GCOV - Varnish Cache


/*-
 * Copyright (c) 2006 Verdens Gang AS
 * Copyright (c) 2006-2015 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.
 *
 */

#include "config.h"

#include <stdlib.h>
#include <netinet/in.h>
#include <netinet/tcp.h>

#include "cache/cache_int.h"

#include "acceptor/cache_acceptor.h"
#include "acceptor/acceptor_priv.h"
#include "acceptor/acceptor_uds.h"

#include "cache/cache_transport.h"
#include "cache/cache_pool.h"
#include "common/heritage.h"

#include "vcli_serve.h"
#include "vsa.h"
#include "vtcp.h"
#include "vtim.h"

/*--------------------------------------------------------------------
 * We want to get out of any kind of trouble-hit TCP connections as fast
 * as absolutely possible, so we set them LINGER disabled, so that even if
 * there are outstanding write data on the socket, a close(2) will return
 * immediately.
 */
static const struct linger disable_so_linger = {
        .l_onoff        =       0,
};

/*
 * We turn on keepalives by default to assist in detecting clients that have
 * hung up on connections returning from waitinglists
 */
static const unsigned enable_so_keepalive = 1;

/*--------------------------------------------------------------------
 * UDS options we want to control
 */

static struct sock_opt sock_opts[] = {
        /* Note: Setting the mod counter to something not-zero is needed
         * to force the setsockopt() calls on startup */

        SOCK_OPT(SOL_SOCKET, SO_LINGER, struct linger)
        SOCK_OPT(SOL_SOCKET, SO_KEEPALIVE, int)
        SOCK_OPT(SOL_SOCKET, SO_SNDTIMEO, struct timeval)
        SOCK_OPT(SOL_SOCKET, SO_RCVTIMEO, struct timeval)

#undef SOCK_OPT
};

static const int n_sock_opts = vcountof(sock_opts);

/*--------------------------------------------------------------------
 * Some kernels have bugs/limitations with respect to which options are
 * inherited from the accept/listen socket, so we have to keep track of
 * which, if any, sockopts we have to set on the accepted socket.
 */

static int
vca_uds_sockopt_init(void)
{
        struct sock_opt *so;
        union sock_arg tmp;
        int n, chg = 0;
        size_t sz;

        memset(&tmp, 0, sizeof tmp);

        for (n = 0; n < n_sock_opts; n++) {
                so = &sock_opts[n];

#define SET_VAL(nm, so, fld, val)                                       \
        do {                                                            \
                if (!vstrcmp(#nm, so->strname)) {                       \
                        assert(so->sz == sizeof so->arg->fld);          \
                        so->arg->fld = (val);                           \
                }                                                       \
        } while (0)

#define NEW_VAL(nm, so, fld, val)                                       \
        do {                                                            \
                if (!vstrcmp(#nm, so->strname)) {                       \
                        sz = sizeof tmp.fld;                            \
                        assert(so->sz == sz);                           \
                        tmp.fld = (val);                                \
                        if (memcmp(&so->arg->fld, &(tmp.fld), sz)) {    \
                                memcpy(&so->arg->fld, &(tmp.fld), sz);  \
                                so->mod++;                              \
                                chg = 1;                                \
                        }                                               \
                }                                                       \
        } while (0)

                SET_VAL(SO_LINGER, so, lg, disable_so_linger);
                SET_VAL(SO_KEEPALIVE, so, i, enable_so_keepalive);
                NEW_VAL(SO_SNDTIMEO, so, tv,
                    VTIM_timeval_sock(cache_param->idle_send_timeout));
                NEW_VAL(SO_RCVTIMEO, so, tv,
                    VTIM_timeval_sock(cache_param->timeout_idle));
        }

        return (chg);
}

static void
vca_uds_sockopt_test(const struct listen_sock *ls, const struct sess *sp)
{
        struct conn_heritage *ch;
        struct sock_opt *so;
        union sock_arg tmp;
        socklen_t l;
        int i, n;

        CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);
        CHECK_OBJ_NOTNULL(sp, SESS_MAGIC);

        for (n = 0; n < n_sock_opts; n++) {
                so = &sock_opts[n];
                ch = &ls->conn_heritage[n];

                if (ch->sess_set) {
                        VSL(SLT_Debug, sp->vxid,
                            "sockopt: Not testing nonhereditary %s for %s=%s",
                            so->strname, ls->name, ls->endpoint);
                        continue;
                }

                memset(&tmp, 0, sizeof tmp);
                l = so->sz;
                i = getsockopt(sp->fd, so->level, so->optname, &tmp, &l);

                if (i == 0 && memcmp(&tmp, so->arg, so->sz)) {
                        VSL(SLT_Debug, sp->vxid,
                            "sockopt: Test confirmed %s non heredity for %s=%s",
                            so->strname, ls->name, ls->endpoint);
                        ch->sess_set = 1;
                }

                if (i && errno != ENOPROTOOPT)
                        VTCP_Assert(i);
        }
}

static void
vca_uds_sockopt_set(const struct listen_sock *ls, const struct sess *sp)
{
        struct conn_heritage *ch;
        struct sock_opt *so;
        vxid_t vxid;
        int n, sock;

        CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);

        if (sp != NULL) {
                CHECK_OBJ(sp, SESS_MAGIC);
                sock = sp->fd;
                vxid = sp->vxid;
        } else {
                sock = ls->sock;
                vxid = NO_VXID;
        }

        for (n = 0; n < n_sock_opts; n++) {
                so = &sock_opts[n];
                ch = &ls->conn_heritage[n];

                if (sp == NULL && ch->listen_mod == so->mod) {
                        VSL(SLT_Debug, vxid,
                            "sockopt: Not setting unmodified %s for %s=%s",
                            so->strname, ls->name, ls->endpoint);
                        continue;
                }

                if  (sp != NULL && !ch->sess_set) {
                        VSL(SLT_Debug, sp->vxid,
                            "sockopt: %s may be inherited for %s=%s",
                            so->strname, ls->name, ls->endpoint);
                        continue;
                }

                VSL(SLT_Debug, vxid,
                    "sockopt: Setting %s for %s=%s",
                    so->strname, ls->name, ls->endpoint);
                VTCP_Assert(setsockopt(sock,
                    so->level, so->optname, so->arg, so->sz));

                if (sp == NULL)
                        ch->listen_mod = so->mod;
        }
}

static void
vca_uds_init(void)
{

}

static int
vca_uds_listen(struct cli *cli, struct listen_sock *ls)
{

        CHECK_OBJ_NOTNULL(ls->transport, TRANSPORT_MAGIC);
        assert (ls->sock > 0);  // We know where stdin is

        if (listen(ls->sock, cache_param->listen_depth)) {
                VCLI_SetResult(cli, CLIS_CANT);
                VCLI_Out(cli, "Listen failed on socket '%s': %s",
                    ls->endpoint, VAS_errtxt(errno));
                return (-1);
        }

        AZ(ls->conn_heritage);
        ls->conn_heritage = calloc(n_sock_opts,
            sizeof *ls->conn_heritage);
        AN(ls->conn_heritage);

        ls->test_heritage = 1;
        vca_uds_sockopt_set(ls, NULL);

        if (cache_param->accept_filter && VTCP_filter_http(ls->sock))
                VSL(SLT_Error, NO_VXID,
                    "Kernel filtering: sock=%d, errno=%d %s",
                    ls->sock, errno, VAS_errtxt(errno));

        return (0);
}

static void
vca_uds_start(struct cli *cli)
{
        struct listen_sock *ls;

        ASSERT_CLI();

        (void)vca_uds_sockopt_init();

        VTAILQ_FOREACH(ls, &UDS_acceptor.socks, vcalist) {
                CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);

                if (vca_uds_listen(cli, ls))
                        return;
        }
}

static void
vca_uds_event(struct cli *cli, struct listen_sock *ls, enum vca_event event)
{

        (void) ls; // XXX const?
        switch (event) {
        case VCA_EVENT_LADDR:
                CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);
                VCLI_Out(cli, "%s %s -\n", ls->name, ls->endpoint);
                break;
        default:
                WRONG("INVALID VCA_EVENT");
        }
}

static void
vca_mk_uds(struct wrk_accept *wa, struct sess *sp)
{
        struct suckaddr *sa = NULL;
        ssize_t sz;

        (void) wa;
        AN(SES_Reserve_remote_addr(sp, &sa, &sz));
        AN(sa);
        assert(sz == vsa_suckaddr_len);
        AZ(SES_Set_remote_addr(sp, bogo_ip));
        sp->sattr[SA_CLIENT_ADDR] = sp->sattr[SA_REMOTE_ADDR];
        sp->sattr[SA_LOCAL_ADDR] = sp->sattr[SA_REMOTE_ADDR];
        sp->sattr[SA_SERVER_ADDR] = sp->sattr[SA_REMOTE_ADDR];
        AN(SES_Set_String_Attr(sp, SA_CLIENT_IP, "0.0.0.0"));
        AN(SES_Set_String_Attr(sp, SA_CLIENT_PORT, "0"));
}

static void v_matchproto_(task_func_t)
vca_uds_make_session(struct worker *wrk, void *arg)
{
        struct wrk_accept *wa;
        struct sess *sp;
        struct req *req;

        CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
        CAST_OBJ_NOTNULL(wa, arg, WRK_ACCEPT_MAGIC);

        VTCP_blocking(wa->acceptsock);

        /* Turn accepted socket into a session */
        AN(WS_Reservation(wrk->aws));
        sp = SES_New(wrk->pool);
        CHECK_OBJ_NOTNULL(sp, SESS_MAGIC);
        wrk->stats->s_sess++;

        sp->t_open = VTIM_real();
        sp->t_idle = sp->t_open;
        sp->vxid = VXID_Get(wrk, VSL_CLIENTMARKER);

        sp->fd = wa->acceptsock;
        wa->acceptsock = -1;
        sp->listen_sock = wa->acceptlsock;

        assert((size_t)wa->acceptaddrlen <= vsa_suckaddr_len);

        vca_mk_uds(wa, sp);

        AN(wa->acceptlsock->name);
        VSL(SLT_Begin, sp->vxid, "sess 0 %s",
            wa->acceptlsock->transport->name);
        VSL(SLT_SessOpen, sp->vxid, "0.0.0.0 0 %s 0.0.0.0 0 %.6f %d",
            wa->acceptlsock->name, sp->t_open, sp->fd);

        vca_pace_good();
        wrk->stats->sess_conn++;

        if (wa->acceptlsock->test_heritage) {
                vca_uds_sockopt_test(wa->acceptlsock, sp);
                wa->acceptlsock->test_heritage = 0;
        }

        vca_uds_sockopt_set(wa->acceptlsock, sp);

        req = Req_New(sp, NULL);
        CHECK_OBJ_NOTNULL(req, REQ_MAGIC);
        req->htc->rfd = &sp->fd;

        SES_SetTransport(wrk, sp, req, wa->acceptlsock->transport);
        WS_Release(wrk->aws, 0);
}

/*--------------------------------------------------------------------
 * This function accepts on a single socket for a single thread pool.
 *
 * As long as we can stick the accepted connection to another thread
 * we do so, otherwise we put the socket back on the "BACK" pool
 * and handle the new connection ourselves.
 */

static void v_matchproto_(task_func_t)
vca_uds_accept_task(struct worker *wrk, void *arg)
{
        struct listen_sock *ls;
        struct wrk_accept wa;
        struct poolsock *ps;
        int i;

        CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
        CAST_OBJ_NOTNULL(ps, arg, POOLSOCK_MAGIC);
        ls = ps->lsock;
        CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);

        /* Return any cached resources from previous task */
        WRK_Cleanup(wrk);

        while (!pool_accepting)
                VTIM_sleep(.1);

        while (!ps->pool->die) {
                INIT_OBJ(&wa, WRK_ACCEPT_MAGIC);
                wa.acceptlsock = ls;

                vca_pace_check();

                wa.acceptaddrlen = sizeof wa.acceptaddr;
                do {
                        i = accept(ls->sock, (void*)&wa.acceptaddr,
                            &wa.acceptaddrlen);
                } while (i < 0 && errno == EAGAIN && !ps->pool->die);

                if (i < 0 && ps->pool->die)
                        break;

                if (i < 0 && ls->sock == -2) {
                        /* Shut down in progress */
                        sleep(2);
                        continue;
                }

                if (i < 0) {
                        switch (errno) {
                        case ECONNABORTED:
                                wrk->stats->sess_fail_econnaborted++;
                                break;
                        case EINTR:
                                wrk->stats->sess_fail_eintr++;
                                break;
                        case EMFILE:
                                wrk->stats->sess_fail_emfile++;
                                vca_pace_bad();
                                break;
                        case EBADF:
                                wrk->stats->sess_fail_ebadf++;
                                vca_pace_bad();
                                break;
                        case ENOBUFS:
                        case ENOMEM:
                                wrk->stats->sess_fail_enomem++;
                                vca_pace_bad();
                                break;
                        default:
                                wrk->stats->sess_fail_other++;
                                vca_pace_bad();
                                break;
                        }

                        i = errno;
                        wrk->stats->sess_fail++;

                        VSL(SLT_SessError, NO_VXID, "%s 0.0.0.0 0 %d %d \"%s\"",
                            wa.acceptlsock->name, ls->sock, i, VAS_errtxt(i));
                        (void)Pool_TrySumstat(wrk);
                        continue;
                }

                wa.acceptsock = i;

                if (!Pool_Task_Arg(wrk, TASK_QUEUE_REQ,
                    vca_uds_make_session, &wa, sizeof wa)) {
                        /*
                         * We couldn't get another thread, so we will handle
                         * the request in this worker thread, but first we
                         * must reschedule the listening task so it will be
                         * taken up by another thread again.
                         */
                        if (!ps->pool->die) {
                                AZ(Pool_Task(wrk->pool, ps->task,
                                    TASK_QUEUE_VCA));
                                return;
                        }
                }
                if (!ps->pool->die && DO_DEBUG(DBG_SLOW_ACCEPTOR))
                        VTIM_sleep(2.0);

        }

        VSL(SLT_Debug, NO_VXID, "XXX Accept thread dies %p", ps);
        FREE_OBJ(ps);
}

static void
vca_uds_accept(struct pool *pp)
{
        struct listen_sock *ls;
        struct poolsock *ps;

        VTAILQ_FOREACH(ls, &UDS_acceptor.socks, vcalist) {
                CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);

                ALLOC_OBJ(ps, POOLSOCK_MAGIC);
                AN(ps);
                ps->lsock = ls;
                ps->task->func = vca_uds_accept_task;
                ps->task->priv = ps;
                ps->pool = pp;
                VTAILQ_INSERT_TAIL(&pp->poolsocks, ps, list);
                AZ(Pool_Task(pp, ps->task, TASK_QUEUE_VCA));
        }
}

static void
vca_uds_update(struct lock *shut_mtx)
{
        struct listen_sock *ls;

        if (!vca_uds_sockopt_init())
                return;

        Lck_Lock(shut_mtx);

        VTAILQ_FOREACH(ls, &UDS_acceptor.socks, vcalist) {
                CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);

                if (ls->sock == -2)
                        continue;       // VCA_Shutdown
                assert (ls->sock > 0);
                vca_uds_sockopt_set(ls, NULL);
                /* If one of the options on a socket has
                 * changed, also force a retest of whether
                 * the values are inherited to the
                 * accepted sockets. This should then
                 * catch any false positives from previous
                 * tests that could happen if the set
                 * value of an option happened to just be
                 * the OS default for that value, and
                 * wasn't actually inherited from the
                 * listening socket. */
                ls->test_heritage = 1;
        }

        Lck_Unlock(shut_mtx);
}

static void
vca_uds_shutdown(void)
{
        struct listen_sock *ls;
        int i;

        VTAILQ_FOREACH(ls, &UDS_acceptor.socks, vcalist) {
                CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);

                i = ls->sock;
                ls->sock = -2;
                (void)close(i);
        }
}

struct acceptor UDS_acceptor = {
        .magic          = ACCEPTOR_MAGIC,
        .name           = "uds",
        .config         = vca_uds_config,
        .init           = vca_uds_init,
        .open           = vca_uds_open,
        .reopen         = vca_uds_reopen,
        .start          = vca_uds_start,
        .event          = vca_uds_event,
        .accept         = vca_uds_accept,
        .update         = vca_uds_update,
        .shutdown       = vca_uds_shutdown,
};

		vinyl-cache/bin/vinyld/acceptor/cache_acceptor_uds.c
0		/*-
1		* Copyright (c) 2006 Verdens Gang AS
2		* Copyright (c) 2006-2015 Varnish Software AS
3		* All rights reserved.
4		*
5		* Author: Poul-Henning Kamp <phk@phk.freebsd.dk>
6		*
7		* SPDX-License-Identifier: BSD-2-Clause
8		*
9		* Redistribution and use in source and binary forms, with or without
10		* modification, are permitted provided that the following conditions
11		* are met:
12		* 1. Redistributions of source code must retain the above copyright
13		* notice, this list of conditions and the following disclaimer.
14		* 2. Redistributions in binary form must reproduce the above copyright
15		* notice, this list of conditions and the following disclaimer in the
16		* documentation and/or other materials provided with the distribution.
17		*
18		* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19		* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20		* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21		* ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
22		* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23		* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24		* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25		* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26		* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27		* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28		* SUCH DAMAGE.
29		*
30		*/
31
32		#include "config.h"
33
34		#include <stdlib.h>
35		#include <netinet/in.h>
36		#include <netinet/tcp.h>
37
38		#include "cache/cache_int.h"
39
40		#include "acceptor/cache_acceptor.h"
41		#include "acceptor/acceptor_priv.h"
42		#include "acceptor/acceptor_uds.h"
43
44		#include "cache/cache_transport.h"
45		#include "cache/cache_pool.h"
46		#include "common/heritage.h"
47
48		#include "vcli_serve.h"
49		#include "vsa.h"
50		#include "vtcp.h"
51		#include "vtim.h"
52
53		/*--------------------------------------------------------------------
54		* We want to get out of any kind of trouble-hit TCP connections as fast
55		* as absolutely possible, so we set them LINGER disabled, so that even if
56		* there are outstanding write data on the socket, a close(2) will return
57		* immediately.
58		*/
59		static const struct linger disable_so_linger = {
60		.l_onoff = 0,
61		};
62
63		/*
64		* We turn on keepalives by default to assist in detecting clients that have
65		* hung up on connections returning from waitinglists
66		*/
67		static const unsigned enable_so_keepalive = 1;
68
69		/*--------------------------------------------------------------------
70		* UDS options we want to control
71		*/
72
73		static struct sock_opt sock_opts[] = {
74		/* Note: Setting the mod counter to something not-zero is needed
75		* to force the setsockopt() calls on startup */
76
77		SOCK_OPT(SOL_SOCKET, SO_LINGER, struct linger)
78		SOCK_OPT(SOL_SOCKET, SO_KEEPALIVE, int)
79		SOCK_OPT(SOL_SOCKET, SO_SNDTIMEO, struct timeval)
80		SOCK_OPT(SOL_SOCKET, SO_RCVTIMEO, struct timeval)
81
82		#undef SOCK_OPT
83		};
84
85		static const int n_sock_opts = vcountof(sock_opts);
86
87		/*--------------------------------------------------------------------
88		* Some kernels have bugs/limitations with respect to which options are
89		* inherited from the accept/listen socket, so we have to keep track of
90		* which, if any, sockopts we have to set on the accepted socket.
91		*/
92
93		static int
94	10555	vca_uds_sockopt_init(void)
95		{
96		struct sock_opt *so;
97		union sock_arg tmp;
98	10555	int n, chg = 0;
99		size_t sz;
100
101	10555	memset(&tmp, 0, sizeof tmp);
102
103	52775	for (n = 0; n < n_sock_opts; n++) {
104	42220	so = &sock_opts[n];
105
106		#define SET_VAL(nm, so, fld, val) \
107		do { \
108		if (!vstrcmp(#nm, so->strname)) { \
109		assert(so->sz == sizeof so->arg->fld); \
110		so->arg->fld = (val); \
111		} \
112		} while (0)
113
114		#define NEW_VAL(nm, so, fld, val) \
115		do { \
116		if (!vstrcmp(#nm, so->strname)) { \
117		sz = sizeof tmp.fld; \
118		assert(so->sz == sz); \
119		tmp.fld = (val); \
120		if (memcmp(&so->arg->fld, &(tmp.fld), sz)) { \
121		memcpy(&so->arg->fld, &(tmp.fld), sz); \
122		so->mod++; \
123		chg = 1; \
124		} \
125		} \
126		} while (0)
127
128	42220	SET_VAL(SO_LINGER, so, lg, disable_so_linger);
129	42220	SET_VAL(SO_KEEPALIVE, so, i, enable_so_keepalive);
130	42220	NEW_VAL(SO_SNDTIMEO, so, tv,
131		VTIM_timeval_sock(cache_param->idle_send_timeout));
132	42220	NEW_VAL(SO_RCVTIMEO, so, tv,
133		VTIM_timeval_sock(cache_param->timeout_idle));
134	42220	}
135
136	10555	return (chg);
137		}
138
139		static void
140	98	vca_uds_sockopt_test(const struct listen_sock ls, const struct sess sp)
141		{
142		struct conn_heritage *ch;
143		struct sock_opt *so;
144		union sock_arg tmp;
145		socklen_t l;
146		int i, n;
147
148	98	CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);
149	98	CHECK_OBJ_NOTNULL(sp, SESS_MAGIC);
150
151	490	for (n = 0; n < n_sock_opts; n++) {
152	392	so = &sock_opts[n];
153	392	ch = &ls->conn_heritage[n];
154
155	392	if (ch->sess_set) {
156	4	VSL(SLT_Debug, sp->vxid,
157		"sockopt: Not testing nonhereditary %s for %s=%s",
158	2	so->strname, ls->name, ls->endpoint);
159	2	continue;
160		}
161
162	390	memset(&tmp, 0, sizeof tmp);
163	390	l = so->sz;
164	390	i = getsockopt(sp->fd, so->level, so->optname, &tmp, &l);
165
166	390	if (i == 0 && memcmp(&tmp, so->arg, so->sz)) {
167	192	VSL(SLT_Debug, sp->vxid,
168		"sockopt: Test confirmed %s non heredity for %s=%s",
169	96	so->strname, ls->name, ls->endpoint);
170	96	ch->sess_set = 1;
171	96	}
172
173	390	if (i && errno != ENOPROTOOPT)
174	0	VTCP_Assert(i);
175	390	}
176	98	}
177
178		static void
179	1468	vca_uds_sockopt_set(const struct listen_sock ls, const struct sess sp)
180		{
181		struct conn_heritage *ch;
182		struct sock_opt *so;
183		vxid_t vxid;
184		int n, sock;
185
186	1468	CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);
187
188	1468	if (sp != NULL) {
189	1360	CHECK_OBJ(sp, SESS_MAGIC);
190	1360	sock = sp->fd;
191	1360	vxid = sp->vxid;
192	1360	} else {
193	108	sock = ls->sock;
194	108	vxid = NO_VXID;
195		}
196
197	7340	for (n = 0; n < n_sock_opts; n++) {
198	5872	so = &sock_opts[n];
199	5872	ch = &ls->conn_heritage[n];
200
201	5872	if (sp == NULL && ch->listen_mod == so->mod) {
202	12	VSL(SLT_Debug, vxid,
203		"sockopt: Not setting unmodified %s for %s=%s",
204	12	so->strname, ls->name, ls->endpoint);
205	12	continue;
206		}
207
208	5860	if (sp != NULL && !ch->sess_set) {
209	8160	VSL(SLT_Debug, sp->vxid,
210		"sockopt: %s may be inherited for %s=%s",
211	4080	so->strname, ls->name, ls->endpoint);
212	4080	continue;
213		}
214
215	1780	VSL(SLT_Debug, vxid,
216		"sockopt: Setting %s for %s=%s",
217	1780	so->strname, ls->name, ls->endpoint);
218	1780	VTCP_Assert(setsockopt(sock,
219		so->level, so->optname, so->arg, so->sz));
220
221	1780	if (sp == NULL)
222	420	ch->listen_mod = so->mod;
223	1780	}
224	1468	}
225
226		static void
227	3892	vca_uds_init(void)
228		{
229
230	3892	}
231
232		static int
233	104	vca_uds_listen(struct cli cli, struct listen_sock ls)
234		{
235
236	104	CHECK_OBJ_NOTNULL(ls->transport, TRANSPORT_MAGIC);
237	104	assert (ls->sock > 0); // We know where stdin is
238
239	104	if (listen(ls->sock, cache_param->listen_depth)) {
240	0	VCLI_SetResult(cli, CLIS_CANT);
241	0	VCLI_Out(cli, "Listen failed on socket '%s': %s",
242	0	ls->endpoint, VAS_errtxt(errno));
243	0	return (-1);
244		}
245
246	104	AZ(ls->conn_heritage);
247	104	ls->conn_heritage = calloc(n_sock_opts,
248		sizeof *ls->conn_heritage);
249	104	AN(ls->conn_heritage);
250
251	104	ls->test_heritage = 1;
252	104	vca_uds_sockopt_set(ls, NULL);
253
254	104	if (cache_param->accept_filter && VTCP_filter_http(ls->sock))
255	200	VSL(SLT_Error, NO_VXID,
256		"Kernel filtering: sock=%d, errno=%d %s",
257	100	ls->sock, errno, VAS_errtxt(errno));
258
259	104	return (0);
260	104	}
261
262		static void
263	3880	vca_uds_start(struct cli *cli)
264		{
265		struct listen_sock *ls;
266
267	3880	ASSERT_CLI();
268
269	3880	(void)vca_uds_sockopt_init();
270
271	3984	VTAILQ_FOREACH(ls, &UDS_acceptor.socks, vcalist) {
272	104	CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);
273
274	104	if (vca_uds_listen(cli, ls))
275	0	return;
276	104	}
277	3880	}
278
279		static void
280	208	vca_uds_event(struct cli cli, struct listen_sock ls, enum vca_event event)
281		{
282
283	208	(void) ls; // XXX const?
284	208	switch (event) {
285		case VCA_EVENT_LADDR:
286	208	CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);
287	208	VCLI_Out(cli, "%s %s -\n", ls->name, ls->endpoint);
288	208	break;
289		default:
290	0	WRONG("INVALID VCA_EVENT");
291	0	}
292	208	}
293
294		static void
295	1360	vca_mk_uds(struct wrk_accept wa, struct sess sp)
296		{
297	1360	struct suckaddr *sa = NULL;
298		ssize_t sz;
299
300	1360	(void) wa;
301	1360	AN(SES_Reserve_remote_addr(sp, &sa, &sz));
302	1360	AN(sa);
303	1360	assert(sz == vsa_suckaddr_len);
304	1360	AZ(SES_Set_remote_addr(sp, bogo_ip));
305	1360	sp->sattr[SA_CLIENT_ADDR] = sp->sattr[SA_REMOTE_ADDR];
306	1360	sp->sattr[SA_LOCAL_ADDR] = sp->sattr[SA_REMOTE_ADDR];
307	1360	sp->sattr[SA_SERVER_ADDR] = sp->sattr[SA_REMOTE_ADDR];
308	1360	AN(SES_Set_String_Attr(sp, SA_CLIENT_IP, "0.0.0.0"));
309	1360	AN(SES_Set_String_Attr(sp, SA_CLIENT_PORT, "0"));
310	1360	}
311
312		static void v_matchproto_(task_func_t)
313	1360	vca_uds_make_session(struct worker wrk, void arg)
314		{
315		struct wrk_accept *wa;
316		struct sess *sp;
317		struct req *req;
318
319	1360	CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
320	1360	CAST_OBJ_NOTNULL(wa, arg, WRK_ACCEPT_MAGIC);
321
322	1360	VTCP_blocking(wa->acceptsock);
323
324		/* Turn accepted socket into a session */
325	1360	AN(WS_Reservation(wrk->aws));
326	1360	sp = SES_New(wrk->pool);
327	1360	CHECK_OBJ_NOTNULL(sp, SESS_MAGIC);
328	1360	wrk->stats->s_sess++;
329
330	1360	sp->t_open = VTIM_real();
331	1360	sp->t_idle = sp->t_open;
332	1360	sp->vxid = VXID_Get(wrk, VSL_CLIENTMARKER);
333
334	1360	sp->fd = wa->acceptsock;
335	1360	wa->acceptsock = -1;
336	1360	sp->listen_sock = wa->acceptlsock;
337
338	1360	assert((size_t)wa->acceptaddrlen <= vsa_suckaddr_len);
339
340	1360	vca_mk_uds(wa, sp);
341
342	1360	AN(wa->acceptlsock->name);
343	2720	VSL(SLT_Begin, sp->vxid, "sess 0 %s",
344	1360	wa->acceptlsock->transport->name);
345	2720	VSL(SLT_SessOpen, sp->vxid, "0.0.0.0 0 %s 0.0.0.0 0 %.6f %d",
346	1360	wa->acceptlsock->name, sp->t_open, sp->fd);
347
348	1360	vca_pace_good();
349	1360	wrk->stats->sess_conn++;
350
351	1360	if (wa->acceptlsock->test_heritage) {
352	98	vca_uds_sockopt_test(wa->acceptlsock, sp);
353	98	wa->acceptlsock->test_heritage = 0;
354	98	}
355
356	1360	vca_uds_sockopt_set(wa->acceptlsock, sp);
357
358	1360	req = Req_New(sp, NULL);
359	1360	CHECK_OBJ_NOTNULL(req, REQ_MAGIC);
360	1360	req->htc->rfd = &sp->fd;
361
362	1360	SES_SetTransport(wrk, sp, req, wa->acceptlsock->transport);
363	1360	WS_Release(wrk->aws, 0);
364	1360	}
365
366		/*--------------------------------------------------------------------
367		* This function accepts on a single socket for a single thread pool.
368		*
369		* As long as we can stick the accepted connection to another thread
370		* we do so, otherwise we put the socket back on the "BACK" pool
371		* and handle the new connection ourselves.
372		*/
373
374		static void v_matchproto_(task_func_t)
375	2	vca_uds_accept_task(struct worker wrk, void arg)
376		{
377		struct listen_sock *ls;
378		struct wrk_accept wa;
379		struct poolsock *ps;
380		int i;
381
382	2	CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
383	2	CAST_OBJ_NOTNULL(ps, arg, POOLSOCK_MAGIC);
384	2	ls = ps->lsock;
385	2	CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);
386
387		/* Return any cached resources from previous task */
388	2	WRK_Cleanup(wrk);
389
390	175	while (!pool_accepting)
391	173	VTIM_sleep(.1);
392
393	1364	while (!ps->pool->die) {
394	1362	INIT_OBJ(&wa, WRK_ACCEPT_MAGIC);
395	1362	wa.acceptlsock = ls;
396
397	1362	vca_pace_check();
398
399	1362	wa.acceptaddrlen = sizeof wa.acceptaddr;
400	1362	do {
401	3132	i = accept(ls->sock, (void*)&wa.acceptaddr,
402	1566	&wa.acceptaddrlen);
403	1566	} while (i < 0 && errno == EAGAIN && !ps->pool->die);
404
405	1362	if (i < 0 && ps->pool->die)
406	0	break;
407
408	1362	if (i < 0 && ls->sock == -2) {
409		/* Shut down in progress */
410	2	sleep(2);
411	2	continue;
412		}
413
414	1360	if (i < 0) {
415	0	switch (errno) {
416		case ECONNABORTED:
417	0	wrk->stats->sess_fail_econnaborted++;
418	0	break;
419		case EINTR:
420	0	wrk->stats->sess_fail_eintr++;
421	0	break;
422		case EMFILE:
423	0	wrk->stats->sess_fail_emfile++;
424	0	vca_pace_bad();
425	0	break;
426		case EBADF:
427	0	wrk->stats->sess_fail_ebadf++;
428	0	vca_pace_bad();
429	0	break;
430		case ENOBUFS:
431		case ENOMEM:
432	0	wrk->stats->sess_fail_enomem++;
433	0	vca_pace_bad();
434	0	break;
435		default:
436	0	wrk->stats->sess_fail_other++;
437	0	vca_pace_bad();
438	0	break;
439		}
440
441	0	i = errno;
442	0	wrk->stats->sess_fail++;
443
444	0	VSL(SLT_SessError, NO_VXID, "%s 0.0.0.0 0 %d %d \"%s\"",
445	0	wa.acceptlsock->name, ls->sock, i, VAS_errtxt(i));
446	0	(void)Pool_TrySumstat(wrk);
447	0	continue;
448		}
449
450	1360	wa.acceptsock = i;
451
452	1360	if (!Pool_Task_Arg(wrk, TASK_QUEUE_REQ,
453		vca_uds_make_session, &wa, sizeof wa)) {
454		/*
455		* We couldn't get another thread, so we will handle
456		* the request in this worker thread, but first we
457		* must reschedule the listening task so it will be
458		* taken up by another thread again.
459		*/
460	1	if (!ps->pool->die) {
461	0	AZ(Pool_Task(wrk->pool, ps->task,
462		TASK_QUEUE_VCA));
463	0	return;
464		}
465	1	}
466	1360	if (!ps->pool->die && DO_DEBUG(DBG_SLOW_ACCEPTOR))
467	0	VTIM_sleep(2.0);
468
469		}
470
471	2	VSL(SLT_Debug, NO_VXID, "XXX Accept thread dies %p", ps);
472	2	FREE_OBJ(ps);
473	2	}
474
475		static void
476	7696	vca_uds_accept(struct pool *pp)
477		{
478		struct listen_sock *ls;
479		struct poolsock *ps;
480
481	7904	VTAILQ_FOREACH(ls, &UDS_acceptor.socks, vcalist) {
482	208	CHECK_OBJ_NOTNULL(ls, LISTEN_SOCK_MAGIC);
483
484	208	ALLOC_OBJ(ps, POOLSOCK_MAGIC);
485	208	AN(ps);
486	208	ps->lsock = ls;
487	208	ps->task->func = vca_uds_accept_task;
488	208	ps->task->priv = ps;
489	208	ps->pool = pp;
490	208	VTAILQ_INSERT_TAIL(&pp->poolsocks, ps, list);
491	208	AZ(Pool_Task(pp, ps->task, TASK_QUEUE_VCA));
492	208	}
493	7696	}
494
495		static void
496	6675	vca_uds_update(struct lock *shut_mtx)
497		{
498		struct listen_sock *ls;
499