INITIALIZER(conn_pools);


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

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);
}

/*--------------------------------------------------------------------
 * 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);
}

/*--------------------------------------------------------------------
 * 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);
        VTAILQ_REMOVE(&conn_pools, cp, list);
        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_WR);
                cp->n_kill++;
        }
        while (cp->n_kill) {
                Lck_Unlock(&cp->mtx);
                (void)usleep(20000);
                Lck_Lock(&cp->mtx);
        }
        Lck_Unlock(&cp->mtx);
        Lck_Delete(&cp->mtx);
        AZ(cp->n_conn);
        AZ(cp->n_kill);
        free(cp->endpoint);
        FREE_OBJ(cp);
}

/*--------------------------------------------------------------------
 * 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);
}

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

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);

        /*
         * In heavy use of dynamic backends, traversing this list
         * can become expensive.  In order to not do so twice we
         * pessimistically create the necessary pool, and discard
         * it on a hit. (XXX: Consider hash or tree ?)
         */

        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);
        VTAILQ_FOREACH(cp2, &conn_pools, list) {
                CHECK_OBJ_NOTNULL(cp2, CONN_POOL_MAGIC);
                assert(cp2->refcnt > 0);
                if (!memcmp(digest, cp2->ident, sizeof cp2->ident))
                        break;
        }
        if (cp2 == NULL)
                VTAILQ_INSERT_HEAD(&conn_pools, cp, list);
        else
                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
51		/*--------------------------------------------------------------------
52		*/
53
54		struct pfd {
55		unsigned magic;
56		#define PFD_MAGIC 0x0c5e6593
57		int fd;
58		VTAILQ_ENTRY(pfd) list;
59		VCL_IP addr;
60		uint8_t state;
61		struct waited waited[1];
62		struct conn_pool *conn_pool;
63
64		pthread_cond_t *cond;
65		};
66
67		/*--------------------------------------------------------------------
68		*/
69
70		typedef int cp_open_f(const struct conn_pool , vtim_dur tmo, VCL_IP ap);
71		typedef void cp_close_f(struct pfd *);
72		typedef void cp_name_f(const struct pfd , char , unsigned, char *, unsigned);
73
74		struct cp_methods {
75		cp_open_f *open;
76		cp_close_f *close;
77		cp_name_f *local_name;
78		cp_name_f *remote_name;
79		};
80
81		struct conn_pool {
82		unsigned magic;
83		#define CONN_POOL_MAGIC 0x85099bc3
84
85		const struct cp_methods *methods;
86
87		struct vrt_endpoint *endpoint;
88		char ident[VSHA256_DIGEST_LENGTH];
89
90		VTAILQ_ENTRY(conn_pool) list;
91		int refcnt;
92		struct lock mtx;
93
94		VTAILQ_HEAD(, pfd) connlist;
95		int n_conn;
96
97		int n_kill;
98
99		int n_used;
100
101		vtim_mono holddown;
102		int holddown_errno;
103		};
104
105		static struct lock conn_pools_mtx;
106
107		static VTAILQ_HEAD(, conn_pool) conn_pools =
108		VTAILQ_HEAD_INITIALIZER(conn_pools);
109
110
111		/*--------------------------------------------------------------------
112		*/
113
114		unsigned
115	244272	PFD_State(const struct pfd *p)
116		{
117	244272	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
118	244272	return (p->state);
119		}
120
121		int *
122	102848	PFD_Fd(struct pfd *p)
123		{
124	102848	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
125	102848	return (&(p->fd));
126		}
127
128		void
129	51150	PFD_LocalName(const struct pfd p, char abuf, unsigned alen, char *pbuf,
130		unsigned plen)
131		{
132	51150	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
133	51150	CHECK_OBJ_NOTNULL(p->conn_pool, CONN_POOL_MAGIC);
134	51150	p->conn_pool->methods->local_name(p, abuf, alen, pbuf, plen);
135	51150	}
136
137		void
138	51149	PFD_RemoteName(const struct pfd p, char abuf, unsigned alen, char *pbuf,
139		unsigned plen)
140		{
141	51149	CHECK_OBJ_NOTNULL(p, PFD_MAGIC);
142	51149	CHECK_OBJ_NOTNULL(p->conn_pool, CONN_POOL_MAGIC);
143	51149	p->conn_pool->methods->remote_name(p, abuf, alen, pbuf, plen);
144	51149	}
145
146		/*--------------------------------------------------------------------
147		* Waiter-handler
148		*/
149
150		static void v_matchproto_(waiter_handle_f)
151	41698	vcp_handle(struct waited *w, enum wait_event ev, vtim_real now)
152		{
153		struct pfd *pfd;
154		struct conn_pool *cp;
155
156	41698	CHECK_OBJ_NOTNULL(w, WAITED_MAGIC);
157	41698	CAST_OBJ_NOTNULL(pfd, w->priv1, PFD_MAGIC);
158	41698	(void)ev;
159	41698	(void)now;
160	41698	CHECK_OBJ_NOTNULL(pfd->conn_pool, CONN_POOL_MAGIC);
161	41698	cp = pfd->conn_pool;
162
163	41698	Lck_Lock(&cp->mtx);
164
165	41698	switch (pfd->state) {
166		case PFD_STATE_STOLEN:
167	18772	pfd->state = PFD_STATE_USED;
168	18772	VTAILQ_REMOVE(&cp->connlist, pfd, list);
169	18772	AN(pfd->cond);
170	18772	PTOK(pthread_cond_signal(pfd->cond));
171	18772	break;
172		case PFD_STATE_AVAIL:
173	22851	cp->methods->close(pfd);
174	22851	VTAILQ_REMOVE(&cp->connlist, pfd, list);
175	22851	cp->n_conn--;
176	22851	FREE_OBJ(pfd);
177	22851	break;
178		case PFD_STATE_CLEANUP:
179	75	cp->methods->close(pfd);
180	75	cp->n_kill--;
181	75	memset(pfd, 0x11, sizeof *pfd);
182	75	free(pfd);
183	75	break;
184		default:
185	0	WRONG("Wrong pfd state");
186	0	}
187	41698	Lck_Unlock(&cp->mtx);
188	41698	}
189
190
191		/*--------------------------------------------------------------------
192		*/
193
194		void
195	850	VCP_AddRef(struct conn_pool *cp)
196		{
197	850	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
198
199	850	Lck_Lock(&conn_pools_mtx);
200	850	assert(cp->refcnt > 0);
201	850	cp->refcnt++;
202	850	Lck_Unlock(&conn_pools_mtx);
203	850	}
204
205		/*--------------------------------------------------------------------
206		* Release Conn pool, destroy if last reference.
207		*/
208
209		void
210	2096	VCP_Rel(struct conn_pool **cpp)
211		{
212		struct conn_pool *cp;
213		struct pfd pfd, pfd2;
214
215	2096	TAKE_OBJ_NOTNULL(cp, cpp, CONN_POOL_MAGIC);
216
217	2096	Lck_Lock(&conn_pools_mtx);
218	2096	assert(cp->refcnt > 0);
219	2096	if (--cp->refcnt > 0) {
220	1446	Lck_Unlock(&conn_pools_mtx);
221	1446	return;
222		}
223	650	AZ(cp->n_used);
224	650	VTAILQ_REMOVE(&conn_pools, cp, list);
225	650	Lck_Unlock(&conn_pools_mtx);
226
227	650	Lck_Lock(&cp->mtx);
228	675	VTAILQ_FOREACH_SAFE(pfd, &cp->connlist, list, pfd2) {
229	25	VTAILQ_REMOVE(&cp->connlist, pfd, list);
230	25	cp->n_conn--;
231	25	assert(pfd->state == PFD_STATE_AVAIL);
232	25	pfd->state = PFD_STATE_CLEANUP;
233	25	(void)shutdown(pfd->fd, SHUT_WR);
234	25	cp->n_kill++;
235	25	}
236	675	while (cp->n_kill) {
237	25	Lck_Unlock(&cp->mtx);
238	25	(void)usleep(20000);
239	25	Lck_Lock(&cp->mtx);
240		}
241	650	Lck_Unlock(&cp->mtx);
242	650	Lck_Delete(&cp->mtx);
243	650	AZ(cp->n_conn);
244	650	AZ(cp->n_kill);
245	650	free(cp->endpoint);
246	650	FREE_OBJ(cp);
247	2096	}
248
249		/*--------------------------------------------------------------------
250		* Recycle a connection.
251		*/
252
253		void
254	42070	VCP_Recycle(const struct worker wrk, struct pfd *pfdp)
255		{
256		struct pfd *pfd;
257		struct conn_pool *cp;
258	42070	int i = 0;
259
260	42070	CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
261	42070	TAKE_OBJ_NOTNULL(pfd, pfdp, PFD_MAGIC);
262	42070	cp = pfd->conn_pool;
263	42070	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
264
265	42070	assert(pfd->state == PFD_STATE_USED);
266	42070	assert(pfd->fd > 0);
267
268	42070	Lck_Lock(&cp->mtx);
269	42070	cp->n_used--;
270
271	42070	pfd->waited->priv1 = pfd;
272	42070	pfd->waited->fd = pfd->fd;
273	42070	pfd->waited->idle = VTIM_real();
274	42070	pfd->state = PFD_STATE_AVAIL;
275	42070	pfd->waited->func = vcp_handle;
276	42070	pfd->waited->tmo = cache_param->backend_idle_timeout;
277	42070	if (Wait_Enter(wrk->pool->waiter, pfd->waited)) {
278	0	cp->methods->close(pfd);
279	0	memset(pfd, 0x33, sizeof *pfd);
280	0	free(pfd);
281		// XXX: stats
282	0	pfd = NULL;
283	0	} else {
284	42070	VTAILQ_INSERT_HEAD(&cp->connlist, pfd, list);
285	42070	i++;
286		}
287
288	42070	if (pfd != NULL)
289	42070	cp->n_conn++;
290	42070	Lck_Unlock(&cp->mtx);
291
292	42070	if (i && DO_DEBUG(DBG_VTC_MODE)) {
293		/*
294		* In varnishtest we do not have the luxury of using
295		* multiple backend connections, so whenever we end up
296		* in the "pending" case, take a short nap to let the
297		* waiter catch up and put the pfd back into circulations.
298		*
299		* In particular ESI:include related tests suffer random
300		* failures without this.
301		*
302		* In normal operation, the only effect is that we will
303		* have N+1 backend connections rather than N, which is
304		* entirely harmless.
305		*/
306	42070	(void)usleep(10000);
307	42070	}
308	42070	}
309
310		/*--------------------------------------------------------------------
311		* Open a new connection from pool.
312		*/
313
314		int
315	37875	VCP_Open(struct conn_pool cp, vtim_dur tmo, VCL_IP ap, int *err)
316		{
317		int r;
318		vtim_mono h;
319
320	37875	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
321	37875	AN(err);
322
323	37875	while (cp->holddown > 0) {
324	534	Lck_Lock(&cp->mtx);
325	534	if (cp->holddown == 0) {
326	0	Lck_Unlock(&cp->mtx);
327	0	break;
328		}
329
330	534	if (VTIM_mono() >= cp->holddown) {
331	65	cp->holddown = 0;
332	65	Lck_Unlock(&cp->mtx);
333	65	break;
334		}
335
336	469	*err = 0;
337	469	errno = cp->holddown_errno;
338	469	Lck_Unlock(&cp->mtx);
339	469	return (-1);
340		}
341
342	37406	*err = errno = 0;
343	37406	r = cp->methods->open(cp, tmo, ap);
344
345	37406	if (r >= 0 && errno == 0 && cp->endpoint->preamble != NULL &&
346	225	cp->endpoint->preamble->len > 0) {
347	675	if (write(r, cp->endpoint->preamble->blob,
348	450	cp->endpoint->preamble->len) !=
349	225	cp->endpoint->preamble->len) {
350	0	*err = errno;
351	0	closefd(&r);
352	0	}
353	225	} else {
354	37181	*err = errno;
355		}
356
357	37406	if (r >= 0)
358	36796	return (r);
359
360	610	h = 0;
361
362	610	switch (errno) {
363		case EACCES:
364		case EPERM:
365	0	h = cache_param->backend_local_error_holddown;
366	0	break;
367		case EADDRNOTAVAIL:
368	0	h = cache_param->backend_local_error_holddown;
369	0	break;
370		case ECONNREFUSED:
371	610	h = cache_param->backend_remote_error_holddown;
372	610	break;
373		case ENETUNREACH:
374	0	h = cache_param->backend_remote_error_holddown;
375	0	break;
376		default:
377	0	break;
378		}
379
380	610	if (h == 0)
381	0	return (r);
382
383	610	Lck_Lock(&cp->mtx);
384	610	h += VTIM_mono();
385	610	if (cp->holddown == 0 \|\| h < cp->holddown) {
386	550	cp->holddown = h;
387	550	cp->holddown_errno = errno;
388	550	}
389
390	610	Lck_Unlock(&cp->mtx);
391
392	610	return (r);
393	37875	}
394
395		/*--------------------------------------------------------------------
396		* Close a connection.
397		*/
398
399		void
400	9030	VCP_Close(struct pfd **pfdp)
401		{
402		struct pfd *pfd;
403		struct conn_pool *cp;
404
405	9030	TAKE_OBJ_NOTNULL(pfd, pfdp, PFD_MAGIC);
406	9030	cp = pfd->conn_pool;
407	9030	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
408
409	9030	assert(pfd->fd > 0);
410
411	9030	Lck_Lock(&cp->mtx);
412	9030	assert(pfd->state == PFD_STATE_USED \|\| pfd->state == PFD_STATE_STOLEN);
413	9030	cp->n_used--;
414	9030	if (pfd->state == PFD_STATE_STOLEN) {
415	50	(void)shutdown(pfd->fd, SHUT_RDWR);
416	50	VTAILQ_REMOVE(&cp->connlist, pfd, list);
417	50	pfd->state = PFD_STATE_CLEANUP;
418	50	cp->n_kill++;
419	50	} else {
420	8980	assert(pfd->state == PFD_STATE_USED);
421	8980	cp->methods->close(pfd);
422	8980	memset(pfd, 0x44, sizeof *pfd);
423	8980	free(pfd);
424		}
425	9030	Lck_Unlock(&cp->mtx);
426	9030	}
427
428		/*--------------------------------------------------------------------
429		* Get a connection, possibly recycled
430		*/
431
432		struct pfd *
433	51926	VCP_Get(struct conn_pool cp, vtim_dur tmo, struct worker wrk,
434		unsigned force_fresh, int *err)
435		{
436		struct pfd *pfd;
437
438	51926	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
439	51926	CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
440	51926	AN(err);
441
442	51926	*err = 0;
443	51926	Lck_Lock(&cp->mtx);
444	51926	pfd = VTAILQ_FIRST(&cp->connlist);
445	51926	CHECK_OBJ_ORNULL(pfd, PFD_MAGIC);
446	51926	if (force_fresh \|\| pfd == NULL \|\| pfd->state == PFD_STATE_STOLEN) {
447	33104	pfd = NULL;
448	33104	} else {
449	18822	assert(pfd->conn_pool == cp);
450	18822	assert(pfd->state == PFD_STATE_AVAIL);
451	18822	VTAILQ_REMOVE(&cp->connlist, pfd, list);
452	18822	VTAILQ_INSERT_TAIL(&cp->connlist, pfd, list);
453	18822	cp->n_conn--;
454	18822	VSC_C_main->backend_reuse++;
455	18822	pfd->state = PFD_STATE_STOLEN;
456	18822	pfd->cond = &wrk->cond;
457		}
458	51926	cp->n_used++; // Opening mostly works
459	51926	Lck_Unlock(&cp->mtx);
460
461	51926	if (pfd != NULL)
462	18822	return (pfd);
463
464	33104	ALLOC_OBJ(pfd, PFD_MAGIC);
465	33104	AN(pfd);
466	33104	INIT_OBJ(pfd->waited, WAITED_MAGIC);
467	33104	pfd->state = PFD_STATE_USED;
468	33104	pfd->conn_pool = cp;
469	33104	pfd->fd = VCP_Open(cp, tmo, &pfd->addr, err);
470	33104	if (pfd->fd < 0) {
471	776	FREE_OBJ(pfd);
472	776	Lck_Lock(&cp->mtx);
473	776	cp->n_used--; // Nope, didn't work after all.
474	776	Lck_Unlock(&cp->mtx);
475	776	} else
476	32328	VSC_C_main->backend_conn++;
477
478	33104	return (pfd);
479	51926	}
480
481		/*--------------------------------------------------------------------
482		*/
483
484		int
485	18722	VCP_Wait(struct worker wrk, struct pfd pfd, vtim_real when)
486		{
487		struct conn_pool *cp;
488		int r;
489
490	18722	CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
491	18722	CHECK_OBJ_NOTNULL(pfd, PFD_MAGIC);
492	18722	cp = pfd->conn_pool;
493	18722	CHECK_OBJ_NOTNULL(cp, CONN_POOL_MAGIC);
494	18722	assert(pfd->cond == &wrk->cond);
495	18722	Lck_Lock(&cp->mtx);
496	37419	while (pfd->state == PFD_STATE_STOLEN) {
497	18722	r = Lck_CondWaitUntil(&wrk->cond, &cp->mtx, when);
498	18722	if (r != 0) {
499	25	if (r == EINTR)