Home | History | Annotate | Line # | Download | only in kern
uipc_accf.c revision 1.5
      1 /*	$NetBSD: uipc_accf.c,v 1.5 2008/10/14 13:45:26 ad Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software developed for The NetBSD Foundation
      8  * by Andrew Doran.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*-
     33  * Copyright (c) 2000 Paycounter, Inc.
     34  * Copyright (c) 2005 Robert N. M. Watson
     35  * Author: Alfred Perlstein <alfred (at) paycounter.com>, <alfred (at) FreeBSD.org>
     36  * All rights reserved.
     37  *
     38  * Redistribution and use in source and binary forms, with or without
     39  * modification, are permitted provided that the following conditions
     40  * are met:
     41  * 1. Redistributions of source code must retain the above copyright
     42  *    notice, this list of conditions and the following disclaimer.
     43  * 2. Redistributions in binary form must reproduce the above copyright
     44  *    notice, this list of conditions and the following disclaimer in the
     45  *    documentation and/or other materials provided with the distribution.
     46  *
     47  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     57  * SUCH DAMAGE.
     58  */
     59 
     60 #include <sys/cdefs.h>
     61 __KERNEL_RCSID(0, "$NetBSD: uipc_accf.c,v 1.5 2008/10/14 13:45:26 ad Exp $");
     62 
     63 #define ACCEPT_FILTER_MOD
     64 
     65 #include <sys/param.h>
     66 #include <sys/systm.h>
     67 #include <sys/domain.h>
     68 #include <sys/kernel.h>
     69 #include <sys/lock.h>
     70 #include <sys/kmem.h>
     71 #include <sys/mbuf.h>
     72 #include <sys/lkm.h>
     73 #include <sys/rwlock.h>
     74 #include <sys/protosw.h>
     75 #include <sys/sysctl.h>
     76 #include <sys/socket.h>
     77 #include <sys/socketvar.h>
     78 #include <sys/queue.h>
     79 #include <sys/once.h>
     80 #include <sys/atomic.h>
     81 
     82 static krwlock_t accept_filter_lock;
     83 
     84 static LIST_HEAD(, accept_filter) accept_filtlsthd =
     85     LIST_HEAD_INITIALIZER(&accept_filtlsthd);
     86 
     87 /*
     88  * Names of Accept filter sysctl objects
     89  */
     90 SYSCTL_SETUP(sysctl_net_inet_accf_setup, "sysctl net.inet.accf subtree setup")
     91 {
     92 
     93 	sysctl_createv(clog, 0, NULL, NULL,
     94 		       CTLFLAG_PERMANENT,
     95 		       CTLTYPE_NODE, "net", NULL,
     96 		       NULL, 0, NULL, 0,
     97 		       CTL_NET, CTL_EOL);
     98 	sysctl_createv(clog, 0, NULL, NULL,
     99 		       CTLFLAG_PERMANENT,
    100 		       CTLTYPE_NODE, "inet", NULL,
    101 		       NULL, 0, NULL, 0,
    102 		       CTL_NET, PF_INET, CTL_EOL);
    103 	sysctl_createv(clog, 0, NULL, NULL,
    104 		       CTLFLAG_PERMANENT,
    105 		       CTLTYPE_NODE, "accf",
    106 		       SYSCTL_DESCR("Accept filters"),
    107 		       NULL, 0, NULL, 0,
    108 		       CTL_NET, PF_INET, SO_ACCEPTFILTER, CTL_EOL);
    109 }
    110 
    111 int
    112 accept_filt_add(struct accept_filter *filt)
    113 {
    114 	struct accept_filter *p;
    115 
    116 	rw_enter(&accept_filter_lock, RW_WRITER);
    117 	LIST_FOREACH(p, &accept_filtlsthd, accf_next) {
    118 		if (strcmp(p->accf_name, filt->accf_name) == 0)  {
    119 			rw_exit(&accept_filter_lock);
    120 			return EEXIST;
    121 		}
    122 	}
    123 	LIST_INSERT_HEAD(&accept_filtlsthd, filt, accf_next);
    124 	rw_exit(&accept_filter_lock);
    125 
    126 	return 0;
    127 }
    128 
    129 int
    130 accept_filt_del(struct accept_filter *p)
    131 {
    132 
    133 	rw_enter(&accept_filter_lock, RW_WRITER);
    134 	if (p->accf_refcnt != 0) {
    135 		rw_exit(&accept_filter_lock);
    136 		return EBUSY;
    137 	}
    138 	LIST_REMOVE(p, accf_next);
    139 	rw_exit(&accept_filter_lock);
    140 
    141 	return 0;
    142 }
    143 
    144 struct accept_filter *
    145 accept_filt_get(char *name)
    146 {
    147 	struct accept_filter *p;
    148 
    149 	rw_enter(&accept_filter_lock, RW_READER);
    150 	LIST_FOREACH(p, &accept_filtlsthd, accf_next) {
    151 		if (strcmp(p->accf_name, name) == 0) {
    152 			atomic_inc_uint(&p->accf_refcnt);
    153 			break;
    154 		}
    155 	}
    156 	rw_exit(&accept_filter_lock);
    157 
    158 	return p;
    159 }
    160 
    161 /*
    162  * Accept filter initialization routine.
    163  * This should be called only once.
    164  */
    165 
    166 static int
    167 accept_filter_init0(void)
    168 {
    169 
    170 	rw_init(&accept_filter_lock);
    171 
    172 	return 0;
    173 }
    174 
    175 /*
    176  * Initialization routine: This can also be replaced with
    177  * accept_filt_generic_mod_event for attaching new accept filter.
    178  */
    179 
    180 void
    181 accept_filter_init(void)
    182 {
    183 	static ONCE_DECL(accept_filter_init_once);
    184 
    185 	RUN_ONCE(&accept_filter_init_once, accept_filter_init0);
    186 }
    187 
    188 int
    189 accept_filt_generic_mod_event(struct lkm_table *lkmtp, int event, void *data)
    190 {
    191 	struct accept_filter *accfp = (struct accept_filter *) data;
    192 	int error;
    193 
    194 	switch (event) {
    195 	case LKM_E_LOAD:
    196 		accept_filter_init();
    197 		error = accept_filt_add(accfp);
    198 		break;
    199 
    200 	case LKM_E_UNLOAD:
    201 		error = accept_filt_del(accfp);
    202 		break;
    203 
    204 	case LKM_E_STAT:
    205 		error = 0;
    206 		break;
    207 
    208 	default:
    209 		error = EOPNOTSUPP;
    210 		break;
    211 	}
    212 
    213 	return error;
    214 }
    215 
    216 int
    217 accept_filt_getopt(struct socket *so, struct sockopt *sopt)
    218 {
    219 	struct accept_filter_arg afa;
    220 	int error;
    221 
    222 	KASSERT(solocked(so));
    223 
    224 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
    225 		error = EINVAL;
    226 		goto out;
    227 	}
    228 	if ((so->so_options & SO_ACCEPTFILTER) == 0) {
    229 		error = EINVAL;
    230 		goto out;
    231 	}
    232 
    233 	memset(&afa, 0, sizeof(afa));
    234 	strcpy(afa.af_name, so->so_accf->so_accept_filter->accf_name);
    235 	if (so->so_accf->so_accept_filter_str != NULL)
    236 		strcpy(afa.af_arg, so->so_accf->so_accept_filter_str);
    237 	error = sockopt_set(sopt, &afa, sizeof(afa));
    238 out:
    239 	return error;
    240 }
    241 
    242 /*
    243  * Simple delete case, with socket locked.
    244  */
    245 int
    246 accept_filt_clear(struct socket *so)
    247 {
    248 	struct accept_filter_arg afa;
    249 	struct accept_filter *afp;
    250 	struct socket *so2;
    251 	struct so_accf *af;
    252 
    253 	KASSERT(solocked(so));
    254 
    255 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
    256 		return EINVAL;
    257 	}
    258 	if (so->so_accf != NULL) {
    259 		/* Break in-flight processing. */
    260 		TAILQ_FOREACH(so2, &so->so_q0, so_qe) {
    261 			if (so2->so_upcall == NULL) {
    262 				continue;
    263 			}
    264 			so2->so_upcall = NULL;
    265 			so2->so_upcallarg = NULL;
    266 			so2->so_rcv.sb_flags &= ~SB_UPCALL;
    267 			soqremque(so2, 0);
    268 			soqinsque(so, so2, 1);
    269 			sorwakeup(so);
    270 			cv_broadcast(&so->so_cv);
    271 		}
    272 		af = so->so_accf;
    273 		afp = af->so_accept_filter;
    274 		if (afp != NULL && afp->accf_destroy != NULL) {
    275 			(*afp->accf_destroy)(so);
    276 		}
    277 		if (af->so_accept_filter_str != NULL) {
    278 			kmem_free(af->so_accept_filter_str,
    279 			    sizeof(afa.af_name));
    280 		}
    281 		kmem_free(af, sizeof(*af));
    282 		so->so_accf = NULL;
    283 		atomic_dec_uint(&afp->accf_refcnt);
    284 	}
    285 	so->so_options &= ~SO_ACCEPTFILTER;
    286 	return 0;
    287 }
    288 
    289 /*
    290  * setsockopt() for accept filters.  Called with the socket unlocked,
    291  * will always return it locked.
    292  */
    293 int
    294 accept_filt_setopt(struct socket *so, const struct sockopt *sopt)
    295 {
    296 	struct accept_filter_arg afa;
    297 	struct accept_filter *afp;
    298 	struct so_accf *newaf;
    299 	int error;
    300 
    301 	if (sopt == NULL || sopt->sopt_size == 0) {
    302 		solock(so);
    303 		return accept_filt_clear(so);
    304 	}
    305 
    306 	/*
    307 	 * Pre-allocate any memory we may need later to avoid blocking at
    308 	 * untimely moments.  This does not optimize for invalid arguments.
    309 	 */
    310 	error = sockopt_get(sopt, &afa, sizeof(afa));
    311 	if (error) {
    312 		solock(so);
    313 		return error;
    314 	}
    315 	afa.af_name[sizeof(afa.af_name)-1] = '\0';
    316 	afa.af_arg[sizeof(afa.af_arg)-1] = '\0';
    317 	afp = accept_filt_get(afa.af_name);
    318 	if (afp == NULL) {
    319 		solock(so);
    320 		return ENOENT;
    321 	}
    322 	/*
    323 	 * Allocate the new accept filter instance storage.  We may
    324 	 * have to free it again later if we fail to attach it.  If
    325 	 * attached properly, 'newaf' is NULLed to avoid a free()
    326 	 * while in use.
    327 	 */
    328 	newaf = kmem_zalloc(sizeof(*newaf), KM_SLEEP);
    329 	if (afp->accf_create != NULL && afa.af_name[0] != '\0') {
    330 		/*
    331 		 * FreeBSD did a variable-size allocation here
    332 		 * with the actual string length from afa.af_name
    333 		 * but it is so short, why bother tracking it?
    334 		 * XXX as others have noted, this is an API mistake;
    335 		 * XXX accept_filter_arg should have a mandatory namelen.
    336 		 * XXX (but it's a bit too late to fix that now)
    337 		 */
    338 		newaf->so_accept_filter_str =
    339 		    kmem_alloc(sizeof(afa.af_name), KM_SLEEP);
    340 		strcpy(newaf->so_accept_filter_str, afa.af_name);
    341 	}
    342 
    343 	/*
    344 	 * Require a listen socket; don't try to replace an existing filter
    345 	 * without first removing it.
    346 	 */
    347 	solock(so);
    348 	if ((so->so_options & SO_ACCEPTCONN) == 0 || so->so_accf != NULL) {
    349 		error = EINVAL;
    350 		goto out;
    351 	}
    352 
    353 	/*
    354 	 * Invoke the accf_create() method of the filter if required.  The
    355 	 * socket lock is held over this call, so create methods for filters
    356 	 * shouldn't block.
    357 	 */
    358 	if (afp->accf_create != NULL) {
    359 		newaf->so_accept_filter_arg =
    360 		    (*afp->accf_create)(so, afa.af_arg);
    361 		if (newaf->so_accept_filter_arg == NULL) {
    362 			error = EINVAL;
    363 			goto out;
    364 		}
    365 	}
    366 	newaf->so_accept_filter = afp;
    367 	so->so_accf = newaf;
    368 	so->so_options |= SO_ACCEPTFILTER;
    369 	newaf = NULL;
    370 out:
    371 	if (newaf != NULL) {
    372 		if (newaf->so_accept_filter_str != NULL)
    373 			kmem_free(newaf->so_accept_filter_str,
    374 			    sizeof(afa.af_name));
    375 		kmem_free(newaf, sizeof(*newaf));
    376 		atomic_dec_uint(&afp->accf_refcnt);
    377 	}
    378 	return error;
    379 }
    380