Home | History | Annotate | Line # | Download | only in npf
npf_os.c revision 1.18 
      1 /*-
      2  * Copyright (c) 2009-2016 The NetBSD Foundation, Inc.
      3  * All rights reserved.
      4  *
      5  * This material is based upon work partially supported by The
      6  * NetBSD Foundation under a contract with Mindaugas Rasiukevicius.
      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 NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     18  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     19  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     20  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     21  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     22  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     23  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     24  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     25  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     26  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     27  * POSSIBILITY OF SUCH DAMAGE.
     28  */
     29 
     30 /*
     31  * NPF main: dynamic load/initialisation and unload routines.
     32  */
     33 
     34 #ifdef _KERNEL
     35 #include <sys/cdefs.h>
     36 __KERNEL_RCSID(0, "$NetBSD: npf_os.c,v 1.18 2020/05/30 14:16:56 rmind Exp $");
     37 
     38 #ifdef _KERNEL_OPT
     39 #include "pf.h"
     40 #if NPF > 0
     41 #error "NPF and PF are mutually exclusive; please select one"
     42 #endif
     43 #endif
     44 
     45 #include <sys/param.h>
     46 #include <sys/types.h>
     47 
     48 #include <sys/conf.h>
     49 #include <sys/kauth.h>
     50 #include <sys/kmem.h>
     51 #include <sys/lwp.h>
     52 #include <sys/module.h>
     53 #include <sys/pserialize.h>
     54 #include <sys/socketvar.h>
     55 #include <sys/uio.h>
     56 
     57 #include <netinet/in.h>
     58 #include <netinet6/in6_var.h>
     59 #endif
     60 
     61 #include "npf_impl.h"
     62 #include "npfkern.h"
     63 
     64 #ifdef _KERNEL
     65 #ifndef _MODULE
     66 #include "opt_modular.h"
     67 #include "opt_net_mpsafe.h"
     68 #endif
     69 #include "ioconf.h"
     70 #endif
     71 
     72 /*
     73  * Module and device structures.
     74  */
     75 #ifndef _MODULE
     76 /*
     77  * Modular kernels load drivers too early, and we need percpu to be inited
     78  * So we make this misc; a better way would be to have early boot and late
     79  * boot drivers.
     80  */
     81 MODULE(MODULE_CLASS_MISC, npf, "bpf");
     82 #else
     83 /* This module autoloads via /dev/npf so it needs to be a driver */
     84 MODULE(MODULE_CLASS_DRIVER, npf, "bpf");
     85 #endif
     86 
     87 #define	NPF_IOCTL_DATA_LIMIT	(4 * 1024 * 1024)
     88 
     89 static int	npf_pfil_register(bool);
     90 static void	npf_pfil_unregister(bool);
     91 
     92 static int	npf_dev_open(dev_t, int, int, lwp_t *);
     93 static int	npf_dev_close(dev_t, int, int, lwp_t *);
     94 static int	npf_dev_ioctl(dev_t, u_long, void *, int, lwp_t *);
     95 static int	npf_dev_poll(dev_t, int, lwp_t *);
     96 static int	npf_dev_read(dev_t, struct uio *, int);
     97 
     98 const struct cdevsw npf_cdevsw = {
     99 	.d_open = npf_dev_open,
    100 	.d_close = npf_dev_close,
    101 	.d_read = npf_dev_read,
    102 	.d_write = nowrite,
    103 	.d_ioctl = npf_dev_ioctl,
    104 	.d_stop = nostop,
    105 	.d_tty = notty,
    106 	.d_poll = npf_dev_poll,
    107 	.d_mmap = nommap,
    108 	.d_kqfilter = nokqfilter,
    109 	.d_discard = nodiscard,
    110 	.d_flag = D_OTHER | D_MPSAFE
    111 };
    112 
    113 static const char *	npf_ifop_getname(npf_t *, ifnet_t *);
    114 static ifnet_t *	npf_ifop_lookup(npf_t *, const char *);
    115 static void		npf_ifop_flush(npf_t *, void *);
    116 static void *		npf_ifop_getmeta(npf_t *, const ifnet_t *);
    117 static void		npf_ifop_setmeta(npf_t *, ifnet_t *, void *);
    118 
    119 static const unsigned	nworkers = 1;
    120 
    121 static bool		pfil_registered = false;
    122 static pfil_head_t *	npf_ph_if = NULL;
    123 static pfil_head_t *	npf_ph_inet = NULL;
    124 static pfil_head_t *	npf_ph_inet6 = NULL;
    125 
    126 static const npf_ifops_t kern_ifops = {
    127 	.getname	= npf_ifop_getname,
    128 	.lookup		= npf_ifop_lookup,
    129 	.flush		= npf_ifop_flush,
    130 	.getmeta	= npf_ifop_getmeta,
    131 	.setmeta	= npf_ifop_setmeta,
    132 };
    133 
    134 static int
    135 npf_fini(void)
    136 {
    137 	npf_t *npf = npf_getkernctx();
    138 
    139 	/* At first, detach device and remove pfil hooks. */
    140 #ifdef _MODULE
    141 	devsw_detach(NULL, &npf_cdevsw);
    142 #endif
    143 	npf_pfil_unregister(true);
    144 	npfk_destroy(npf);
    145 	npfk_sysfini();
    146 	return 0;
    147 }
    148 
    149 static int
    150 npf_init(void)
    151 {
    152 	npf_t *npf;
    153 	int error = 0;
    154 
    155 	error = npfk_sysinit(nworkers);
    156 	if (error)
    157 		return error;
    158 	npf = npfk_create(0, NULL, &kern_ifops, NULL);
    159 	npf_setkernctx(npf);
    160 	npf_pfil_register(true);
    161 
    162 #ifdef _MODULE
    163 	devmajor_t bmajor = NODEVMAJOR, cmajor = NODEVMAJOR;
    164 
    165 	/* Attach /dev/npf device. */
    166 	error = devsw_attach("npf", NULL, &bmajor, &npf_cdevsw, &cmajor);
    167 	if (error) {
    168 		/* It will call devsw_detach(), which is safe. */
    169 		(void)npf_fini();
    170 	}
    171 #endif
    172 	return error;
    173 }
    174 
    175 
    176 /*
    177  * Module interface.
    178  */
    179 static int
    180 npf_modcmd(modcmd_t cmd, void *arg)
    181 {
    182 	switch (cmd) {
    183 	case MODULE_CMD_INIT:
    184 		return npf_init();
    185 	case MODULE_CMD_FINI:
    186 		return npf_fini();
    187 	case MODULE_CMD_AUTOUNLOAD:
    188 		if (npf_autounload_p()) {
    189 			return EBUSY;
    190 		}
    191 		break;
    192 	default:
    193 		return ENOTTY;
    194 	}
    195 	return 0;
    196 }
    197 
    198 void
    199 npfattach(int nunits)
    200 {
    201 	/* Nothing */
    202 }
    203 
    204 static int
    205 npf_dev_open(dev_t dev, int flag, int mode, lwp_t *l)
    206 {
    207 	/* Available only for super-user. */
    208 	if (kauth_authorize_network(l->l_cred, KAUTH_NETWORK_FIREWALL,
    209 	    KAUTH_REQ_NETWORK_FIREWALL_FW, NULL, NULL, NULL)) {
    210 		return EPERM;
    211 	}
    212 	return 0;
    213 }
    214 
    215 static int
    216 npf_dev_close(dev_t dev, int flag, int mode, lwp_t *l)
    217 {
    218 	return 0;
    219 }
    220 
    221 static int
    222 npf_stats_export(npf_t *npf, void *data)
    223 {
    224 	uint64_t *fullst, *uptr = *(uint64_t **)data;
    225 	int error;
    226 
    227 	fullst = kmem_alloc(NPF_STATS_SIZE, KM_SLEEP);
    228 	npfk_stats(npf, fullst); /* will zero the buffer */
    229 	error = copyout(fullst, uptr, NPF_STATS_SIZE);
    230 	kmem_free(fullst, NPF_STATS_SIZE);
    231 	return error;
    232 }
    233 
    234 /*
    235  * npfctl_switch: enable or disable packet inspection.
    236  */
    237 static int
    238 npfctl_switch(void *data)
    239 {
    240 	const bool onoff = *(int *)data ? true : false;
    241 	int error;
    242 
    243 	if (onoff) {
    244 		/* Enable: add pfil hooks. */
    245 		error = npf_pfil_register(false);
    246 	} else {
    247 		/* Disable: remove pfil hooks. */
    248 		npf_pfil_unregister(false);
    249 		error = 0;
    250 	}
    251 	return error;
    252 }
    253 
    254 static int
    255 npf_dev_ioctl(dev_t dev, u_long cmd, void *data, int flag, lwp_t *l)
    256 {
    257 	npf_t *npf = npf_getkernctx();
    258 	nvlist_t *req, *resp;
    259 	int error;
    260 
    261 	/* Available only for super-user. */
    262 	if (kauth_authorize_network(l->l_cred, KAUTH_NETWORK_FIREWALL,
    263 	    KAUTH_REQ_NETWORK_FIREWALL_FW, NULL, NULL, NULL)) {
    264 		return EPERM;
    265 	}
    266 
    267 	switch (cmd) {
    268 	case IOC_NPF_VERSION:
    269 		*(int *)data = NPF_VERSION;
    270 		return 0;
    271 
    272 	case IOC_NPF_SWITCH:
    273 		return npfctl_switch(data);
    274 
    275 	case IOC_NPF_TABLE:
    276 		return npfctl_table(npf, data);
    277 
    278 	case IOC_NPF_STATS:
    279 		return npf_stats_export(npf, data);
    280 	}
    281 
    282 	error = nvlist_copyin(data, &req, NPF_IOCTL_DATA_LIMIT);
    283 	if (__predict_false(error)) {
    284 #ifdef __NetBSD__
    285 		/* Until the version bump. */
    286 		if (cmd != IOC_NPF_SAVE) {
    287 			return error;
    288 		}
    289 		req = nvlist_create(0);
    290 #else
    291 		return error;
    292 #endif
    293 	}
    294 	resp = nvlist_create(0);
    295 	npfctl_run_op(npf, cmd, req, resp);
    296 	error = nvlist_copyout(data, resp);
    297 	nvlist_destroy(resp);
    298 	nvlist_destroy(req);
    299 
    300 	return error;
    301 }
    302 
    303 static int
    304 npf_dev_poll(dev_t dev, int events, lwp_t *l)
    305 {
    306 	return ENOTSUP;
    307 }
    308 
    309 static int
    310 npf_dev_read(dev_t dev, struct uio *uio, int flag)
    311 {
    312 	return ENOTSUP;
    313 }
    314 
    315 bool
    316 npf_autounload_p(void)
    317 {
    318 	npf_t *npf = npf_getkernctx();
    319 	return !npf_active_p() && npf_default_pass(npf);
    320 }
    321 
    322 /*
    323  * Interface operations.
    324  */
    325 
    326 static const char *
    327 npf_ifop_getname(npf_t *npf __unused, ifnet_t *ifp)
    328 {
    329 	return ifp->if_xname;
    330 }
    331 
    332 static ifnet_t *
    333 npf_ifop_lookup(npf_t *npf __unused, const char *name)
    334 {
    335 	return ifunit(name);
    336 }
    337 
    338 static void
    339 npf_ifop_flush(npf_t *npf __unused, void *arg)
    340 {
    341 	ifnet_t *ifp;
    342 
    343 	KERNEL_LOCK(1, NULL);
    344 	IFNET_GLOBAL_LOCK();
    345 	IFNET_WRITER_FOREACH(ifp) {
    346 		ifp->if_npf_private = arg;
    347 	}
    348 	IFNET_GLOBAL_UNLOCK();
    349 	KERNEL_UNLOCK_ONE(NULL);
    350 }
    351 
    352 static void *
    353 npf_ifop_getmeta(npf_t *npf __unused, const ifnet_t *ifp)
    354 {
    355 	return ifp->if_npf_private;
    356 }
    357 
    358 static void
    359 npf_ifop_setmeta(npf_t *npf __unused, ifnet_t *ifp, void *arg)
    360 {
    361 	ifp->if_npf_private = arg;
    362 }
    363 
    364 #ifdef _KERNEL
    365 
    366 /*
    367  * Wrapper of the main packet handler to pass the kernel NPF context.
    368  */
    369 static int
    370 npfos_packet_handler(void *arg, struct mbuf **mp, ifnet_t *ifp, int di)
    371 {
    372 	npf_t *npf = npf_getkernctx();
    373 	return npfk_packet_handler(npf, mp, ifp, di);
    374 }
    375 
    376 /*
    377  * npf_ifhook: hook handling interface changes.
    378  */
    379 static void
    380 npf_ifhook(void *arg, unsigned long cmd, void *arg2)
    381 {
    382 	npf_t *npf = npf_getkernctx();
    383 	ifnet_t *ifp = arg2;
    384 
    385 	switch (cmd) {
    386 	case PFIL_IFNET_ATTACH:
    387 		npfk_ifmap_attach(npf, ifp);
    388 		npf_ifaddr_sync(npf, ifp);
    389 		break;
    390 	case PFIL_IFNET_DETACH:
    391 		npfk_ifmap_detach(npf, ifp);
    392 		npf_ifaddr_flush(npf, ifp);
    393 		break;
    394 	}
    395 }
    396 
    397 static void
    398 npf_ifaddrhook(void *arg, u_long cmd, void *arg2)
    399 {
    400 	npf_t *npf = npf_getkernctx();
    401 	struct ifaddr *ifa = arg2;
    402 
    403 	switch (cmd) {
    404 	case SIOCSIFADDR:
    405 	case SIOCAIFADDR:
    406 	case SIOCDIFADDR:
    407 #ifdef INET6
    408 	case SIOCSIFADDR_IN6:
    409 	case SIOCAIFADDR_IN6:
    410 	case SIOCDIFADDR_IN6:
    411 #endif
    412 		KASSERT(ifa != NULL);
    413 		break;
    414 	default:
    415 		return;
    416 	}
    417 	npf_ifaddr_sync(npf, ifa->ifa_ifp);
    418 }
    419 
    420 /*
    421  * npf_pfil_register: register pfil(9) hooks.
    422  */
    423 static int
    424 npf_pfil_register(bool init)
    425 {
    426 	npf_t *npf = npf_getkernctx();
    427 	int error = 0;
    428 
    429 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    430 
    431 	/* Init: interface re-config and attach/detach hook. */
    432 	if (!npf_ph_if) {
    433 		npf_ph_if = pfil_head_get(PFIL_TYPE_IFNET, 0);
    434 		if (!npf_ph_if) {
    435 			error = ENOENT;
    436 			goto out;
    437 		}
    438 
    439 		error = pfil_add_ihook(npf_ifhook, NULL,
    440 		    PFIL_IFNET, npf_ph_if);
    441 		KASSERT(error == 0);
    442 
    443 		error = pfil_add_ihook(npf_ifaddrhook, NULL,
    444 		    PFIL_IFADDR, npf_ph_if);
    445 		KASSERT(error == 0);
    446 	}
    447 	if (init) {
    448 		goto out;
    449 	}
    450 
    451 	/* Check if pfil hooks are not already registered. */
    452 	if (pfil_registered) {
    453 		error = EEXIST;
    454 		goto out;
    455 	}
    456 
    457 	/* Capture points of the activity in the IP layer. */
    458 	npf_ph_inet = pfil_head_get(PFIL_TYPE_AF, (void *)AF_INET);
    459 	npf_ph_inet6 = pfil_head_get(PFIL_TYPE_AF, (void *)AF_INET6);
    460 	if (!npf_ph_inet && !npf_ph_inet6) {
    461 		error = ENOENT;
    462 		goto out;
    463 	}
    464 
    465 	/* Packet IN/OUT handlers for IP layer. */
    466 	if (npf_ph_inet) {
    467 		error = pfil_add_hook(npfos_packet_handler, npf,
    468 		    PFIL_ALL, npf_ph_inet);
    469 		KASSERT(error == 0);
    470 	}
    471 	if (npf_ph_inet6) {
    472 		error = pfil_add_hook(npfos_packet_handler, npf,
    473 		    PFIL_ALL, npf_ph_inet6);
    474 		KASSERT(error == 0);
    475 	}
    476 
    477 	/*
    478 	 * It is necessary to re-sync all/any interface address tables,
    479 	 * since we did not listen for any changes.
    480 	 */
    481 	npf_ifaddr_syncall(npf);
    482 	pfil_registered = true;
    483 out:
    484 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    485 
    486 	return error;
    487 }
    488 
    489 /*
    490  * npf_pfil_unregister: unregister pfil(9) hooks.
    491  */
    492 static void
    493 npf_pfil_unregister(bool fini)
    494 {
    495 	npf_t *npf = npf_getkernctx();
    496 
    497 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    498 
    499 	if (fini && npf_ph_if) {
    500 		(void)pfil_remove_ihook(npf_ifhook, NULL,
    501 		    PFIL_IFNET, npf_ph_if);
    502 		(void)pfil_remove_ihook(npf_ifaddrhook, NULL,
    503 		    PFIL_IFADDR, npf_ph_if);
    504 	}
    505 	if (npf_ph_inet) {
    506 		(void)pfil_remove_hook(npfos_packet_handler, npf,
    507 		    PFIL_ALL, npf_ph_inet);
    508 	}
    509 	if (npf_ph_inet6) {
    510 		(void)pfil_remove_hook(npfos_packet_handler, npf,
    511 		    PFIL_ALL, npf_ph_inet6);
    512 	}
    513 	pfil_registered = false;
    514 
    515 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    516 }
    517 
    518 bool
    519 npf_active_p(void)
    520 {
    521 	return pfil_registered;
    522 }
    523 
    524 #endif
    525 
    526 #ifdef __NetBSD__
    527 
    528 /*
    529  * Epoch-Based Reclamation (EBR) wrappers: in NetBSD, we rely on the
    530  * passive serialization mechanism (see pserialize(9) manual page),
    531  * which provides sufficient guarantees for NPF.
    532  */
    533 
    534 ebr_t *
    535 npf_ebr_create(void)
    536 {
    537 	return pserialize_create();
    538 }
    539 
    540 void
    541 npf_ebr_destroy(ebr_t *ebr)
    542 {
    543 	pserialize_destroy(ebr);
    544 }
    545 
    546 void
    547 npf_ebr_register(ebr_t *ebr)
    548 {
    549 	KASSERT(ebr != NULL); (void)ebr;
    550 }
    551 
    552 void
    553 npf_ebr_unregister(ebr_t *ebr)
    554 {
    555 	KASSERT(ebr != NULL); (void)ebr;
    556 }
    557 
    558 int
    559 npf_ebr_enter(ebr_t *ebr)
    560 {
    561 	KASSERT(ebr != NULL); (void)ebr;
    562 	return pserialize_read_enter();
    563 }
    564 
    565 void
    566 npf_ebr_exit(ebr_t *ebr, int s)
    567 {
    568 	KASSERT(ebr != NULL); (void)ebr;
    569 	pserialize_read_exit(s);
    570 }
    571 
    572 void
    573 npf_ebr_full_sync(ebr_t *ebr)
    574 {
    575 	pserialize_perform(ebr);
    576 }
    577 
    578 bool
    579 npf_ebr_incrit_p(ebr_t *ebr)
    580 {
    581 	KASSERT(ebr != NULL); (void)ebr;
    582 	return pserialize_in_read_section();
    583 }
    584 
    585 #endif
    586