Home | History | Annotate | Line # | Download | only in kern
uipc_socket.c revision 1.151.6.5
      1  1.151.6.1       mjf /*	$NetBSD: uipc_socket.c,v 1.151.6.5 2009/01/17 13:29:20 mjf Exp $	*/
      2       1.64   thorpej 
      3       1.64   thorpej /*-
      4  1.151.6.1       mjf  * Copyright (c) 2002, 2007, 2008 The NetBSD Foundation, Inc.
      5       1.64   thorpej  * All rights reserved.
      6       1.64   thorpej  *
      7       1.64   thorpej  * This code is derived from software contributed to The NetBSD Foundation
      8       1.64   thorpej  * by Jason R. Thorpe of Wasabi Systems, Inc.
      9       1.64   thorpej  *
     10       1.64   thorpej  * Redistribution and use in source and binary forms, with or without
     11       1.64   thorpej  * modification, are permitted provided that the following conditions
     12       1.64   thorpej  * are met:
     13       1.64   thorpej  * 1. Redistributions of source code must retain the above copyright
     14       1.64   thorpej  *    notice, this list of conditions and the following disclaimer.
     15       1.64   thorpej  * 2. Redistributions in binary form must reproduce the above copyright
     16       1.64   thorpej  *    notice, this list of conditions and the following disclaimer in the
     17       1.64   thorpej  *    documentation and/or other materials provided with the distribution.
     18       1.64   thorpej  *
     19       1.64   thorpej  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20       1.64   thorpej  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21       1.64   thorpej  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22       1.64   thorpej  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23       1.64   thorpej  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24       1.64   thorpej  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25       1.64   thorpej  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26       1.64   thorpej  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27       1.64   thorpej  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28       1.64   thorpej  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29       1.64   thorpej  * POSSIBILITY OF SUCH DAMAGE.
     30       1.64   thorpej  */
     31       1.16       cgd 
     32        1.1       cgd /*
     33  1.151.6.2       mjf  * Copyright (c) 2004 The FreeBSD Foundation
     34  1.151.6.2       mjf  * Copyright (c) 2004 Robert Watson
     35       1.15   mycroft  * Copyright (c) 1982, 1986, 1988, 1990, 1993
     36       1.15   mycroft  *	The Regents of the University of California.  All rights reserved.
     37        1.1       cgd  *
     38        1.1       cgd  * Redistribution and use in source and binary forms, with or without
     39        1.1       cgd  * modification, are permitted provided that the following conditions
     40        1.1       cgd  * are met:
     41        1.1       cgd  * 1. Redistributions of source code must retain the above copyright
     42        1.1       cgd  *    notice, this list of conditions and the following disclaimer.
     43        1.1       cgd  * 2. Redistributions in binary form must reproduce the above copyright
     44        1.1       cgd  *    notice, this list of conditions and the following disclaimer in the
     45        1.1       cgd  *    documentation and/or other materials provided with the distribution.
     46       1.85       agc  * 3. Neither the name of the University nor the names of its contributors
     47        1.1       cgd  *    may be used to endorse or promote products derived from this software
     48        1.1       cgd  *    without specific prior written permission.
     49        1.1       cgd  *
     50        1.1       cgd  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     51        1.1       cgd  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     52        1.1       cgd  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     53        1.1       cgd  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     54        1.1       cgd  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     55        1.1       cgd  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     56        1.1       cgd  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     57        1.1       cgd  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     58        1.1       cgd  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     59        1.1       cgd  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     60        1.1       cgd  * SUCH DAMAGE.
     61        1.1       cgd  *
     62       1.32      fvdl  *	@(#)uipc_socket.c	8.6 (Berkeley) 5/2/95
     63        1.1       cgd  */
     64       1.59     lukem 
     65       1.59     lukem #include <sys/cdefs.h>
     66  1.151.6.1       mjf __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.151.6.5 2009/01/17 13:29:20 mjf Exp $");
     67       1.64   thorpej 
     68  1.151.6.5       mjf #include "opt_compat_netbsd.h"
     69       1.64   thorpej #include "opt_sock_counters.h"
     70       1.64   thorpej #include "opt_sosend_loan.h"
     71       1.81    martin #include "opt_mbuftrace.h"
     72       1.84     ragge #include "opt_somaxkva.h"
     73  1.151.6.2       mjf #include "opt_multiprocessor.h"	/* XXX */
     74        1.1       cgd 
     75        1.9   mycroft #include <sys/param.h>
     76        1.9   mycroft #include <sys/systm.h>
     77        1.9   mycroft #include <sys/proc.h>
     78        1.9   mycroft #include <sys/file.h>
     79      1.142    dyoung #include <sys/filedesc.h>
     80  1.151.6.5       mjf #include <sys/kmem.h>
     81        1.9   mycroft #include <sys/mbuf.h>
     82        1.9   mycroft #include <sys/domain.h>
     83        1.9   mycroft #include <sys/kernel.h>
     84        1.9   mycroft #include <sys/protosw.h>
     85        1.9   mycroft #include <sys/socket.h>
     86        1.9   mycroft #include <sys/socketvar.h>
     87       1.21  christos #include <sys/signalvar.h>
     88        1.9   mycroft #include <sys/resourcevar.h>
     89  1.151.6.5       mjf #include <sys/uidinfo.h>
     90       1.72  jdolecek #include <sys/event.h>
     91       1.89  christos #include <sys/poll.h>
     92      1.118      elad #include <sys/kauth.h>
     93      1.136        ad #include <sys/mutex.h>
     94      1.136        ad #include <sys/condvar.h>
     95       1.37   thorpej 
     96  1.151.6.5       mjf #ifdef COMPAT_50
     97  1.151.6.5       mjf #include <compat/sys/time.h>
     98  1.151.6.5       mjf #define	SO_OSNDTIMEO	0x1005
     99  1.151.6.5       mjf #define	SO_ORCVTIMEO	0x1006
    100  1.151.6.5       mjf #endif
    101  1.151.6.5       mjf 
    102       1.64   thorpej #include <uvm/uvm.h>
    103       1.64   thorpej 
    104       1.77   thorpej MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
    105       1.77   thorpej MALLOC_DEFINE(M_SONAME, "soname", "socket name");
    106       1.37   thorpej 
    107      1.142    dyoung extern const struct fileops socketops;
    108      1.142    dyoung 
    109       1.54     lukem extern int	somaxconn;			/* patchable (XXX sysctl) */
    110       1.54     lukem int		somaxconn = SOMAXCONN;
    111  1.151.6.2       mjf kmutex_t	*softnet_lock;
    112       1.49  jonathan 
    113       1.64   thorpej #ifdef SOSEND_COUNTERS
    114       1.64   thorpej #include <sys/device.h>
    115       1.64   thorpej 
    116      1.113   thorpej static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    117       1.64   thorpej     NULL, "sosend", "loan big");
    118      1.113   thorpej static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    119       1.64   thorpej     NULL, "sosend", "copy big");
    120      1.113   thorpej static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    121       1.64   thorpej     NULL, "sosend", "copy small");
    122      1.113   thorpej static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    123       1.64   thorpej     NULL, "sosend", "kva limit");
    124       1.64   thorpej 
    125       1.64   thorpej #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
    126       1.64   thorpej 
    127      1.101      matt EVCNT_ATTACH_STATIC(sosend_loan_big);
    128      1.101      matt EVCNT_ATTACH_STATIC(sosend_copy_big);
    129      1.101      matt EVCNT_ATTACH_STATIC(sosend_copy_small);
    130      1.101      matt EVCNT_ATTACH_STATIC(sosend_kvalimit);
    131       1.64   thorpej #else
    132       1.64   thorpej 
    133       1.64   thorpej #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
    134       1.64   thorpej 
    135       1.64   thorpej #endif /* SOSEND_COUNTERS */
    136       1.64   thorpej 
    137      1.119      yamt static struct callback_entry sokva_reclaimerentry;
    138        1.1       cgd 
    139  1.151.6.2       mjf #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
    140      1.121      yamt int sock_loan_thresh = -1;
    141       1.71   thorpej #else
    142      1.121      yamt int sock_loan_thresh = 4096;
    143       1.65   thorpej #endif
    144       1.64   thorpej 
    145      1.136        ad static kmutex_t so_pendfree_lock;
    146      1.113   thorpej static struct mbuf *so_pendfree;
    147       1.64   thorpej 
    148       1.84     ragge #ifndef SOMAXKVA
    149       1.84     ragge #define	SOMAXKVA (16 * 1024 * 1024)
    150       1.84     ragge #endif
    151       1.84     ragge int somaxkva = SOMAXKVA;
    152      1.113   thorpej static int socurkva;
    153      1.136        ad static kcondvar_t socurkva_cv;
    154       1.64   thorpej 
    155       1.64   thorpej #define	SOCK_LOAN_CHUNK		65536
    156       1.64   thorpej 
    157      1.117      yamt static size_t sodopendfree(void);
    158      1.117      yamt static size_t sodopendfreel(void);
    159       1.93      yamt 
    160  1.151.6.5       mjf static void sysctl_kern_somaxkva_setup(void);
    161  1.151.6.5       mjf static struct sysctllog *socket_sysctllog;
    162  1.151.6.5       mjf 
    163      1.113   thorpej static vsize_t
    164      1.129      yamt sokvareserve(struct socket *so, vsize_t len)
    165       1.80      yamt {
    166       1.98  christos 	int error;
    167       1.80      yamt 
    168      1.136        ad 	mutex_enter(&so_pendfree_lock);
    169       1.80      yamt 	while (socurkva + len > somaxkva) {
    170       1.93      yamt 		size_t freed;
    171       1.93      yamt 
    172       1.93      yamt 		/*
    173       1.93      yamt 		 * try to do pendfree.
    174       1.93      yamt 		 */
    175       1.93      yamt 
    176      1.117      yamt 		freed = sodopendfreel();
    177       1.93      yamt 
    178       1.93      yamt 		/*
    179       1.93      yamt 		 * if some kva was freed, try again.
    180       1.93      yamt 		 */
    181       1.93      yamt 
    182       1.93      yamt 		if (freed)
    183       1.80      yamt 			continue;
    184       1.93      yamt 
    185       1.80      yamt 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
    186      1.136        ad 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
    187       1.98  christos 		if (error) {
    188       1.98  christos 			len = 0;
    189       1.98  christos 			break;
    190       1.98  christos 		}
    191       1.80      yamt 	}
    192       1.93      yamt 	socurkva += len;
    193      1.136        ad 	mutex_exit(&so_pendfree_lock);
    194       1.98  christos 	return len;
    195       1.95      yamt }
    196       1.95      yamt 
    197      1.113   thorpej static void
    198       1.95      yamt sokvaunreserve(vsize_t len)
    199       1.95      yamt {
    200       1.95      yamt 
    201      1.136        ad 	mutex_enter(&so_pendfree_lock);
    202       1.95      yamt 	socurkva -= len;
    203      1.136        ad 	cv_broadcast(&socurkva_cv);
    204      1.136        ad 	mutex_exit(&so_pendfree_lock);
    205       1.95      yamt }
    206       1.95      yamt 
    207       1.95      yamt /*
    208       1.95      yamt  * sokvaalloc: allocate kva for loan.
    209       1.95      yamt  */
    210       1.95      yamt 
    211       1.95      yamt vaddr_t
    212       1.95      yamt sokvaalloc(vsize_t len, struct socket *so)
    213       1.95      yamt {
    214       1.95      yamt 	vaddr_t lva;
    215       1.95      yamt 
    216       1.95      yamt 	/*
    217       1.95      yamt 	 * reserve kva.
    218       1.95      yamt 	 */
    219       1.95      yamt 
    220       1.98  christos 	if (sokvareserve(so, len) == 0)
    221       1.98  christos 		return 0;
    222       1.93      yamt 
    223       1.93      yamt 	/*
    224       1.93      yamt 	 * allocate kva.
    225       1.93      yamt 	 */
    226       1.80      yamt 
    227      1.109      yamt 	lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
    228       1.95      yamt 	if (lva == 0) {
    229       1.95      yamt 		sokvaunreserve(len);
    230       1.80      yamt 		return (0);
    231       1.95      yamt 	}
    232       1.80      yamt 
    233       1.80      yamt 	return lva;
    234       1.80      yamt }
    235       1.80      yamt 
    236       1.93      yamt /*
    237       1.93      yamt  * sokvafree: free kva for loan.
    238       1.93      yamt  */
    239       1.93      yamt 
    240       1.80      yamt void
    241       1.80      yamt sokvafree(vaddr_t sva, vsize_t len)
    242       1.80      yamt {
    243       1.93      yamt 
    244       1.93      yamt 	/*
    245       1.93      yamt 	 * free kva.
    246       1.93      yamt 	 */
    247       1.80      yamt 
    248      1.109      yamt 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
    249       1.93      yamt 
    250       1.93      yamt 	/*
    251       1.93      yamt 	 * unreserve kva.
    252       1.93      yamt 	 */
    253       1.93      yamt 
    254       1.95      yamt 	sokvaunreserve(len);
    255       1.80      yamt }
    256       1.80      yamt 
    257       1.64   thorpej static void
    258      1.134  christos sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
    259       1.64   thorpej {
    260  1.151.6.1       mjf 	vaddr_t sva, eva;
    261       1.64   thorpej 	vsize_t len;
    262  1.151.6.1       mjf 	int npgs;
    263  1.151.6.1       mjf 
    264  1.151.6.1       mjf 	KASSERT(pgs != NULL);
    265       1.64   thorpej 
    266       1.64   thorpej 	eva = round_page((vaddr_t) buf + size);
    267       1.64   thorpej 	sva = trunc_page((vaddr_t) buf);
    268       1.64   thorpej 	len = eva - sva;
    269       1.64   thorpej 	npgs = len >> PAGE_SHIFT;
    270       1.64   thorpej 
    271       1.64   thorpej 	pmap_kremove(sva, len);
    272       1.64   thorpej 	pmap_update(pmap_kernel());
    273       1.64   thorpej 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
    274       1.80      yamt 	sokvafree(sva, len);
    275       1.64   thorpej }
    276       1.64   thorpej 
    277       1.64   thorpej static size_t
    278  1.151.6.1       mjf sodopendfree(void)
    279       1.64   thorpej {
    280       1.93      yamt 	size_t rv;
    281       1.64   thorpej 
    282  1.151.6.2       mjf 	if (__predict_true(so_pendfree == NULL))
    283  1.151.6.2       mjf 		return 0;
    284  1.151.6.2       mjf 
    285      1.136        ad 	mutex_enter(&so_pendfree_lock);
    286      1.117      yamt 	rv = sodopendfreel();
    287      1.136        ad 	mutex_exit(&so_pendfree_lock);
    288       1.93      yamt 
    289       1.93      yamt 	return rv;
    290       1.93      yamt }
    291       1.93      yamt 
    292       1.93      yamt /*
    293       1.93      yamt  * sodopendfreel: free mbufs on "pendfree" list.
    294      1.136        ad  * unlock and relock so_pendfree_lock when freeing mbufs.
    295       1.93      yamt  *
    296      1.136        ad  * => called with so_pendfree_lock held.
    297       1.93      yamt  */
    298       1.93      yamt 
    299       1.93      yamt static size_t
    300  1.151.6.1       mjf sodopendfreel(void)
    301       1.93      yamt {
    302      1.137        ad 	struct mbuf *m, *next;
    303       1.93      yamt 	size_t rv = 0;
    304       1.93      yamt 
    305      1.136        ad 	KASSERT(mutex_owned(&so_pendfree_lock));
    306       1.64   thorpej 
    307      1.137        ad 	while (so_pendfree != NULL) {
    308       1.64   thorpej 		m = so_pendfree;
    309       1.93      yamt 		so_pendfree = NULL;
    310      1.136        ad 		mutex_exit(&so_pendfree_lock);
    311       1.93      yamt 
    312       1.93      yamt 		for (; m != NULL; m = next) {
    313       1.93      yamt 			next = m->m_next;
    314  1.151.6.1       mjf 			KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
    315  1.151.6.1       mjf 			KASSERT(m->m_ext.ext_refcnt == 0);
    316       1.93      yamt 
    317       1.93      yamt 			rv += m->m_ext.ext_size;
    318  1.151.6.1       mjf 			sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
    319       1.93      yamt 			    m->m_ext.ext_size);
    320      1.145        ad 			pool_cache_put(mb_cache, m);
    321       1.93      yamt 		}
    322       1.64   thorpej 
    323      1.136        ad 		mutex_enter(&so_pendfree_lock);
    324       1.64   thorpej 	}
    325       1.64   thorpej 
    326       1.64   thorpej 	return (rv);
    327       1.64   thorpej }
    328       1.64   thorpej 
    329       1.80      yamt void
    330      1.134  christos soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
    331       1.64   thorpej {
    332       1.64   thorpej 
    333  1.151.6.1       mjf 	KASSERT(m != NULL);
    334       1.64   thorpej 
    335       1.93      yamt 	/*
    336       1.93      yamt 	 * postpone freeing mbuf.
    337       1.93      yamt 	 *
    338       1.93      yamt 	 * we can't do it in interrupt context
    339       1.93      yamt 	 * because we need to put kva back to kernel_map.
    340       1.93      yamt 	 */
    341       1.93      yamt 
    342      1.136        ad 	mutex_enter(&so_pendfree_lock);
    343       1.92      yamt 	m->m_next = so_pendfree;
    344       1.92      yamt 	so_pendfree = m;
    345      1.136        ad 	cv_broadcast(&socurkva_cv);
    346      1.136        ad 	mutex_exit(&so_pendfree_lock);
    347       1.64   thorpej }
    348       1.64   thorpej 
    349       1.64   thorpej static long
    350       1.64   thorpej sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
    351       1.64   thorpej {
    352       1.64   thorpej 	struct iovec *iov = uio->uio_iov;
    353       1.64   thorpej 	vaddr_t sva, eva;
    354       1.64   thorpej 	vsize_t len;
    355  1.151.6.1       mjf 	vaddr_t lva;
    356  1.151.6.1       mjf 	int npgs, error;
    357  1.151.6.1       mjf 	vaddr_t va;
    358  1.151.6.1       mjf 	int i;
    359       1.64   thorpej 
    360      1.116      yamt 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
    361       1.64   thorpej 		return (0);
    362       1.64   thorpej 
    363       1.64   thorpej 	if (iov->iov_len < (size_t) space)
    364       1.64   thorpej 		space = iov->iov_len;
    365       1.64   thorpej 	if (space > SOCK_LOAN_CHUNK)
    366       1.64   thorpej 		space = SOCK_LOAN_CHUNK;
    367       1.64   thorpej 
    368       1.64   thorpej 	eva = round_page((vaddr_t) iov->iov_base + space);
    369       1.64   thorpej 	sva = trunc_page((vaddr_t) iov->iov_base);
    370       1.64   thorpej 	len = eva - sva;
    371       1.64   thorpej 	npgs = len >> PAGE_SHIFT;
    372       1.64   thorpej 
    373       1.79   thorpej 	KASSERT(npgs <= M_EXT_MAXPAGES);
    374       1.79   thorpej 
    375       1.80      yamt 	lva = sokvaalloc(len, so);
    376       1.64   thorpej 	if (lva == 0)
    377       1.80      yamt 		return 0;
    378       1.64   thorpej 
    379      1.116      yamt 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
    380       1.79   thorpej 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
    381       1.64   thorpej 	if (error) {
    382       1.80      yamt 		sokvafree(lva, len);
    383       1.64   thorpej 		return (0);
    384       1.64   thorpej 	}
    385       1.64   thorpej 
    386       1.64   thorpej 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
    387       1.79   thorpej 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
    388       1.79   thorpej 		    VM_PROT_READ);
    389       1.64   thorpej 	pmap_update(pmap_kernel());
    390       1.64   thorpej 
    391       1.64   thorpej 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
    392       1.64   thorpej 
    393      1.134  christos 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
    394       1.79   thorpej 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
    395       1.64   thorpej 
    396       1.64   thorpej 	uio->uio_resid -= space;
    397       1.64   thorpej 	/* uio_offset not updated, not set/used for write(2) */
    398      1.134  christos 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
    399       1.64   thorpej 	uio->uio_iov->iov_len -= space;
    400       1.64   thorpej 	if (uio->uio_iov->iov_len == 0) {
    401       1.64   thorpej 		uio->uio_iov++;
    402       1.64   thorpej 		uio->uio_iovcnt--;
    403       1.64   thorpej 	}
    404       1.64   thorpej 
    405       1.64   thorpej 	return (space);
    406       1.64   thorpej }
    407       1.64   thorpej 
    408      1.119      yamt static int
    409      1.129      yamt sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
    410      1.119      yamt {
    411      1.119      yamt 
    412      1.119      yamt 	KASSERT(ce == &sokva_reclaimerentry);
    413      1.119      yamt 	KASSERT(obj == NULL);
    414      1.119      yamt 
    415      1.119      yamt 	sodopendfree();
    416      1.119      yamt 	if (!vm_map_starved_p(kernel_map)) {
    417      1.119      yamt 		return CALLBACK_CHAIN_ABORT;
    418      1.119      yamt 	}
    419      1.119      yamt 	return CALLBACK_CHAIN_CONTINUE;
    420      1.119      yamt }
    421      1.119      yamt 
    422      1.142    dyoung struct mbuf *
    423      1.147    dyoung getsombuf(struct socket *so, int type)
    424      1.142    dyoung {
    425      1.142    dyoung 	struct mbuf *m;
    426      1.142    dyoung 
    427      1.147    dyoung 	m = m_get(M_WAIT, type);
    428      1.142    dyoung 	MCLAIM(m, so->so_mowner);
    429      1.142    dyoung 	return m;
    430      1.142    dyoung }
    431      1.142    dyoung 
    432      1.119      yamt void
    433      1.119      yamt soinit(void)
    434      1.119      yamt {
    435      1.119      yamt 
    436  1.151.6.5       mjf 	sysctl_kern_somaxkva_setup();
    437  1.151.6.5       mjf 
    438      1.148        ad 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
    439  1.151.6.2       mjf 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    440      1.136        ad 	cv_init(&socurkva_cv, "sokva");
    441  1.151.6.2       mjf 	soinit2();
    442      1.136        ad 
    443      1.119      yamt 	/* Set the initial adjusted socket buffer size. */
    444      1.119      yamt 	if (sb_max_set(sb_max))
    445      1.119      yamt 		panic("bad initial sb_max value: %lu", sb_max);
    446      1.119      yamt 
    447      1.119      yamt 	callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
    448      1.119      yamt 	    &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
    449      1.119      yamt }
    450      1.119      yamt 
    451        1.1       cgd /*
    452        1.1       cgd  * Socket operation routines.
    453        1.1       cgd  * These routines are called by the routines in
    454        1.1       cgd  * sys_socket.c or from a system process, and
    455        1.1       cgd  * implement the semantics of socket operations by
    456        1.1       cgd  * switching out to the protocol specific routines.
    457        1.1       cgd  */
    458        1.1       cgd /*ARGSUSED*/
    459        1.3    andrew int
    460  1.151.6.2       mjf socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
    461  1.151.6.2       mjf 	 struct socket *lockso)
    462        1.1       cgd {
    463       1.99      matt 	const struct protosw	*prp;
    464       1.54     lukem 	struct socket	*so;
    465      1.115      yamt 	uid_t		uid;
    466  1.151.6.2       mjf 	int		error;
    467  1.151.6.2       mjf 	kmutex_t	*lock;
    468        1.1       cgd 
    469      1.132      elad 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
    470      1.132      elad 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
    471      1.132      elad 	    KAUTH_ARG(proto));
    472      1.140    dyoung 	if (error != 0)
    473      1.140    dyoung 		return error;
    474      1.127      elad 
    475        1.1       cgd 	if (proto)
    476        1.1       cgd 		prp = pffindproto(dom, proto, type);
    477        1.1       cgd 	else
    478        1.1       cgd 		prp = pffindtype(dom, type);
    479      1.140    dyoung 	if (prp == NULL) {
    480      1.120  ginsbach 		/* no support for domain */
    481      1.120  ginsbach 		if (pffinddomain(dom) == 0)
    482      1.140    dyoung 			return EAFNOSUPPORT;
    483      1.120  ginsbach 		/* no support for socket type */
    484      1.120  ginsbach 		if (proto == 0 && type != 0)
    485      1.140    dyoung 			return EPROTOTYPE;
    486      1.140    dyoung 		return EPROTONOSUPPORT;
    487      1.120  ginsbach 	}
    488      1.140    dyoung 	if (prp->pr_usrreq == NULL)
    489      1.140    dyoung 		return EPROTONOSUPPORT;
    490        1.1       cgd 	if (prp->pr_type != type)
    491      1.140    dyoung 		return EPROTOTYPE;
    492  1.151.6.2       mjf 
    493  1.151.6.2       mjf 	so = soget(true);
    494        1.1       cgd 	so->so_type = type;
    495        1.1       cgd 	so->so_proto = prp;
    496       1.33      matt 	so->so_send = sosend;
    497       1.33      matt 	so->so_receive = soreceive;
    498       1.78      matt #ifdef MBUFTRACE
    499       1.78      matt 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
    500       1.78      matt 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
    501       1.78      matt 	so->so_mowner = &prp->pr_domain->dom_mowner;
    502       1.78      matt #endif
    503      1.138     rmind 	uid = kauth_cred_geteuid(l->l_cred);
    504      1.115      yamt 	so->so_uidinfo = uid_find(uid);
    505  1.151.6.3       mjf 	so->so_egid = kauth_cred_getegid(l->l_cred);
    506  1.151.6.3       mjf 	so->so_cpid = l->l_proc->p_pid;
    507  1.151.6.2       mjf 	if (lockso != NULL) {
    508  1.151.6.2       mjf 		/* Caller wants us to share a lock. */
    509  1.151.6.2       mjf 		lock = lockso->so_lock;
    510  1.151.6.2       mjf 		so->so_lock = lock;
    511  1.151.6.2       mjf 		mutex_obj_hold(lock);
    512  1.151.6.2       mjf 		mutex_enter(lock);
    513  1.151.6.2       mjf 	} else {
    514  1.151.6.2       mjf 		/* Lock assigned and taken during PRU_ATTACH. */
    515  1.151.6.2       mjf 	}
    516      1.140    dyoung 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
    517      1.140    dyoung 	    (struct mbuf *)(long)proto, NULL, l);
    518  1.151.6.2       mjf 	KASSERT(solocked(so));
    519      1.140    dyoung 	if (error != 0) {
    520        1.1       cgd 		so->so_state |= SS_NOFDREF;
    521        1.1       cgd 		sofree(so);
    522      1.140    dyoung 		return error;
    523        1.1       cgd 	}
    524  1.151.6.2       mjf 	sounlock(so);
    525        1.1       cgd 	*aso = so;
    526      1.140    dyoung 	return 0;
    527        1.1       cgd }
    528        1.1       cgd 
    529      1.142    dyoung /* On success, write file descriptor to fdout and return zero.  On
    530      1.142    dyoung  * failure, return non-zero; *fdout will be undefined.
    531      1.142    dyoung  */
    532      1.142    dyoung int
    533      1.142    dyoung fsocreate(int domain, struct socket **sop, int type, int protocol,
    534      1.142    dyoung     struct lwp *l, int *fdout)
    535      1.142    dyoung {
    536      1.142    dyoung 	struct socket	*so;
    537      1.142    dyoung 	struct file	*fp;
    538      1.142    dyoung 	int		fd, error;
    539      1.142    dyoung 
    540  1.151.6.1       mjf 	if ((error = fd_allocfile(&fp, &fd)) != 0)
    541      1.142    dyoung 		return (error);
    542      1.142    dyoung 	fp->f_flag = FREAD|FWRITE;
    543      1.142    dyoung 	fp->f_type = DTYPE_SOCKET;
    544      1.142    dyoung 	fp->f_ops = &socketops;
    545  1.151.6.2       mjf 	error = socreate(domain, &so, type, protocol, l, NULL);
    546      1.142    dyoung 	if (error != 0) {
    547  1.151.6.1       mjf 		fd_abort(curproc, fp, fd);
    548      1.142    dyoung 	} else {
    549      1.142    dyoung 		if (sop != NULL)
    550      1.142    dyoung 			*sop = so;
    551      1.142    dyoung 		fp->f_data = so;
    552  1.151.6.1       mjf 		fd_affix(curproc, fp, fd);
    553      1.142    dyoung 		*fdout = fd;
    554      1.142    dyoung 	}
    555      1.142    dyoung 	return error;
    556      1.142    dyoung }
    557      1.142    dyoung 
    558        1.3    andrew int
    559      1.114  christos sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
    560        1.1       cgd {
    561  1.151.6.2       mjf 	int	error;
    562        1.1       cgd 
    563  1.151.6.2       mjf 	solock(so);
    564      1.140    dyoung 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
    565  1.151.6.2       mjf 	sounlock(so);
    566      1.140    dyoung 	return error;
    567        1.1       cgd }
    568        1.1       cgd 
    569        1.3    andrew int
    570      1.150      elad solisten(struct socket *so, int backlog, struct lwp *l)
    571        1.1       cgd {
    572  1.151.6.2       mjf 	int	error;
    573        1.1       cgd 
    574  1.151.6.2       mjf 	solock(so);
    575  1.151.6.1       mjf 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
    576  1.151.6.2       mjf 	    SS_ISDISCONNECTING)) != 0) {
    577  1.151.6.2       mjf 	    	sounlock(so);
    578  1.151.6.1       mjf 		return (EOPNOTSUPP);
    579  1.151.6.2       mjf 	}
    580      1.140    dyoung 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
    581      1.150      elad 	    NULL, NULL, l);
    582      1.140    dyoung 	if (error != 0) {
    583  1.151.6.2       mjf 		sounlock(so);
    584      1.140    dyoung 		return error;
    585        1.1       cgd 	}
    586       1.63      matt 	if (TAILQ_EMPTY(&so->so_q))
    587        1.1       cgd 		so->so_options |= SO_ACCEPTCONN;
    588        1.1       cgd 	if (backlog < 0)
    589        1.1       cgd 		backlog = 0;
    590       1.49  jonathan 	so->so_qlimit = min(backlog, somaxconn);
    591  1.151.6.2       mjf 	sounlock(so);
    592      1.140    dyoung 	return 0;
    593        1.1       cgd }
    594        1.1       cgd 
    595       1.21  christos void
    596       1.54     lukem sofree(struct socket *so)
    597        1.1       cgd {
    598  1.151.6.2       mjf 	u_int refs;
    599  1.151.6.2       mjf 
    600  1.151.6.2       mjf 	KASSERT(solocked(so));
    601        1.1       cgd 
    602  1.151.6.2       mjf 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
    603  1.151.6.2       mjf 		sounlock(so);
    604        1.1       cgd 		return;
    605  1.151.6.2       mjf 	}
    606       1.43   mycroft 	if (so->so_head) {
    607       1.43   mycroft 		/*
    608       1.43   mycroft 		 * We must not decommission a socket that's on the accept(2)
    609       1.43   mycroft 		 * queue.  If we do, then accept(2) may hang after select(2)
    610       1.43   mycroft 		 * indicated that the listening socket was ready.
    611       1.43   mycroft 		 */
    612  1.151.6.2       mjf 		if (!soqremque(so, 0)) {
    613  1.151.6.2       mjf 			sounlock(so);
    614       1.43   mycroft 			return;
    615  1.151.6.2       mjf 		}
    616       1.43   mycroft 	}
    617       1.98  christos 	if (so->so_rcv.sb_hiwat)
    618      1.110  christos 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
    619       1.98  christos 		    RLIM_INFINITY);
    620       1.98  christos 	if (so->so_snd.sb_hiwat)
    621      1.110  christos 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
    622       1.98  christos 		    RLIM_INFINITY);
    623       1.98  christos 	sbrelease(&so->so_snd, so);
    624  1.151.6.2       mjf 	KASSERT(!cv_has_waiters(&so->so_cv));
    625  1.151.6.2       mjf 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
    626  1.151.6.2       mjf 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
    627        1.1       cgd 	sorflush(so);
    628  1.151.6.2       mjf 	refs = so->so_aborting;	/* XXX */
    629  1.151.6.5       mjf 	/* Remove acccept filter if one is present. */
    630  1.151.6.4       mjf 	if (so->so_accf != NULL)
    631  1.151.6.5       mjf 		(void)accept_filt_clear(so);
    632  1.151.6.2       mjf 	sounlock(so);
    633  1.151.6.2       mjf 	if (refs == 0)		/* XXX */
    634  1.151.6.2       mjf 		soput(so);
    635        1.1       cgd }
    636        1.1       cgd 
    637        1.1       cgd /*
    638        1.1       cgd  * Close a socket on last file table reference removal.
    639        1.1       cgd  * Initiate disconnect if connected.
    640        1.1       cgd  * Free socket when disconnect complete.
    641        1.1       cgd  */
    642        1.3    andrew int
    643       1.54     lukem soclose(struct socket *so)
    644        1.1       cgd {
    645       1.54     lukem 	struct socket	*so2;
    646  1.151.6.2       mjf 	int		error;
    647  1.151.6.2       mjf 	int		error2;
    648        1.1       cgd 
    649       1.54     lukem 	error = 0;
    650  1.151.6.2       mjf 	solock(so);
    651        1.1       cgd 	if (so->so_options & SO_ACCEPTCONN) {
    652  1.151.6.5       mjf 		for (;;) {
    653  1.151.6.5       mjf 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
    654  1.151.6.2       mjf 				KASSERT(solocked2(so, so2));
    655  1.151.6.2       mjf 				(void) soqremque(so2, 0);
    656  1.151.6.2       mjf 				/* soabort drops the lock. */
    657  1.151.6.2       mjf 				(void) soabort(so2);
    658  1.151.6.2       mjf 				solock(so);
    659  1.151.6.5       mjf 				continue;
    660  1.151.6.2       mjf 			}
    661  1.151.6.5       mjf 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
    662  1.151.6.2       mjf 				KASSERT(solocked2(so, so2));
    663  1.151.6.2       mjf 				(void) soqremque(so2, 1);
    664  1.151.6.2       mjf 				/* soabort drops the lock. */
    665  1.151.6.2       mjf 				(void) soabort(so2);
    666  1.151.6.2       mjf 				solock(so);
    667  1.151.6.5       mjf 				continue;
    668  1.151.6.2       mjf 			}
    669  1.151.6.5       mjf 			break;
    670  1.151.6.5       mjf 		}
    671        1.1       cgd 	}
    672        1.1       cgd 	if (so->so_pcb == 0)
    673        1.1       cgd 		goto discard;
    674        1.1       cgd 	if (so->so_state & SS_ISCONNECTED) {
    675        1.1       cgd 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
    676        1.1       cgd 			error = sodisconnect(so);
    677        1.1       cgd 			if (error)
    678        1.1       cgd 				goto drop;
    679        1.1       cgd 		}
    680        1.1       cgd 		if (so->so_options & SO_LINGER) {
    681      1.151        ad 			if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
    682        1.1       cgd 				goto drop;
    683       1.21  christos 			while (so->so_state & SS_ISCONNECTED) {
    684  1.151.6.2       mjf 				error = sowait(so, so->so_linger * hz);
    685       1.21  christos 				if (error)
    686        1.1       cgd 					break;
    687       1.21  christos 			}
    688        1.1       cgd 		}
    689        1.1       cgd 	}
    690       1.54     lukem  drop:
    691        1.1       cgd 	if (so->so_pcb) {
    692  1.151.6.2       mjf 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
    693      1.140    dyoung 		    NULL, NULL, NULL, NULL);
    694        1.1       cgd 		if (error == 0)
    695        1.1       cgd 			error = error2;
    696        1.1       cgd 	}
    697       1.54     lukem  discard:
    698        1.1       cgd 	if (so->so_state & SS_NOFDREF)
    699        1.1       cgd 		panic("soclose: NOFDREF");
    700        1.1       cgd 	so->so_state |= SS_NOFDREF;
    701        1.1       cgd 	sofree(so);
    702        1.1       cgd 	return (error);
    703        1.1       cgd }
    704        1.1       cgd 
    705        1.1       cgd /*
    706  1.151.6.2       mjf  * Must be called with the socket locked..  Will return with it unlocked.
    707        1.1       cgd  */
    708        1.3    andrew int
    709       1.54     lukem soabort(struct socket *so)
    710        1.1       cgd {
    711  1.151.6.2       mjf 	u_int refs;
    712      1.139      yamt 	int error;
    713  1.151.6.2       mjf 
    714  1.151.6.2       mjf 	KASSERT(solocked(so));
    715      1.139      yamt 	KASSERT(so->so_head == NULL);
    716  1.151.6.2       mjf 
    717  1.151.6.2       mjf 	so->so_aborting++;		/* XXX */
    718      1.140    dyoung 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
    719      1.140    dyoung 	    NULL, NULL, NULL);
    720  1.151.6.2       mjf 	refs = --so->so_aborting;	/* XXX */
    721  1.151.6.2       mjf 	if (error || (refs == 0)) {
    722      1.139      yamt 		sofree(so);
    723  1.151.6.2       mjf 	} else {
    724  1.151.6.2       mjf 		sounlock(so);
    725      1.139      yamt 	}
    726      1.139      yamt 	return error;
    727        1.1       cgd }
    728        1.1       cgd 
    729        1.3    andrew int
    730       1.54     lukem soaccept(struct socket *so, struct mbuf *nam)
    731        1.1       cgd {
    732  1.151.6.2       mjf 	int	error;
    733  1.151.6.2       mjf 
    734  1.151.6.2       mjf 	KASSERT(solocked(so));
    735        1.1       cgd 
    736       1.54     lukem 	error = 0;
    737        1.1       cgd 	if ((so->so_state & SS_NOFDREF) == 0)
    738        1.1       cgd 		panic("soaccept: !NOFDREF");
    739        1.1       cgd 	so->so_state &= ~SS_NOFDREF;
    740       1.55   thorpej 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
    741       1.55   thorpej 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
    742       1.41   mycroft 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
    743      1.140    dyoung 		    NULL, nam, NULL, NULL);
    744       1.41   mycroft 	else
    745       1.53    itojun 		error = ECONNABORTED;
    746       1.52    itojun 
    747        1.1       cgd 	return (error);
    748        1.1       cgd }
    749        1.1       cgd 
    750        1.3    andrew int
    751      1.114  christos soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
    752        1.1       cgd {
    753  1.151.6.2       mjf 	int		error;
    754  1.151.6.2       mjf 
    755  1.151.6.2       mjf 	KASSERT(solocked(so));
    756        1.1       cgd 
    757        1.1       cgd 	if (so->so_options & SO_ACCEPTCONN)
    758        1.1       cgd 		return (EOPNOTSUPP);
    759        1.1       cgd 	/*
    760        1.1       cgd 	 * If protocol is connection-based, can only connect once.
    761        1.1       cgd 	 * Otherwise, if connected, try to disconnect first.
    762        1.1       cgd 	 * This allows user to disconnect by connecting to, e.g.,
    763        1.1       cgd 	 * a null address.
    764        1.1       cgd 	 */
    765        1.1       cgd 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
    766        1.1       cgd 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
    767        1.1       cgd 	    (error = sodisconnect(so))))
    768        1.1       cgd 		error = EISCONN;
    769        1.1       cgd 	else
    770        1.1       cgd 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
    771      1.140    dyoung 		    NULL, nam, NULL, l);
    772        1.1       cgd 	return (error);
    773        1.1       cgd }
    774        1.1       cgd 
    775        1.3    andrew int
    776       1.54     lukem soconnect2(struct socket *so1, struct socket *so2)
    777        1.1       cgd {
    778  1.151.6.2       mjf 	int	error;
    779  1.151.6.2       mjf 
    780  1.151.6.2       mjf 	KASSERT(solocked2(so1, so2));
    781        1.1       cgd 
    782       1.22   mycroft 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
    783      1.140    dyoung 	    NULL, (struct mbuf *)so2, NULL, NULL);
    784        1.1       cgd 	return (error);
    785        1.1       cgd }
    786        1.1       cgd 
    787        1.3    andrew int
    788       1.54     lukem sodisconnect(struct socket *so)
    789        1.1       cgd {
    790  1.151.6.2       mjf 	int	error;
    791  1.151.6.2       mjf 
    792  1.151.6.2       mjf 	KASSERT(solocked(so));
    793        1.1       cgd 
    794        1.1       cgd 	if ((so->so_state & SS_ISCONNECTED) == 0) {
    795        1.1       cgd 		error = ENOTCONN;
    796  1.151.6.2       mjf 	} else if (so->so_state & SS_ISDISCONNECTING) {
    797        1.1       cgd 		error = EALREADY;
    798  1.151.6.2       mjf 	} else {
    799  1.151.6.2       mjf 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
    800  1.151.6.2       mjf 		    NULL, NULL, NULL, NULL);
    801        1.1       cgd 	}
    802      1.117      yamt 	sodopendfree();
    803        1.1       cgd 	return (error);
    804        1.1       cgd }
    805        1.1       cgd 
    806       1.15   mycroft #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
    807        1.1       cgd /*
    808        1.1       cgd  * Send on a socket.
    809        1.1       cgd  * If send must go all at once and message is larger than
    810        1.1       cgd  * send buffering, then hard error.
    811        1.1       cgd  * Lock against other senders.
    812        1.1       cgd  * If must go all at once and not enough room now, then
    813        1.1       cgd  * inform user that this would block and do nothing.
    814        1.1       cgd  * Otherwise, if nonblocking, send as much as possible.
    815        1.1       cgd  * The data to be sent is described by "uio" if nonzero,
    816        1.1       cgd  * otherwise by the mbuf chain "top" (which must be null
    817        1.1       cgd  * if uio is not).  Data provided in mbuf chain must be small
    818        1.1       cgd  * enough to send all at once.
    819        1.1       cgd  *
    820        1.1       cgd  * Returns nonzero on error, timeout or signal; callers
    821        1.1       cgd  * must check for short counts if EINTR/ERESTART are returned.
    822        1.1       cgd  * Data and control buffers are freed on return.
    823        1.1       cgd  */
    824        1.3    andrew int
    825       1.54     lukem sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
    826      1.114  christos 	struct mbuf *control, int flags, struct lwp *l)
    827        1.1       cgd {
    828       1.54     lukem 	struct mbuf	**mp, *m;
    829      1.114  christos 	struct proc	*p;
    830       1.58  jdolecek 	long		space, len, resid, clen, mlen;
    831       1.58  jdolecek 	int		error, s, dontroute, atomic;
    832       1.54     lukem 
    833      1.114  christos 	p = l->l_proc;
    834      1.117      yamt 	sodopendfree();
    835       1.54     lukem 	clen = 0;
    836  1.151.6.2       mjf 
    837  1.151.6.2       mjf 	/*
    838  1.151.6.2       mjf 	 * solock() provides atomicity of access.  splsoftnet() prevents
    839  1.151.6.2       mjf 	 * protocol processing soft interrupts from interrupting us and
    840  1.151.6.2       mjf 	 * blocking (expensive).
    841  1.151.6.2       mjf 	 */
    842  1.151.6.2       mjf 	s = splsoftnet();
    843  1.151.6.2       mjf 	solock(so);
    844       1.54     lukem 	atomic = sosendallatonce(so) || top;
    845        1.1       cgd 	if (uio)
    846        1.1       cgd 		resid = uio->uio_resid;
    847        1.1       cgd 	else
    848        1.1       cgd 		resid = top->m_pkthdr.len;
    849        1.7       cgd 	/*
    850        1.7       cgd 	 * In theory resid should be unsigned.
    851        1.7       cgd 	 * However, space must be signed, as it might be less than 0
    852        1.7       cgd 	 * if we over-committed, and we must use a signed comparison
    853        1.7       cgd 	 * of space and resid.  On the other hand, a negative resid
    854        1.7       cgd 	 * causes us to loop sending 0-length segments to the protocol.
    855        1.7       cgd 	 */
    856       1.29   mycroft 	if (resid < 0) {
    857       1.29   mycroft 		error = EINVAL;
    858       1.29   mycroft 		goto out;
    859       1.29   mycroft 	}
    860        1.1       cgd 	dontroute =
    861        1.1       cgd 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
    862        1.1       cgd 	    (so->so_proto->pr_flags & PR_ATOMIC);
    863  1.151.6.2       mjf 	l->l_ru.ru_msgsnd++;
    864        1.1       cgd 	if (control)
    865        1.1       cgd 		clen = control->m_len;
    866       1.54     lukem  restart:
    867       1.21  christos 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
    868        1.1       cgd 		goto out;
    869        1.1       cgd 	do {
    870  1.151.6.2       mjf 		if (so->so_state & SS_CANTSENDMORE) {
    871  1.151.6.2       mjf 			error = EPIPE;
    872  1.151.6.2       mjf 			goto release;
    873  1.151.6.2       mjf 		}
    874       1.48   thorpej 		if (so->so_error) {
    875       1.48   thorpej 			error = so->so_error;
    876       1.48   thorpej 			so->so_error = 0;
    877       1.48   thorpej 			goto release;
    878       1.48   thorpej 		}
    879        1.1       cgd 		if ((so->so_state & SS_ISCONNECTED) == 0) {
    880        1.1       cgd 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
    881        1.1       cgd 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
    882  1.151.6.2       mjf 				    !(resid == 0 && clen != 0)) {
    883  1.151.6.2       mjf 					error = ENOTCONN;
    884  1.151.6.2       mjf 					goto release;
    885  1.151.6.2       mjf 				}
    886  1.151.6.2       mjf 			} else if (addr == 0) {
    887  1.151.6.2       mjf 				error = EDESTADDRREQ;
    888  1.151.6.2       mjf 				goto release;
    889  1.151.6.2       mjf 			}
    890        1.1       cgd 		}
    891        1.1       cgd 		space = sbspace(&so->so_snd);
    892        1.1       cgd 		if (flags & MSG_OOB)
    893        1.1       cgd 			space += 1024;
    894       1.21  christos 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
    895  1.151.6.2       mjf 		    clen > so->so_snd.sb_hiwat) {
    896  1.151.6.2       mjf 			error = EMSGSIZE;
    897  1.151.6.2       mjf 			goto release;
    898  1.151.6.2       mjf 		}
    899       1.96   mycroft 		if (space < resid + clen &&
    900        1.1       cgd 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
    901  1.151.6.2       mjf 			if (so->so_nbio) {
    902  1.151.6.2       mjf 				error = EWOULDBLOCK;
    903  1.151.6.2       mjf 				goto release;
    904  1.151.6.2       mjf 			}
    905        1.1       cgd 			sbunlock(&so->so_snd);
    906        1.1       cgd 			error = sbwait(&so->so_snd);
    907        1.1       cgd 			if (error)
    908        1.1       cgd 				goto out;
    909        1.1       cgd 			goto restart;
    910        1.1       cgd 		}
    911        1.1       cgd 		mp = &top;
    912        1.1       cgd 		space -= clen;
    913        1.1       cgd 		do {
    914       1.45        tv 			if (uio == NULL) {
    915       1.45        tv 				/*
    916       1.45        tv 				 * Data is prepackaged in "top".
    917       1.45        tv 				 */
    918       1.45        tv 				resid = 0;
    919       1.45        tv 				if (flags & MSG_EOR)
    920       1.45        tv 					top->m_flags |= M_EOR;
    921       1.45        tv 			} else do {
    922  1.151.6.2       mjf 				sounlock(so);
    923  1.151.6.2       mjf 				splx(s);
    924      1.144    dyoung 				if (top == NULL) {
    925       1.78      matt 					m = m_gethdr(M_WAIT, MT_DATA);
    926       1.45        tv 					mlen = MHLEN;
    927       1.45        tv 					m->m_pkthdr.len = 0;
    928      1.140    dyoung 					m->m_pkthdr.rcvif = NULL;
    929       1.45        tv 				} else {
    930       1.78      matt 					m = m_get(M_WAIT, MT_DATA);
    931       1.45        tv 					mlen = MLEN;
    932       1.45        tv 				}
    933       1.78      matt 				MCLAIM(m, so->so_snd.sb_mowner);
    934      1.121      yamt 				if (sock_loan_thresh >= 0 &&
    935      1.121      yamt 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
    936      1.121      yamt 				    space >= sock_loan_thresh &&
    937       1.64   thorpej 				    (len = sosend_loan(so, uio, m,
    938       1.64   thorpej 						       space)) != 0) {
    939       1.64   thorpej 					SOSEND_COUNTER_INCR(&sosend_loan_big);
    940       1.64   thorpej 					space -= len;
    941       1.64   thorpej 					goto have_data;
    942       1.64   thorpej 				}
    943       1.45        tv 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
    944       1.64   thorpej 					SOSEND_COUNTER_INCR(&sosend_copy_big);
    945       1.78      matt 					m_clget(m, M_WAIT);
    946       1.45        tv 					if ((m->m_flags & M_EXT) == 0)
    947       1.45        tv 						goto nopages;
    948       1.45        tv 					mlen = MCLBYTES;
    949       1.45        tv 					if (atomic && top == 0) {
    950       1.58  jdolecek 						len = lmin(MCLBYTES - max_hdr,
    951       1.54     lukem 						    resid);
    952       1.45        tv 						m->m_data += max_hdr;
    953       1.45        tv 					} else
    954       1.58  jdolecek 						len = lmin(MCLBYTES, resid);
    955       1.45        tv 					space -= len;
    956       1.45        tv 				} else {
    957       1.64   thorpej  nopages:
    958       1.64   thorpej 					SOSEND_COUNTER_INCR(&sosend_copy_small);
    959       1.58  jdolecek 					len = lmin(lmin(mlen, resid), space);
    960       1.45        tv 					space -= len;
    961       1.45        tv 					/*
    962       1.45        tv 					 * For datagram protocols, leave room
    963       1.45        tv 					 * for protocol headers in first mbuf.
    964       1.45        tv 					 */
    965       1.45        tv 					if (atomic && top == 0 && len < mlen)
    966       1.45        tv 						MH_ALIGN(m, len);
    967       1.45        tv 				}
    968      1.144    dyoung 				error = uiomove(mtod(m, void *), (int)len, uio);
    969       1.64   thorpej  have_data:
    970       1.45        tv 				resid = uio->uio_resid;
    971       1.45        tv 				m->m_len = len;
    972       1.45        tv 				*mp = m;
    973       1.45        tv 				top->m_pkthdr.len += len;
    974  1.151.6.2       mjf 				s = splsoftnet();
    975  1.151.6.2       mjf 				solock(so);
    976      1.144    dyoung 				if (error != 0)
    977       1.45        tv 					goto release;
    978       1.45        tv 				mp = &m->m_next;
    979       1.45        tv 				if (resid <= 0) {
    980       1.45        tv 					if (flags & MSG_EOR)
    981       1.45        tv 						top->m_flags |= M_EOR;
    982       1.45        tv 					break;
    983       1.45        tv 				}
    984       1.45        tv 			} while (space > 0 && atomic);
    985      1.108     perry 
    986  1.151.6.2       mjf 			if (so->so_state & SS_CANTSENDMORE) {
    987  1.151.6.2       mjf 				error = EPIPE;
    988  1.151.6.2       mjf 				goto release;
    989  1.151.6.2       mjf 			}
    990       1.45        tv 			if (dontroute)
    991       1.45        tv 				so->so_options |= SO_DONTROUTE;
    992       1.45        tv 			if (resid > 0)
    993       1.45        tv 				so->so_state |= SS_MORETOCOME;
    994       1.46  sommerfe 			error = (*so->so_proto->pr_usrreq)(so,
    995       1.46  sommerfe 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
    996  1.151.6.2       mjf 			    top, addr, control, curlwp);
    997       1.45        tv 			if (dontroute)
    998       1.45        tv 				so->so_options &= ~SO_DONTROUTE;
    999       1.45        tv 			if (resid > 0)
   1000       1.45        tv 				so->so_state &= ~SS_MORETOCOME;
   1001       1.45        tv 			clen = 0;
   1002      1.144    dyoung 			control = NULL;
   1003      1.144    dyoung 			top = NULL;
   1004       1.45        tv 			mp = &top;
   1005      1.144    dyoung 			if (error != 0)
   1006        1.1       cgd 				goto release;
   1007        1.1       cgd 		} while (resid && space > 0);
   1008        1.1       cgd 	} while (resid);
   1009        1.1       cgd 
   1010       1.54     lukem  release:
   1011        1.1       cgd 	sbunlock(&so->so_snd);
   1012       1.54     lukem  out:
   1013  1.151.6.2       mjf 	sounlock(so);
   1014  1.151.6.2       mjf 	splx(s);
   1015        1.1       cgd 	if (top)
   1016        1.1       cgd 		m_freem(top);
   1017        1.1       cgd 	if (control)
   1018        1.1       cgd 		m_freem(control);
   1019        1.1       cgd 	return (error);
   1020        1.1       cgd }
   1021        1.1       cgd 
   1022        1.1       cgd /*
   1023  1.151.6.2       mjf  * Following replacement or removal of the first mbuf on the first
   1024  1.151.6.2       mjf  * mbuf chain of a socket buffer, push necessary state changes back
   1025  1.151.6.2       mjf  * into the socket buffer so that other consumers see the values
   1026  1.151.6.2       mjf  * consistently.  'nextrecord' is the callers locally stored value of
   1027  1.151.6.2       mjf  * the original value of sb->sb_mb->m_nextpkt which must be restored
   1028  1.151.6.2       mjf  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
   1029  1.151.6.2       mjf  */
   1030  1.151.6.2       mjf static void
   1031  1.151.6.2       mjf sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
   1032  1.151.6.2       mjf {
   1033  1.151.6.2       mjf 
   1034  1.151.6.2       mjf 	KASSERT(solocked(sb->sb_so));
   1035  1.151.6.2       mjf 
   1036  1.151.6.2       mjf 	/*
   1037  1.151.6.2       mjf 	 * First, update for the new value of nextrecord.  If necessary,
   1038  1.151.6.2       mjf 	 * make it the first record.
   1039  1.151.6.2       mjf 	 */
   1040  1.151.6.2       mjf 	if (sb->sb_mb != NULL)
   1041  1.151.6.2       mjf 		sb->sb_mb->m_nextpkt = nextrecord;
   1042  1.151.6.2       mjf 	else
   1043  1.151.6.2       mjf 		sb->sb_mb = nextrecord;
   1044  1.151.6.2       mjf 
   1045  1.151.6.2       mjf         /*
   1046  1.151.6.2       mjf          * Now update any dependent socket buffer fields to reflect
   1047  1.151.6.2       mjf          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
   1048  1.151.6.2       mjf          * the addition of a second clause that takes care of the
   1049  1.151.6.2       mjf          * case where sb_mb has been updated, but remains the last
   1050  1.151.6.2       mjf          * record.
   1051  1.151.6.2       mjf          */
   1052  1.151.6.2       mjf         if (sb->sb_mb == NULL) {
   1053  1.151.6.2       mjf                 sb->sb_mbtail = NULL;
   1054  1.151.6.2       mjf                 sb->sb_lastrecord = NULL;
   1055  1.151.6.2       mjf         } else if (sb->sb_mb->m_nextpkt == NULL)
   1056  1.151.6.2       mjf                 sb->sb_lastrecord = sb->sb_mb;
   1057  1.151.6.2       mjf }
   1058  1.151.6.2       mjf 
   1059  1.151.6.2       mjf /*
   1060        1.1       cgd  * Implement receive operations on a socket.
   1061        1.1       cgd  * We depend on the way that records are added to the sockbuf
   1062        1.1       cgd  * by sbappend*.  In particular, each record (mbufs linked through m_next)
   1063        1.1       cgd  * must begin with an address if the protocol so specifies,
   1064        1.1       cgd  * followed by an optional mbuf or mbufs containing ancillary data,
   1065        1.1       cgd  * and then zero or more mbufs of data.
   1066        1.1       cgd  * In order to avoid blocking network interrupts for the entire time here,
   1067        1.1       cgd  * we splx() while doing the actual copy to user space.
   1068        1.1       cgd  * Although the sockbuf is locked, new data may still be appended,
   1069        1.1       cgd  * and thus we must maintain consistency of the sockbuf during that time.
   1070        1.1       cgd  *
   1071        1.1       cgd  * The caller may receive the data as a single mbuf chain by supplying
   1072        1.1       cgd  * an mbuf **mp0 for use in returning the chain.  The uio is then used
   1073        1.1       cgd  * only for the count in uio_resid.
   1074        1.1       cgd  */
   1075        1.3    andrew int
   1076       1.54     lukem soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
   1077       1.54     lukem 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
   1078        1.1       cgd {
   1079      1.116      yamt 	struct lwp *l = curlwp;
   1080  1.151.6.2       mjf 	struct mbuf	*m, **mp, *mt;
   1081      1.146    dyoung 	int atomic, flags, len, error, s, offset, moff, type, orig_resid;
   1082       1.99      matt 	const struct protosw	*pr;
   1083       1.54     lukem 	struct mbuf	*nextrecord;
   1084       1.67        he 	int		mbuf_removed = 0;
   1085      1.146    dyoung 	const struct domain *dom;
   1086       1.64   thorpej 
   1087       1.54     lukem 	pr = so->so_proto;
   1088      1.146    dyoung 	atomic = pr->pr_flags & PR_ATOMIC;
   1089      1.146    dyoung 	dom = pr->pr_domain;
   1090        1.1       cgd 	mp = mp0;
   1091       1.54     lukem 	type = 0;
   1092       1.54     lukem 	orig_resid = uio->uio_resid;
   1093      1.102  jonathan 
   1094      1.144    dyoung 	if (paddr != NULL)
   1095      1.144    dyoung 		*paddr = NULL;
   1096      1.144    dyoung 	if (controlp != NULL)
   1097      1.144    dyoung 		*controlp = NULL;
   1098      1.144    dyoung 	if (flagsp != NULL)
   1099        1.1       cgd 		flags = *flagsp &~ MSG_EOR;
   1100        1.1       cgd 	else
   1101        1.1       cgd 		flags = 0;
   1102       1.66     enami 
   1103       1.66     enami 	if ((flags & MSG_DONTWAIT) == 0)
   1104      1.117      yamt 		sodopendfree();
   1105       1.66     enami 
   1106        1.1       cgd 	if (flags & MSG_OOB) {
   1107        1.1       cgd 		m = m_get(M_WAIT, MT_DATA);
   1108  1.151.6.2       mjf 		solock(so);
   1109       1.17       cgd 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
   1110      1.140    dyoung 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
   1111  1.151.6.2       mjf 		sounlock(so);
   1112        1.1       cgd 		if (error)
   1113        1.1       cgd 			goto bad;
   1114        1.1       cgd 		do {
   1115      1.134  christos 			error = uiomove(mtod(m, void *),
   1116        1.1       cgd 			    (int) min(uio->uio_resid, m->m_len), uio);
   1117        1.1       cgd 			m = m_free(m);
   1118      1.144    dyoung 		} while (uio->uio_resid > 0 && error == 0 && m);
   1119       1.54     lukem  bad:
   1120      1.144    dyoung 		if (m != NULL)
   1121        1.1       cgd 			m_freem(m);
   1122      1.144    dyoung 		return error;
   1123        1.1       cgd 	}
   1124      1.144    dyoung 	if (mp != NULL)
   1125      1.140    dyoung 		*mp = NULL;
   1126  1.151.6.2       mjf 
   1127  1.151.6.2       mjf 	/*
   1128  1.151.6.2       mjf 	 * solock() provides atomicity of access.  splsoftnet() prevents
   1129  1.151.6.2       mjf 	 * protocol processing soft interrupts from interrupting us and
   1130  1.151.6.2       mjf 	 * blocking (expensive).
   1131  1.151.6.2       mjf 	 */
   1132  1.151.6.2       mjf 	s = splsoftnet();
   1133  1.151.6.2       mjf 	solock(so);
   1134        1.1       cgd 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
   1135      1.140    dyoung 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
   1136        1.1       cgd 
   1137       1.54     lukem  restart:
   1138  1.151.6.2       mjf 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
   1139  1.151.6.2       mjf 		sounlock(so);
   1140  1.151.6.2       mjf 		splx(s);
   1141      1.144    dyoung 		return error;
   1142  1.151.6.2       mjf 	}
   1143        1.1       cgd 
   1144        1.1       cgd 	m = so->so_rcv.sb_mb;
   1145        1.1       cgd 	/*
   1146        1.1       cgd 	 * If we have less data than requested, block awaiting more
   1147        1.1       cgd 	 * (subject to any timeout) if:
   1148       1.15   mycroft 	 *   1. the current count is less than the low water mark,
   1149        1.1       cgd 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
   1150       1.15   mycroft 	 *	receive operation at once if we block (resid <= hiwat), or
   1151       1.15   mycroft 	 *   3. MSG_DONTWAIT is not set.
   1152        1.1       cgd 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
   1153        1.1       cgd 	 * we have to do the receive in sections, and thus risk returning
   1154        1.1       cgd 	 * a short count if a timeout or signal occurs after we start.
   1155        1.1       cgd 	 */
   1156      1.144    dyoung 	if (m == NULL ||
   1157      1.144    dyoung 	    ((flags & MSG_DONTWAIT) == 0 &&
   1158      1.144    dyoung 	     so->so_rcv.sb_cc < uio->uio_resid &&
   1159      1.144    dyoung 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
   1160      1.144    dyoung 	      ((flags & MSG_WAITALL) &&
   1161      1.144    dyoung 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
   1162      1.146    dyoung 	     m->m_nextpkt == NULL && !atomic)) {
   1163        1.1       cgd #ifdef DIAGNOSTIC
   1164      1.144    dyoung 		if (m == NULL && so->so_rcv.sb_cc)
   1165        1.1       cgd 			panic("receive 1");
   1166        1.1       cgd #endif
   1167        1.1       cgd 		if (so->so_error) {
   1168      1.144    dyoung 			if (m != NULL)
   1169       1.15   mycroft 				goto dontblock;
   1170        1.1       cgd 			error = so->so_error;
   1171        1.1       cgd 			if ((flags & MSG_PEEK) == 0)
   1172        1.1       cgd 				so->so_error = 0;
   1173        1.1       cgd 			goto release;
   1174        1.1       cgd 		}
   1175        1.1       cgd 		if (so->so_state & SS_CANTRCVMORE) {
   1176      1.144    dyoung 			if (m != NULL)
   1177       1.15   mycroft 				goto dontblock;
   1178        1.1       cgd 			else
   1179        1.1       cgd 				goto release;
   1180        1.1       cgd 		}
   1181      1.144    dyoung 		for (; m != NULL; m = m->m_next)
   1182        1.1       cgd 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
   1183        1.1       cgd 				m = so->so_rcv.sb_mb;
   1184        1.1       cgd 				goto dontblock;
   1185        1.1       cgd 			}
   1186        1.1       cgd 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
   1187        1.1       cgd 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
   1188        1.1       cgd 			error = ENOTCONN;
   1189        1.1       cgd 			goto release;
   1190        1.1       cgd 		}
   1191        1.1       cgd 		if (uio->uio_resid == 0)
   1192        1.1       cgd 			goto release;
   1193      1.151        ad 		if (so->so_nbio || (flags & MSG_DONTWAIT)) {
   1194        1.1       cgd 			error = EWOULDBLOCK;
   1195        1.1       cgd 			goto release;
   1196        1.1       cgd 		}
   1197       1.69   thorpej 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
   1198       1.69   thorpej 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
   1199        1.1       cgd 		sbunlock(&so->so_rcv);
   1200        1.1       cgd 		error = sbwait(&so->so_rcv);
   1201  1.151.6.2       mjf 		if (error != 0) {
   1202  1.151.6.2       mjf 			sounlock(so);
   1203  1.151.6.2       mjf 			splx(s);
   1204      1.144    dyoung 			return error;
   1205  1.151.6.2       mjf 		}
   1206        1.1       cgd 		goto restart;
   1207        1.1       cgd 	}
   1208       1.54     lukem  dontblock:
   1209       1.69   thorpej 	/*
   1210       1.69   thorpej 	 * On entry here, m points to the first record of the socket buffer.
   1211  1.151.6.2       mjf 	 * From this point onward, we maintain 'nextrecord' as a cache of the
   1212  1.151.6.2       mjf 	 * pointer to the next record in the socket buffer.  We must keep the
   1213  1.151.6.2       mjf 	 * various socket buffer pointers and local stack versions of the
   1214  1.151.6.2       mjf 	 * pointers in sync, pushing out modifications before dropping the
   1215  1.151.6.2       mjf 	 * socket lock, and re-reading them when picking it up.
   1216  1.151.6.2       mjf 	 *
   1217  1.151.6.2       mjf 	 * Otherwise, we will race with the network stack appending new data
   1218  1.151.6.2       mjf 	 * or records onto the socket buffer by using inconsistent/stale
   1219  1.151.6.2       mjf 	 * versions of the field, possibly resulting in socket buffer
   1220  1.151.6.2       mjf 	 * corruption.
   1221  1.151.6.2       mjf 	 *
   1222  1.151.6.2       mjf 	 * By holding the high-level sblock(), we prevent simultaneous
   1223  1.151.6.2       mjf 	 * readers from pulling off the front of the socket buffer.
   1224       1.69   thorpej 	 */
   1225      1.144    dyoung 	if (l != NULL)
   1226  1.151.6.1       mjf 		l->l_ru.ru_msgrcv++;
   1227       1.69   thorpej 	KASSERT(m == so->so_rcv.sb_mb);
   1228       1.69   thorpej 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
   1229       1.69   thorpej 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
   1230        1.1       cgd 	nextrecord = m->m_nextpkt;
   1231        1.1       cgd 	if (pr->pr_flags & PR_ADDR) {
   1232        1.1       cgd #ifdef DIAGNOSTIC
   1233        1.1       cgd 		if (m->m_type != MT_SONAME)
   1234        1.1       cgd 			panic("receive 1a");
   1235        1.1       cgd #endif
   1236        1.3    andrew 		orig_resid = 0;
   1237        1.1       cgd 		if (flags & MSG_PEEK) {
   1238        1.1       cgd 			if (paddr)
   1239        1.1       cgd 				*paddr = m_copy(m, 0, m->m_len);
   1240        1.1       cgd 			m = m->m_next;
   1241        1.1       cgd 		} else {
   1242        1.1       cgd 			sbfree(&so->so_rcv, m);
   1243       1.67        he 			mbuf_removed = 1;
   1244      1.144    dyoung 			if (paddr != NULL) {
   1245        1.1       cgd 				*paddr = m;
   1246        1.1       cgd 				so->so_rcv.sb_mb = m->m_next;
   1247      1.144    dyoung 				m->m_next = NULL;
   1248        1.1       cgd 				m = so->so_rcv.sb_mb;
   1249        1.1       cgd 			} else {
   1250        1.1       cgd 				MFREE(m, so->so_rcv.sb_mb);
   1251        1.1       cgd 				m = so->so_rcv.sb_mb;
   1252        1.1       cgd 			}
   1253  1.151.6.2       mjf 			sbsync(&so->so_rcv, nextrecord);
   1254        1.1       cgd 		}
   1255        1.1       cgd 	}
   1256  1.151.6.2       mjf 
   1257  1.151.6.2       mjf 	/*
   1258  1.151.6.2       mjf 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
   1259  1.151.6.2       mjf 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
   1260  1.151.6.2       mjf 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
   1261  1.151.6.2       mjf 	 * perform externalization (or freeing if controlp == NULL).
   1262  1.151.6.2       mjf 	 */
   1263  1.151.6.2       mjf 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
   1264  1.151.6.2       mjf 		struct mbuf *cm = NULL, *cmn;
   1265  1.151.6.2       mjf 		struct mbuf **cme = &cm;
   1266  1.151.6.2       mjf 
   1267  1.151.6.2       mjf 		do {
   1268  1.151.6.2       mjf 			if (flags & MSG_PEEK) {
   1269  1.151.6.2       mjf 				if (controlp != NULL) {
   1270  1.151.6.2       mjf 					*controlp = m_copy(m, 0, m->m_len);
   1271  1.151.6.2       mjf 					controlp = &(*controlp)->m_next;
   1272  1.151.6.2       mjf 				}
   1273  1.151.6.2       mjf 				m = m->m_next;
   1274  1.151.6.2       mjf 			} else {
   1275  1.151.6.2       mjf 				sbfree(&so->so_rcv, m);
   1276        1.1       cgd 				so->so_rcv.sb_mb = m->m_next;
   1277      1.144    dyoung 				m->m_next = NULL;
   1278  1.151.6.2       mjf 				*cme = m;
   1279  1.151.6.2       mjf 				cme = &(*cme)->m_next;
   1280        1.1       cgd 				m = so->so_rcv.sb_mb;
   1281  1.151.6.2       mjf 			}
   1282  1.151.6.2       mjf 		} while (m != NULL && m->m_type == MT_CONTROL);
   1283  1.151.6.2       mjf 		if ((flags & MSG_PEEK) == 0)
   1284  1.151.6.2       mjf 			sbsync(&so->so_rcv, nextrecord);
   1285  1.151.6.2       mjf 		for (; cm != NULL; cm = cmn) {
   1286  1.151.6.2       mjf 			cmn = cm->m_next;
   1287  1.151.6.2       mjf 			cm->m_next = NULL;
   1288  1.151.6.2       mjf 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
   1289  1.151.6.2       mjf 			if (controlp != NULL) {
   1290  1.151.6.2       mjf 				if (dom->dom_externalize != NULL &&
   1291  1.151.6.2       mjf 				    type == SCM_RIGHTS) {
   1292  1.151.6.2       mjf 					sounlock(so);
   1293  1.151.6.2       mjf 					splx(s);
   1294  1.151.6.2       mjf 					error = (*dom->dom_externalize)(cm, l);
   1295  1.151.6.2       mjf 					s = splsoftnet();
   1296  1.151.6.2       mjf 					solock(so);
   1297  1.151.6.2       mjf 				}
   1298  1.151.6.2       mjf 				*controlp = cm;
   1299  1.151.6.2       mjf 				while (*controlp != NULL)
   1300  1.151.6.2       mjf 					controlp = &(*controlp)->m_next;
   1301        1.1       cgd 			} else {
   1302      1.106    itojun 				/*
   1303      1.106    itojun 				 * Dispose of any SCM_RIGHTS message that went
   1304      1.106    itojun 				 * through the read path rather than recv.
   1305      1.106    itojun 				 */
   1306  1.151.6.2       mjf 				if (dom->dom_dispose != NULL &&
   1307  1.151.6.2       mjf 				    type == SCM_RIGHTS) {
   1308  1.151.6.2       mjf 				    	sounlock(so);
   1309  1.151.6.2       mjf 					(*dom->dom_dispose)(cm);
   1310  1.151.6.2       mjf 					solock(so);
   1311  1.151.6.2       mjf 				}
   1312  1.151.6.2       mjf 				m_freem(cm);
   1313        1.1       cgd 			}
   1314        1.1       cgd 		}
   1315  1.151.6.2       mjf 		if (m != NULL)
   1316  1.151.6.2       mjf 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
   1317  1.151.6.2       mjf 		else
   1318  1.151.6.2       mjf 			nextrecord = so->so_rcv.sb_mb;
   1319  1.151.6.2       mjf 		orig_resid = 0;
   1320        1.1       cgd 	}
   1321       1.69   thorpej 
   1322  1.151.6.2       mjf 	/* If m is non-NULL, we have some data to read. */
   1323  1.151.6.2       mjf 	if (__predict_true(m != NULL)) {
   1324        1.1       cgd 		type = m->m_type;
   1325        1.1       cgd 		if (type == MT_OOBDATA)
   1326        1.1       cgd 			flags |= MSG_OOB;
   1327        1.1       cgd 	}
   1328       1.69   thorpej 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
   1329       1.69   thorpej 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
   1330       1.69   thorpej 
   1331        1.1       cgd 	moff = 0;
   1332        1.1       cgd 	offset = 0;
   1333      1.144    dyoung 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
   1334        1.1       cgd 		if (m->m_type == MT_OOBDATA) {
   1335        1.1       cgd 			if (type != MT_OOBDATA)
   1336        1.1       cgd 				break;
   1337        1.1       cgd 		} else if (type == MT_OOBDATA)
   1338        1.1       cgd 			break;
   1339        1.1       cgd #ifdef DIAGNOSTIC
   1340        1.1       cgd 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
   1341        1.1       cgd 			panic("receive 3");
   1342        1.1       cgd #endif
   1343        1.1       cgd 		so->so_state &= ~SS_RCVATMARK;
   1344        1.1       cgd 		len = uio->uio_resid;
   1345        1.1       cgd 		if (so->so_oobmark && len > so->so_oobmark - offset)
   1346        1.1       cgd 			len = so->so_oobmark - offset;
   1347        1.1       cgd 		if (len > m->m_len - moff)
   1348        1.1       cgd 			len = m->m_len - moff;
   1349        1.1       cgd 		/*
   1350        1.1       cgd 		 * If mp is set, just pass back the mbufs.
   1351        1.1       cgd 		 * Otherwise copy them out via the uio, then free.
   1352        1.1       cgd 		 * Sockbuf must be consistent here (points to current mbuf,
   1353        1.1       cgd 		 * it points to next record) when we drop priority;
   1354        1.1       cgd 		 * we must note any additions to the sockbuf when we
   1355        1.1       cgd 		 * block interrupts again.
   1356        1.1       cgd 		 */
   1357      1.144    dyoung 		if (mp == NULL) {
   1358       1.69   thorpej 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
   1359       1.69   thorpej 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
   1360  1.151.6.2       mjf 			sounlock(so);
   1361        1.1       cgd 			splx(s);
   1362      1.134  christos 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
   1363       1.20   mycroft 			s = splsoftnet();
   1364  1.151.6.2       mjf 			solock(so);
   1365      1.144    dyoung 			if (error != 0) {
   1366       1.67        he 				/*
   1367       1.67        he 				 * If any part of the record has been removed
   1368       1.67        he 				 * (such as the MT_SONAME mbuf, which will
   1369       1.67        he 				 * happen when PR_ADDR, and thus also
   1370       1.67        he 				 * PR_ATOMIC, is set), then drop the entire
   1371       1.67        he 				 * record to maintain the atomicity of the
   1372       1.67        he 				 * receive operation.
   1373       1.67        he 				 *
   1374       1.67        he 				 * This avoids a later panic("receive 1a")
   1375       1.67        he 				 * when compiled with DIAGNOSTIC.
   1376       1.67        he 				 */
   1377      1.146    dyoung 				if (m && mbuf_removed && atomic)
   1378       1.67        he 					(void) sbdroprecord(&so->so_rcv);
   1379       1.67        he 
   1380       1.57  jdolecek 				goto release;
   1381       1.67        he 			}
   1382        1.1       cgd 		} else
   1383        1.1       cgd 			uio->uio_resid -= len;
   1384        1.1       cgd 		if (len == m->m_len - moff) {
   1385        1.1       cgd 			if (m->m_flags & M_EOR)
   1386        1.1       cgd 				flags |= MSG_EOR;
   1387        1.1       cgd 			if (flags & MSG_PEEK) {
   1388        1.1       cgd 				m = m->m_next;
   1389        1.1       cgd 				moff = 0;
   1390        1.1       cgd 			} else {
   1391        1.1       cgd 				nextrecord = m->m_nextpkt;
   1392        1.1       cgd 				sbfree(&so->so_rcv, m);
   1393        1.1       cgd 				if (mp) {
   1394        1.1       cgd 					*mp = m;
   1395        1.1       cgd 					mp = &m->m_next;
   1396        1.1       cgd 					so->so_rcv.sb_mb = m = m->m_next;
   1397      1.140    dyoung 					*mp = NULL;
   1398        1.1       cgd 				} else {
   1399        1.1       cgd 					MFREE(m, so->so_rcv.sb_mb);
   1400        1.1       cgd 					m = so->so_rcv.sb_mb;
   1401        1.1       cgd 				}
   1402       1.69   thorpej 				/*
   1403       1.69   thorpej 				 * If m != NULL, we also know that
   1404       1.69   thorpej 				 * so->so_rcv.sb_mb != NULL.
   1405       1.69   thorpej 				 */
   1406       1.69   thorpej 				KASSERT(so->so_rcv.sb_mb == m);
   1407       1.69   thorpej 				if (m) {
   1408        1.1       cgd 					m->m_nextpkt = nextrecord;
   1409       1.69   thorpej 					if (nextrecord == NULL)
   1410       1.69   thorpej 						so->so_rcv.sb_lastrecord = m;
   1411       1.69   thorpej 				} else {
   1412       1.69   thorpej 					so->so_rcv.sb_mb = nextrecord;
   1413       1.70   thorpej 					SB_EMPTY_FIXUP(&so->so_rcv);
   1414       1.69   thorpej 				}
   1415       1.69   thorpej 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
   1416       1.69   thorpej 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
   1417        1.1       cgd 			}
   1418      1.144    dyoung 		} else if (flags & MSG_PEEK)
   1419      1.144    dyoung 			moff += len;
   1420      1.144    dyoung 		else {
   1421  1.151.6.2       mjf 			if (mp != NULL) {
   1422  1.151.6.2       mjf 				mt = m_copym(m, 0, len, M_NOWAIT);
   1423  1.151.6.2       mjf 				if (__predict_false(mt == NULL)) {
   1424  1.151.6.2       mjf 					sounlock(so);
   1425  1.151.6.2       mjf 					mt = m_copym(m, 0, len, M_WAIT);
   1426  1.151.6.2       mjf 					solock(so);
   1427  1.151.6.2       mjf 				}
   1428  1.151.6.2       mjf 				*mp = mt;
   1429  1.151.6.2       mjf 			}
   1430      1.144    dyoung 			m->m_data += len;
   1431      1.144    dyoung 			m->m_len -= len;
   1432      1.144    dyoung 			so->so_rcv.sb_cc -= len;
   1433        1.1       cgd 		}
   1434        1.1       cgd 		if (so->so_oobmark) {
   1435        1.1       cgd 			if ((flags & MSG_PEEK) == 0) {
   1436        1.1       cgd 				so->so_oobmark -= len;
   1437        1.1       cgd 				if (so->so_oobmark == 0) {
   1438        1.1       cgd 					so->so_state |= SS_RCVATMARK;
   1439        1.1       cgd 					break;
   1440        1.1       cgd 				}
   1441        1.7       cgd 			} else {
   1442        1.1       cgd 				offset += len;
   1443        1.7       cgd 				if (offset == so->so_oobmark)
   1444        1.7       cgd 					break;
   1445        1.7       cgd 			}
   1446        1.1       cgd 		}
   1447        1.1       cgd 		if (flags & MSG_EOR)
   1448        1.1       cgd 			break;
   1449        1.1       cgd 		/*
   1450        1.1       cgd 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
   1451        1.1       cgd 		 * we must not quit until "uio->uio_resid == 0" or an error
   1452        1.1       cgd 		 * termination.  If a signal/timeout occurs, return
   1453        1.1       cgd 		 * with a short count but without error.
   1454        1.1       cgd 		 * Keep sockbuf locked against other readers.
   1455        1.1       cgd 		 */
   1456      1.144    dyoung 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
   1457        1.3    andrew 		    !sosendallatonce(so) && !nextrecord) {
   1458        1.1       cgd 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
   1459        1.1       cgd 				break;
   1460       1.68      matt 			/*
   1461       1.68      matt 			 * If we are peeking and the socket receive buffer is
   1462       1.68      matt 			 * full, stop since we can't get more data to peek at.
   1463       1.68      matt 			 */
   1464       1.68      matt 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
   1465       1.68      matt 				break;
   1466       1.68      matt 			/*
   1467       1.68      matt 			 * If we've drained the socket buffer, tell the
   1468       1.68      matt 			 * protocol in case it needs to do something to
   1469       1.68      matt 			 * get it filled again.
   1470       1.68      matt 			 */
   1471       1.68      matt 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
   1472       1.68      matt 				(*pr->pr_usrreq)(so, PRU_RCVD,
   1473      1.140    dyoung 				    NULL, (struct mbuf *)(long)flags, NULL, l);
   1474       1.69   thorpej 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
   1475       1.69   thorpej 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
   1476        1.1       cgd 			error = sbwait(&so->so_rcv);
   1477      1.144    dyoung 			if (error != 0) {
   1478        1.1       cgd 				sbunlock(&so->so_rcv);
   1479  1.151.6.2       mjf 				sounlock(so);
   1480        1.1       cgd 				splx(s);
   1481      1.144    dyoung 				return 0;
   1482        1.1       cgd 			}
   1483       1.21  christos 			if ((m = so->so_rcv.sb_mb) != NULL)
   1484        1.1       cgd 				nextrecord = m->m_nextpkt;
   1485        1.1       cgd 		}
   1486        1.1       cgd 	}
   1487        1.3    andrew 
   1488      1.146    dyoung 	if (m && atomic) {
   1489        1.3    andrew 		flags |= MSG_TRUNC;
   1490        1.3    andrew 		if ((flags & MSG_PEEK) == 0)
   1491        1.3    andrew 			(void) sbdroprecord(&so->so_rcv);
   1492        1.3    andrew 	}
   1493        1.1       cgd 	if ((flags & MSG_PEEK) == 0) {
   1494      1.144    dyoung 		if (m == NULL) {
   1495       1.69   thorpej 			/*
   1496       1.70   thorpej 			 * First part is an inline SB_EMPTY_FIXUP().  Second
   1497       1.69   thorpej 			 * part makes sure sb_lastrecord is up-to-date if
   1498       1.69   thorpej 			 * there is still data in the socket buffer.
   1499       1.69   thorpej 			 */
   1500        1.1       cgd 			so->so_rcv.sb_mb = nextrecord;
   1501       1.69   thorpej 			if (so->so_rcv.sb_mb == NULL) {
   1502       1.69   thorpej 				so->so_rcv.sb_mbtail = NULL;
   1503       1.69   thorpej 				so->so_rcv.sb_lastrecord = NULL;
   1504       1.69   thorpej 			} else if (nextrecord->m_nextpkt == NULL)
   1505       1.69   thorpej 				so->so_rcv.sb_lastrecord = nextrecord;
   1506       1.69   thorpej 		}
   1507       1.69   thorpej 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
   1508       1.69   thorpej 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
   1509        1.1       cgd 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
   1510      1.140    dyoung 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
   1511      1.140    dyoung 			    (struct mbuf *)(long)flags, NULL, l);
   1512        1.1       cgd 	}
   1513        1.3    andrew 	if (orig_resid == uio->uio_resid && orig_resid &&
   1514        1.3    andrew 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
   1515        1.3    andrew 		sbunlock(&so->so_rcv);
   1516        1.3    andrew 		goto restart;
   1517        1.3    andrew 	}
   1518      1.108     perry 
   1519      1.144    dyoung 	if (flagsp != NULL)
   1520        1.1       cgd 		*flagsp |= flags;
   1521       1.54     lukem  release:
   1522        1.1       cgd 	sbunlock(&so->so_rcv);
   1523  1.151.6.2       mjf 	sounlock(so);
   1524        1.1       cgd 	splx(s);
   1525      1.144    dyoung 	return error;
   1526        1.1       cgd }
   1527        1.1       cgd 
   1528       1.14   mycroft int
   1529       1.54     lukem soshutdown(struct socket *so, int how)
   1530        1.1       cgd {
   1531       1.99      matt 	const struct protosw	*pr;
   1532  1.151.6.2       mjf 	int	error;
   1533  1.151.6.2       mjf 
   1534  1.151.6.2       mjf 	KASSERT(solocked(so));
   1535       1.34    kleink 
   1536       1.54     lukem 	pr = so->so_proto;
   1537       1.34    kleink 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
   1538       1.34    kleink 		return (EINVAL);
   1539        1.1       cgd 
   1540  1.151.6.2       mjf 	if (how == SHUT_RD || how == SHUT_RDWR) {
   1541        1.1       cgd 		sorflush(so);
   1542  1.151.6.2       mjf 		error = 0;
   1543  1.151.6.2       mjf 	}
   1544       1.34    kleink 	if (how == SHUT_WR || how == SHUT_RDWR)
   1545  1.151.6.2       mjf 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
   1546      1.140    dyoung 		    NULL, NULL, NULL);
   1547  1.151.6.2       mjf 
   1548  1.151.6.2       mjf 	return error;
   1549        1.1       cgd }
   1550        1.1       cgd 
   1551       1.14   mycroft void
   1552       1.54     lukem sorflush(struct socket *so)
   1553        1.1       cgd {
   1554       1.54     lukem 	struct sockbuf	*sb, asb;
   1555       1.99      matt 	const struct protosw	*pr;
   1556  1.151.6.2       mjf 
   1557  1.151.6.2       mjf 	KASSERT(solocked(so));
   1558        1.1       cgd 
   1559       1.54     lukem 	sb = &so->so_rcv;
   1560       1.54     lukem 	pr = so->so_proto;
   1561        1.1       cgd 	socantrcvmore(so);
   1562  1.151.6.2       mjf 	sb->sb_flags |= SB_NOINTR;
   1563  1.151.6.2       mjf 	(void )sblock(sb, M_WAITOK);
   1564        1.1       cgd 	sbunlock(sb);
   1565        1.1       cgd 	asb = *sb;
   1566       1.86  wrstuden 	/*
   1567       1.86  wrstuden 	 * Clear most of the sockbuf structure, but leave some of the
   1568       1.86  wrstuden 	 * fields valid.
   1569       1.86  wrstuden 	 */
   1570       1.86  wrstuden 	memset(&sb->sb_startzero, 0,
   1571       1.86  wrstuden 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
   1572  1.151.6.2       mjf 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
   1573  1.151.6.2       mjf 		sounlock(so);
   1574        1.1       cgd 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
   1575  1.151.6.2       mjf 		solock(so);
   1576  1.151.6.2       mjf 	}
   1577       1.98  christos 	sbrelease(&asb, so);
   1578        1.1       cgd }
   1579        1.1       cgd 
   1580  1.151.6.4       mjf /*
   1581  1.151.6.4       mjf  * internal set SOL_SOCKET options
   1582  1.151.6.4       mjf  */
   1583      1.142    dyoung static int
   1584  1.151.6.4       mjf sosetopt1(struct socket *so, const struct sockopt *sopt)
   1585        1.1       cgd {
   1586  1.151.6.5       mjf 	int error = EINVAL, optval, opt;
   1587  1.151.6.4       mjf 	struct linger l;
   1588  1.151.6.4       mjf 	struct timeval tv;
   1589      1.142    dyoung 
   1590  1.151.6.5       mjf 	switch ((opt = sopt->sopt_name)) {
   1591      1.142    dyoung 
   1592  1.151.6.4       mjf 	case SO_ACCEPTFILTER:
   1593  1.151.6.5       mjf 		error = accept_filt_setopt(so, sopt);
   1594  1.151.6.5       mjf 		KASSERT(solocked(so));
   1595      1.142    dyoung 		break;
   1596  1.151.6.4       mjf 
   1597  1.151.6.4       mjf   	case SO_LINGER:
   1598  1.151.6.4       mjf  		error = sockopt_get(sopt, &l, sizeof(l));
   1599  1.151.6.5       mjf 		solock(so);
   1600  1.151.6.4       mjf  		if (error)
   1601  1.151.6.5       mjf  			break;
   1602  1.151.6.4       mjf  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
   1603  1.151.6.5       mjf  		    l.l_linger > (INT_MAX / hz)) {
   1604  1.151.6.5       mjf 			error = EDOM;
   1605  1.151.6.5       mjf 			break;
   1606  1.151.6.5       mjf 		}
   1607  1.151.6.4       mjf  		so->so_linger = l.l_linger;
   1608  1.151.6.4       mjf  		if (l.l_onoff)
   1609  1.151.6.4       mjf  			so->so_options |= SO_LINGER;
   1610  1.151.6.4       mjf  		else
   1611  1.151.6.4       mjf  			so->so_options &= ~SO_LINGER;
   1612  1.151.6.5       mjf    		break;
   1613        1.1       cgd 
   1614      1.142    dyoung 	case SO_DEBUG:
   1615      1.142    dyoung 	case SO_KEEPALIVE:
   1616      1.142    dyoung 	case SO_DONTROUTE:
   1617      1.142    dyoung 	case SO_USELOOPBACK:
   1618      1.142    dyoung 	case SO_BROADCAST:
   1619      1.142    dyoung 	case SO_REUSEADDR:
   1620      1.142    dyoung 	case SO_REUSEPORT:
   1621      1.142    dyoung 	case SO_OOBINLINE:
   1622      1.142    dyoung 	case SO_TIMESTAMP:
   1623  1.151.6.4       mjf 		error = sockopt_getint(sopt, &optval);
   1624  1.151.6.5       mjf 		solock(so);
   1625  1.151.6.4       mjf 		if (error)
   1626  1.151.6.5       mjf 			break;
   1627  1.151.6.4       mjf 		if (optval)
   1628  1.151.6.5       mjf 			so->so_options |= opt;
   1629      1.142    dyoung 		else
   1630  1.151.6.5       mjf 			so->so_options &= ~opt;
   1631      1.142    dyoung 		break;
   1632      1.142    dyoung 
   1633      1.142    dyoung 	case SO_SNDBUF:
   1634      1.142    dyoung 	case SO_RCVBUF:
   1635      1.142    dyoung 	case SO_SNDLOWAT:
   1636      1.142    dyoung 	case SO_RCVLOWAT:
   1637  1.151.6.4       mjf 		error = sockopt_getint(sopt, &optval);
   1638  1.151.6.5       mjf 		solock(so);
   1639  1.151.6.4       mjf 		if (error)
   1640  1.151.6.5       mjf 			break;
   1641        1.1       cgd 
   1642      1.142    dyoung 		/*
   1643      1.142    dyoung 		 * Values < 1 make no sense for any of these
   1644      1.142    dyoung 		 * options, so disallow them.
   1645      1.142    dyoung 		 */
   1646  1.151.6.5       mjf 		if (optval < 1) {
   1647  1.151.6.5       mjf 			error = EINVAL;
   1648  1.151.6.5       mjf 			break;
   1649  1.151.6.5       mjf 		}
   1650        1.1       cgd 
   1651  1.151.6.5       mjf 		switch (opt) {
   1652        1.1       cgd 		case SO_SNDBUF:
   1653  1.151.6.5       mjf 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
   1654  1.151.6.5       mjf 				error = ENOBUFS;
   1655  1.151.6.5       mjf 				break;
   1656  1.151.6.5       mjf 			}
   1657  1.151.6.4       mjf 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
   1658  1.151.6.4       mjf 			break;
   1659  1.151.6.4       mjf 
   1660        1.1       cgd 		case SO_RCVBUF:
   1661  1.151.6.5       mjf 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
   1662  1.151.6.5       mjf 				error = ENOBUFS;
   1663  1.151.6.5       mjf 				break;
   1664  1.151.6.5       mjf 			}
   1665  1.151.6.4       mjf 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
   1666      1.142    dyoung 			break;
   1667      1.142    dyoung 
   1668      1.142    dyoung 		/*
   1669      1.142    dyoung 		 * Make sure the low-water is never greater than
   1670      1.142    dyoung 		 * the high-water.
   1671      1.142    dyoung 		 */
   1672        1.1       cgd 		case SO_SNDLOWAT:
   1673  1.151.6.4       mjf 			if (optval > so->so_snd.sb_hiwat)
   1674  1.151.6.4       mjf 				optval = so->so_snd.sb_hiwat;
   1675  1.151.6.4       mjf 
   1676  1.151.6.4       mjf 			so->so_snd.sb_lowat = optval;
   1677      1.142    dyoung 			break;
   1678  1.151.6.4       mjf 
   1679        1.1       cgd 		case SO_RCVLOWAT:
   1680  1.151.6.4       mjf 			if (optval > so->so_rcv.sb_hiwat)
   1681  1.151.6.4       mjf 				optval = so->so_rcv.sb_hiwat;
   1682  1.151.6.4       mjf 
   1683  1.151.6.4       mjf 			so->so_rcv.sb_lowat = optval;
   1684      1.142    dyoung 			break;
   1685      1.142    dyoung 		}
   1686      1.142    dyoung 		break;
   1687       1.28   thorpej 
   1688  1.151.6.5       mjf #ifdef COMPAT_50
   1689  1.151.6.5       mjf 	case SO_OSNDTIMEO:
   1690  1.151.6.5       mjf 	case SO_ORCVTIMEO: {
   1691  1.151.6.5       mjf 		struct timeval50 otv;
   1692  1.151.6.5       mjf 		error = sockopt_get(sopt, &otv, sizeof(otv));
   1693  1.151.6.5       mjf 		if (error)
   1694  1.151.6.5       mjf 			break;
   1695  1.151.6.5       mjf 		timeval50_to_timeval(&otv, &tv);
   1696  1.151.6.5       mjf 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
   1697  1.151.6.5       mjf 		error = 0;
   1698  1.151.6.5       mjf 		/*FALLTHROUGH*/
   1699  1.151.6.5       mjf 	}
   1700  1.151.6.5       mjf #endif /* COMPAT_50 */
   1701  1.151.6.5       mjf 
   1702      1.142    dyoung 	case SO_SNDTIMEO:
   1703      1.142    dyoung 	case SO_RCVTIMEO:
   1704  1.151.6.4       mjf 		if (error)
   1705  1.151.6.5       mjf 			error = sockopt_get(sopt, &tv, sizeof(tv));
   1706  1.151.6.5       mjf 		solock(so);
   1707  1.151.6.5       mjf 		if (error)
   1708  1.151.6.5       mjf 			break;
   1709  1.151.6.4       mjf 
   1710  1.151.6.5       mjf 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
   1711  1.151.6.5       mjf 			error = EDOM;
   1712  1.151.6.5       mjf 			break;
   1713  1.151.6.5       mjf 		}
   1714       1.28   thorpej 
   1715  1.151.6.4       mjf 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
   1716  1.151.6.4       mjf 		if (optval == 0 && tv.tv_usec != 0)
   1717  1.151.6.4       mjf 			optval = 1;
   1718       1.28   thorpej 
   1719  1.151.6.5       mjf 		switch (opt) {
   1720      1.142    dyoung 		case SO_SNDTIMEO:
   1721  1.151.6.4       mjf 			so->so_snd.sb_timeo = optval;
   1722        1.1       cgd 			break;
   1723        1.1       cgd 		case SO_RCVTIMEO:
   1724  1.151.6.4       mjf 			so->so_rcv.sb_timeo = optval;
   1725      1.142    dyoung 			break;
   1726      1.142    dyoung 		}
   1727      1.142    dyoung 		break;
   1728        1.1       cgd 
   1729      1.142    dyoung 	default:
   1730  1.151.6.5       mjf 		solock(so);
   1731  1.151.6.5       mjf 		error = ENOPROTOOPT;
   1732  1.151.6.5       mjf 		break;
   1733      1.142    dyoung 	}
   1734  1.151.6.5       mjf 	KASSERT(solocked(so));
   1735  1.151.6.5       mjf 	return error;
   1736      1.142    dyoung }
   1737        1.1       cgd 
   1738      1.142    dyoung int
   1739  1.151.6.4       mjf sosetopt(struct socket *so, struct sockopt *sopt)
   1740      1.142    dyoung {
   1741      1.142    dyoung 	int error, prerr;
   1742        1.1       cgd 
   1743  1.151.6.5       mjf 	if (sopt->sopt_level == SOL_SOCKET) {
   1744  1.151.6.4       mjf 		error = sosetopt1(so, sopt);
   1745  1.151.6.5       mjf 		KASSERT(solocked(so));
   1746  1.151.6.5       mjf 	} else {
   1747      1.142    dyoung 		error = ENOPROTOOPT;
   1748  1.151.6.5       mjf 		solock(so);
   1749  1.151.6.5       mjf 	}
   1750        1.1       cgd 
   1751      1.142    dyoung 	if ((error == 0 || error == ENOPROTOOPT) &&
   1752      1.142    dyoung 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
   1753      1.142    dyoung 		/* give the protocol stack a shot */
   1754  1.151.6.4       mjf 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
   1755      1.142    dyoung 		if (prerr == 0)
   1756      1.142    dyoung 			error = 0;
   1757      1.142    dyoung 		else if (prerr != ENOPROTOOPT)
   1758      1.142    dyoung 			error = prerr;
   1759  1.151.6.4       mjf 	}
   1760  1.151.6.2       mjf 	sounlock(so);
   1761      1.142    dyoung 	return error;
   1762        1.1       cgd }
   1763        1.1       cgd 
   1764  1.151.6.4       mjf /*
   1765  1.151.6.4       mjf  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
   1766  1.151.6.4       mjf  */
   1767       1.14   mycroft int
   1768  1.151.6.4       mjf so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
   1769  1.151.6.4       mjf     const void *val, size_t valsize)
   1770  1.151.6.4       mjf {
   1771  1.151.6.4       mjf 	struct sockopt sopt;
   1772  1.151.6.4       mjf 	int error;
   1773  1.151.6.4       mjf 
   1774  1.151.6.4       mjf 	KASSERT(valsize == 0 || val != NULL);
   1775  1.151.6.4       mjf 
   1776  1.151.6.4       mjf 	sockopt_init(&sopt, level, name, valsize);
   1777  1.151.6.4       mjf 	sockopt_set(&sopt, val, valsize);
   1778  1.151.6.4       mjf 
   1779  1.151.6.4       mjf 	error = sosetopt(so, &sopt);
   1780  1.151.6.4       mjf 
   1781  1.151.6.4       mjf 	sockopt_destroy(&sopt);
   1782  1.151.6.4       mjf 
   1783  1.151.6.4       mjf 	return error;
   1784  1.151.6.4       mjf }
   1785  1.151.6.4       mjf 
   1786  1.151.6.4       mjf /*
   1787  1.151.6.4       mjf  * internal get SOL_SOCKET options
   1788  1.151.6.4       mjf  */
   1789  1.151.6.4       mjf static int
   1790  1.151.6.4       mjf sogetopt1(struct socket *so, struct sockopt *sopt)
   1791  1.151.6.4       mjf {
   1792  1.151.6.5       mjf 	int error, optval, opt;
   1793  1.151.6.4       mjf 	struct linger l;
   1794  1.151.6.4       mjf 	struct timeval tv;
   1795  1.151.6.4       mjf 
   1796  1.151.6.5       mjf 	switch ((opt = sopt->sopt_name)) {
   1797  1.151.6.4       mjf 
   1798  1.151.6.4       mjf 	case SO_ACCEPTFILTER:
   1799  1.151.6.5       mjf 		error = accept_filt_getopt(so, sopt);
   1800  1.151.6.4       mjf 		break;
   1801  1.151.6.4       mjf 
   1802  1.151.6.4       mjf 	case SO_LINGER:
   1803  1.151.6.4       mjf 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
   1804  1.151.6.4       mjf 		l.l_linger = so->so_linger;
   1805  1.151.6.4       mjf 
   1806  1.151.6.4       mjf 		error = sockopt_set(sopt, &l, sizeof(l));
   1807  1.151.6.4       mjf 		break;
   1808  1.151.6.4       mjf 
   1809  1.151.6.4       mjf 	case SO_USELOOPBACK:
   1810  1.151.6.4       mjf 	case SO_DONTROUTE:
   1811  1.151.6.4       mjf 	case SO_DEBUG:
   1812  1.151.6.4       mjf 	case SO_KEEPALIVE:
   1813  1.151.6.4       mjf 	case SO_REUSEADDR:
   1814  1.151.6.4       mjf 	case SO_REUSEPORT:
   1815  1.151.6.4       mjf 	case SO_BROADCAST:
   1816  1.151.6.4       mjf 	case SO_OOBINLINE:
   1817  1.151.6.4       mjf 	case SO_TIMESTAMP:
   1818  1.151.6.5       mjf 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
   1819  1.151.6.4       mjf 		break;
   1820  1.151.6.4       mjf 
   1821  1.151.6.4       mjf 	case SO_TYPE:
   1822  1.151.6.4       mjf 		error = sockopt_setint(sopt, so->so_type);
   1823  1.151.6.4       mjf 		break;
   1824  1.151.6.4       mjf 
   1825  1.151.6.4       mjf 	case SO_ERROR:
   1826  1.151.6.4       mjf 		error = sockopt_setint(sopt, so->so_error);
   1827  1.151.6.4       mjf 		so->so_error = 0;
   1828  1.151.6.4       mjf 		break;
   1829  1.151.6.4       mjf 
   1830  1.151.6.4       mjf 	case SO_SNDBUF:
   1831  1.151.6.4       mjf 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
   1832  1.151.6.4       mjf 		break;
   1833  1.151.6.4       mjf 
   1834  1.151.6.4       mjf 	case SO_RCVBUF:
   1835  1.151.6.4       mjf 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
   1836  1.151.6.4       mjf 		break;
   1837  1.151.6.4       mjf 
   1838  1.151.6.4       mjf 	case SO_SNDLOWAT:
   1839  1.151.6.4       mjf 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
   1840  1.151.6.4       mjf 		break;
   1841  1.151.6.4       mjf 
   1842  1.151.6.4       mjf 	case SO_RCVLOWAT:
   1843  1.151.6.4       mjf 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
   1844  1.151.6.4       mjf 		break;
   1845  1.151.6.4       mjf 
   1846  1.151.6.5       mjf #ifdef COMPAT_50
   1847  1.151.6.5       mjf 	case SO_OSNDTIMEO:
   1848  1.151.6.5       mjf 	case SO_ORCVTIMEO: {
   1849  1.151.6.5       mjf 		struct timeval50 otv;
   1850  1.151.6.5       mjf 
   1851  1.151.6.5       mjf 		optval = (opt == SO_OSNDTIMEO ?
   1852  1.151.6.5       mjf 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
   1853  1.151.6.5       mjf 
   1854  1.151.6.5       mjf 		otv.tv_sec = optval / hz;
   1855  1.151.6.5       mjf 		otv.tv_usec = (optval % hz) * tick;
   1856  1.151.6.5       mjf 
   1857  1.151.6.5       mjf 		error = sockopt_set(sopt, &otv, sizeof(otv));
   1858  1.151.6.5       mjf 		break;
   1859  1.151.6.5       mjf 	}
   1860  1.151.6.5       mjf #endif /* COMPAT_50 */
   1861  1.151.6.5       mjf 
   1862  1.151.6.4       mjf 	case SO_SNDTIMEO:
   1863  1.151.6.4       mjf 	case SO_RCVTIMEO:
   1864  1.151.6.5       mjf 		optval = (opt == SO_SNDTIMEO ?
   1865  1.151.6.4       mjf 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
   1866  1.151.6.4       mjf 
   1867  1.151.6.4       mjf 		tv.tv_sec = optval / hz;
   1868  1.151.6.4       mjf 		tv.tv_usec = (optval % hz) * tick;
   1869  1.151.6.4       mjf 
   1870  1.151.6.4       mjf 		error = sockopt_set(sopt, &tv, sizeof(tv));
   1871  1.151.6.4       mjf 		break;
   1872  1.151.6.4       mjf 
   1873  1.151.6.4       mjf 	case SO_OVERFLOWED:
   1874  1.151.6.4       mjf 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
   1875  1.151.6.4       mjf 		break;
   1876  1.151.6.4       mjf 
   1877  1.151.6.4       mjf 	default:
   1878  1.151.6.4       mjf 		error = ENOPROTOOPT;
   1879  1.151.6.4       mjf 		break;
   1880  1.151.6.4       mjf 	}
   1881  1.151.6.4       mjf 
   1882  1.151.6.4       mjf 	return (error);
   1883  1.151.6.4       mjf }
   1884  1.151.6.4       mjf 
   1885  1.151.6.4       mjf int
   1886  1.151.6.4       mjf sogetopt(struct socket *so, struct sockopt *sopt)
   1887        1.1       cgd {
   1888  1.151.6.2       mjf 	int		error;
   1889        1.1       cgd 
   1890  1.151.6.2       mjf 	solock(so);
   1891  1.151.6.4       mjf 	if (sopt->sopt_level != SOL_SOCKET) {
   1892        1.1       cgd 		if (so->so_proto && so->so_proto->pr_ctloutput) {
   1893  1.151.6.2       mjf 			error = ((*so->so_proto->pr_ctloutput)
   1894  1.151.6.4       mjf 			    (PRCO_GETOPT, so, sopt));
   1895        1.1       cgd 		} else
   1896  1.151.6.2       mjf 			error = (ENOPROTOOPT);
   1897        1.1       cgd 	} else {
   1898  1.151.6.4       mjf 		error = sogetopt1(so, sopt);
   1899  1.151.6.4       mjf 	}
   1900  1.151.6.4       mjf 	sounlock(so);
   1901  1.151.6.4       mjf 	return (error);
   1902  1.151.6.4       mjf }
   1903        1.1       cgd 
   1904  1.151.6.4       mjf /*
   1905  1.151.6.4       mjf  * alloc sockopt data buffer buffer
   1906  1.151.6.4       mjf  *	- will be released at destroy
   1907  1.151.6.4       mjf  */
   1908  1.151.6.5       mjf static int
   1909  1.151.6.5       mjf sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
   1910  1.151.6.4       mjf {
   1911        1.1       cgd 
   1912  1.151.6.4       mjf 	KASSERT(sopt->sopt_size == 0);
   1913        1.1       cgd 
   1914  1.151.6.5       mjf 	if (len > sizeof(sopt->sopt_buf)) {
   1915  1.151.6.5       mjf 		sopt->sopt_data = kmem_zalloc(len, kmflag);
   1916  1.151.6.5       mjf 		if (sopt->sopt_data == NULL)
   1917  1.151.6.5       mjf 			return ENOMEM;
   1918  1.151.6.5       mjf 	} else
   1919  1.151.6.4       mjf 		sopt->sopt_data = sopt->sopt_buf;
   1920        1.1       cgd 
   1921  1.151.6.4       mjf 	sopt->sopt_size = len;
   1922  1.151.6.5       mjf 	return 0;
   1923  1.151.6.4       mjf }
   1924        1.1       cgd 
   1925  1.151.6.4       mjf /*
   1926  1.151.6.4       mjf  * initialise sockopt storage
   1927  1.151.6.5       mjf  *	- MAY sleep during allocation
   1928  1.151.6.4       mjf  */
   1929  1.151.6.4       mjf void
   1930  1.151.6.4       mjf sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
   1931  1.151.6.4       mjf {
   1932        1.1       cgd 
   1933  1.151.6.4       mjf 	memset(sopt, 0, sizeof(*sopt));
   1934        1.1       cgd 
   1935  1.151.6.4       mjf 	sopt->sopt_level = level;
   1936  1.151.6.4       mjf 	sopt->sopt_name = name;
   1937  1.151.6.5       mjf 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
   1938  1.151.6.4       mjf }
   1939        1.1       cgd 
   1940  1.151.6.4       mjf /*
   1941  1.151.6.4       mjf  * destroy sockopt storage
   1942  1.151.6.4       mjf  *	- will release any held memory references
   1943  1.151.6.4       mjf  */
   1944  1.151.6.4       mjf void
   1945  1.151.6.4       mjf sockopt_destroy(struct sockopt *sopt)
   1946  1.151.6.4       mjf {
   1947        1.1       cgd 
   1948  1.151.6.4       mjf 	if (sopt->sopt_data != sopt->sopt_buf)
   1949  1.151.6.5       mjf 		kmem_free(sopt->sopt_data, sopt->sopt_size);
   1950        1.1       cgd 
   1951  1.151.6.4       mjf 	memset(sopt, 0, sizeof(*sopt));
   1952  1.151.6.4       mjf }
   1953        1.1       cgd 
   1954  1.151.6.4       mjf /*
   1955  1.151.6.4       mjf  * set sockopt value
   1956  1.151.6.4       mjf  *	- value is copied into sockopt
   1957  1.151.6.5       mjf  * 	- memory is allocated when necessary, will not sleep
   1958  1.151.6.4       mjf  */
   1959  1.151.6.4       mjf int
   1960  1.151.6.4       mjf sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
   1961  1.151.6.4       mjf {
   1962  1.151.6.5       mjf 	int error;
   1963      1.107   darrenr 
   1964  1.151.6.5       mjf 	if (sopt->sopt_size == 0) {
   1965  1.151.6.5       mjf 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
   1966  1.151.6.5       mjf 		if (error)
   1967  1.151.6.5       mjf 			return error;
   1968  1.151.6.5       mjf 	}
   1969  1.151.6.4       mjf 
   1970  1.151.6.4       mjf 	KASSERT(sopt->sopt_size == len);
   1971  1.151.6.4       mjf 	memcpy(sopt->sopt_data, buf, len);
   1972  1.151.6.4       mjf 	return 0;
   1973  1.151.6.4       mjf }
   1974  1.151.6.4       mjf 
   1975  1.151.6.4       mjf /*
   1976  1.151.6.4       mjf  * common case of set sockopt integer value
   1977  1.151.6.4       mjf  */
   1978  1.151.6.4       mjf int
   1979  1.151.6.4       mjf sockopt_setint(struct sockopt *sopt, int val)
   1980  1.151.6.4       mjf {
   1981  1.151.6.4       mjf 
   1982  1.151.6.4       mjf 	return sockopt_set(sopt, &val, sizeof(int));
   1983  1.151.6.4       mjf }
   1984  1.151.6.4       mjf 
   1985  1.151.6.4       mjf /*
   1986  1.151.6.4       mjf  * get sockopt value
   1987  1.151.6.4       mjf  *	- correct size must be given
   1988  1.151.6.4       mjf  */
   1989  1.151.6.4       mjf int
   1990  1.151.6.4       mjf sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
   1991  1.151.6.4       mjf {
   1992  1.151.6.4       mjf 
   1993  1.151.6.4       mjf 	if (sopt->sopt_size != len)
   1994  1.151.6.4       mjf 		return EINVAL;
   1995  1.151.6.4       mjf 
   1996  1.151.6.4       mjf 	memcpy(buf, sopt->sopt_data, len);
   1997  1.151.6.4       mjf 	return 0;
   1998  1.151.6.4       mjf }
   1999  1.151.6.4       mjf 
   2000  1.151.6.4       mjf /*
   2001  1.151.6.4       mjf  * common case of get sockopt integer value
   2002  1.151.6.4       mjf  */
   2003  1.151.6.4       mjf int
   2004  1.151.6.4       mjf sockopt_getint(const struct sockopt *sopt, int *valp)
   2005  1.151.6.4       mjf {
   2006  1.151.6.4       mjf 
   2007  1.151.6.4       mjf 	return sockopt_get(sopt, valp, sizeof(int));
   2008  1.151.6.4       mjf }
   2009  1.151.6.4       mjf 
   2010  1.151.6.4       mjf /*
   2011  1.151.6.4       mjf  * set sockopt value from mbuf
   2012  1.151.6.4       mjf  *	- ONLY for legacy code
   2013  1.151.6.4       mjf  *	- mbuf is released by sockopt
   2014  1.151.6.5       mjf  *	- will not sleep
   2015  1.151.6.4       mjf  */
   2016  1.151.6.4       mjf int
   2017  1.151.6.4       mjf sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
   2018  1.151.6.4       mjf {
   2019  1.151.6.4       mjf 	size_t len;
   2020  1.151.6.5       mjf 	int error;
   2021  1.151.6.4       mjf 
   2022  1.151.6.4       mjf 	len = m_length(m);
   2023  1.151.6.4       mjf 
   2024  1.151.6.5       mjf 	if (sopt->sopt_size == 0) {
   2025  1.151.6.5       mjf 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
   2026  1.151.6.5       mjf 		if (error)
   2027  1.151.6.5       mjf 			return error;
   2028  1.151.6.5       mjf 	}
   2029  1.151.6.4       mjf 
   2030  1.151.6.4       mjf 	KASSERT(sopt->sopt_size == len);
   2031  1.151.6.4       mjf 	m_copydata(m, 0, len, sopt->sopt_data);
   2032  1.151.6.4       mjf 	m_freem(m);
   2033  1.151.6.4       mjf 
   2034  1.151.6.4       mjf 	return 0;
   2035  1.151.6.4       mjf }
   2036  1.151.6.4       mjf 
   2037  1.151.6.4       mjf /*
   2038  1.151.6.4       mjf  * get sockopt value into mbuf
   2039  1.151.6.4       mjf  *	- ONLY for legacy code
   2040  1.151.6.4       mjf  *	- mbuf to be released by the caller
   2041  1.151.6.5       mjf  *	- will not sleep
   2042  1.151.6.4       mjf  */
   2043  1.151.6.4       mjf struct mbuf *
   2044  1.151.6.4       mjf sockopt_getmbuf(const struct sockopt *sopt)
   2045  1.151.6.4       mjf {
   2046  1.151.6.4       mjf 	struct mbuf *m;
   2047  1.151.6.4       mjf 
   2048  1.151.6.5       mjf 	if (sopt->sopt_size > MCLBYTES)
   2049  1.151.6.4       mjf 		return NULL;
   2050  1.151.6.4       mjf 
   2051  1.151.6.5       mjf 	m = m_get(M_DONTWAIT, MT_SOOPTS);
   2052  1.151.6.5       mjf 	if (m == NULL)
   2053  1.151.6.4       mjf 		return NULL;
   2054  1.151.6.5       mjf 
   2055  1.151.6.5       mjf 	if (sopt->sopt_size > MLEN) {
   2056  1.151.6.5       mjf 		MCLGET(m, M_DONTWAIT);
   2057  1.151.6.5       mjf 		if ((m->m_flags & M_EXT) == 0) {
   2058  1.151.6.5       mjf 			m_free(m);
   2059  1.151.6.5       mjf 			return NULL;
   2060  1.151.6.5       mjf 		}
   2061        1.1       cgd 	}
   2062  1.151.6.5       mjf 
   2063  1.151.6.5       mjf 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
   2064  1.151.6.5       mjf 	m->m_len = sopt->sopt_size;
   2065  1.151.6.2       mjf 
   2066  1.151.6.4       mjf 	return m;
   2067        1.1       cgd }
   2068        1.1       cgd 
   2069       1.14   mycroft void
   2070       1.54     lukem sohasoutofband(struct socket *so)
   2071        1.1       cgd {
   2072  1.151.6.1       mjf 
   2073       1.90  christos 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
   2074  1.151.6.1       mjf 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, 0);
   2075        1.1       cgd }
   2076       1.72  jdolecek 
   2077       1.72  jdolecek static void
   2078       1.72  jdolecek filt_sordetach(struct knote *kn)
   2079       1.72  jdolecek {
   2080       1.72  jdolecek 	struct socket	*so;
   2081       1.72  jdolecek 
   2082  1.151.6.1       mjf 	so = ((file_t *)kn->kn_obj)->f_data;
   2083  1.151.6.2       mjf 	solock(so);
   2084       1.73  christos 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
   2085       1.73  christos 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
   2086       1.72  jdolecek 		so->so_rcv.sb_flags &= ~SB_KNOTE;
   2087  1.151.6.2       mjf 	sounlock(so);
   2088       1.72  jdolecek }
   2089       1.72  jdolecek 
   2090       1.72  jdolecek /*ARGSUSED*/
   2091       1.72  jdolecek static int
   2092      1.129      yamt filt_soread(struct knote *kn, long hint)
   2093       1.72  jdolecek {
   2094       1.72  jdolecek 	struct socket	*so;
   2095  1.151.6.2       mjf 	int rv;
   2096       1.72  jdolecek 
   2097  1.151.6.1       mjf 	so = ((file_t *)kn->kn_obj)->f_data;
   2098  1.151.6.2       mjf 	if (hint != NOTE_SUBMIT)
   2099  1.151.6.2       mjf 		solock(so);
   2100       1.72  jdolecek 	kn->kn_data = so->so_rcv.sb_cc;
   2101       1.72  jdolecek 	if (so->so_state & SS_CANTRCVMORE) {
   2102      1.108     perry 		kn->kn_flags |= EV_EOF;
   2103       1.72  jdolecek 		kn->kn_fflags = so->so_error;
   2104  1.151.6.2       mjf 		rv = 1;
   2105  1.151.6.2       mjf 	} else if (so->so_error)	/* temporary udp error */
   2106  1.151.6.2       mjf 		rv = 1;
   2107  1.151.6.2       mjf 	else if (kn->kn_sfflags & NOTE_LOWAT)
   2108  1.151.6.2       mjf 		rv = (kn->kn_data >= kn->kn_sdata);
   2109  1.151.6.2       mjf 	else
   2110  1.151.6.2       mjf 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
   2111  1.151.6.2       mjf 	if (hint != NOTE_SUBMIT)
   2112  1.151.6.2       mjf 		sounlock(so);
   2113  1.151.6.2       mjf 	return rv;
   2114       1.72  jdolecek }
   2115       1.72  jdolecek 
   2116       1.72  jdolecek static void
   2117       1.72  jdolecek filt_sowdetach(struct knote *kn)
   2118       1.72  jdolecek {
   2119       1.72  jdolecek 	struct socket	*so;
   2120       1.72  jdolecek 
   2121  1.151.6.1       mjf 	so = ((file_t *)kn->kn_obj)->f_data;
   2122  1.151.6.2       mjf 	solock(so);
   2123       1.73  christos 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
   2124       1.73  christos 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
   2125       1.72  jdolecek 		so->so_snd.sb_flags &= ~SB_KNOTE;
   2126  1.151.6.2       mjf 	sounlock(so);
   2127       1.72  jdolecek }
   2128       1.72  jdolecek 
   2129       1.72  jdolecek /*ARGSUSED*/
   2130       1.72  jdolecek static int
   2131      1.129      yamt filt_sowrite(struct knote *kn, long hint)
   2132       1.72  jdolecek {
   2133       1.72  jdolecek 	struct socket	*so;
   2134  1.151.6.2       mjf 	int rv;
   2135       1.72  jdolecek 
   2136  1.151.6.1       mjf 	so = ((file_t *)kn->kn_obj)->f_data;
   2137  1.151.6.2       mjf 	if (hint != NOTE_SUBMIT)
   2138  1.151.6.2       mjf 		solock(so);
   2139       1.72  jdolecek 	kn->kn_data = sbspace(&so->so_snd);
   2140       1.72  jdolecek 	if (so->so_state & SS_CANTSENDMORE) {
   2141      1.108     perry 		kn->kn_flags |= EV_EOF;
   2142       1.72  jdolecek 		kn->kn_fflags = so->so_error;
   2143  1.151.6.2       mjf 		rv = 1;
   2144  1.151.6.2       mjf 	} else if (so->so_error)	/* temporary udp error */
   2145  1.151.6.2       mjf 		rv = 1;
   2146  1.151.6.2       mjf 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
   2147       1.72  jdolecek 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
   2148  1.151.6.2       mjf 		rv = 0;
   2149  1.151.6.2       mjf 	else if (kn->kn_sfflags & NOTE_LOWAT)
   2150  1.151.6.2       mjf 		rv = (kn->kn_data >= kn->kn_sdata);
   2151  1.151.6.2       mjf 	else
   2152  1.151.6.2       mjf 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
   2153  1.151.6.2       mjf 	if (hint != NOTE_SUBMIT)
   2154  1.151.6.2       mjf 		sounlock(so);
   2155  1.151.6.2       mjf 	return rv;
   2156       1.72  jdolecek }
   2157       1.72  jdolecek 
   2158       1.72  jdolecek /*ARGSUSED*/
   2159       1.72  jdolecek static int
   2160      1.129      yamt filt_solisten(struct knote *kn, long hint)
   2161       1.72  jdolecek {
   2162       1.72  jdolecek 	struct socket	*so;
   2163  1.151.6.2       mjf 	int rv;
   2164       1.72  jdolecek 
   2165  1.151.6.1       mjf 	so = ((file_t *)kn->kn_obj)->f_data;
   2166       1.72  jdolecek 
   2167       1.72  jdolecek 	/*
   2168       1.72  jdolecek 	 * Set kn_data to number of incoming connections, not
   2169       1.72  jdolecek 	 * counting partial (incomplete) connections.
   2170      1.108     perry 	 */
   2171  1.151.6.2       mjf 	if (hint != NOTE_SUBMIT)
   2172  1.151.6.2       mjf 		solock(so);
   2173       1.72  jdolecek 	kn->kn_data = so->so_qlen;
   2174  1.151.6.2       mjf 	rv = (kn->kn_data > 0);
   2175  1.151.6.2       mjf 	if (hint != NOTE_SUBMIT)
   2176  1.151.6.2       mjf 		sounlock(so);
   2177  1.151.6.2       mjf 	return rv;
   2178       1.72  jdolecek }
   2179       1.72  jdolecek 
   2180       1.72  jdolecek static const struct filterops solisten_filtops =
   2181       1.72  jdolecek 	{ 1, NULL, filt_sordetach, filt_solisten };
   2182       1.72  jdolecek static const struct filterops soread_filtops =
   2183       1.72  jdolecek 	{ 1, NULL, filt_sordetach, filt_soread };
   2184       1.72  jdolecek static const struct filterops sowrite_filtops =
   2185       1.72  jdolecek 	{ 1, NULL, filt_sowdetach, filt_sowrite };
   2186       1.72  jdolecek 
   2187       1.72  jdolecek int
   2188      1.129      yamt soo_kqfilter(struct file *fp, struct knote *kn)
   2189       1.72  jdolecek {
   2190       1.72  jdolecek 	struct socket	*so;
   2191       1.72  jdolecek 	struct sockbuf	*sb;
   2192       1.72  jdolecek 
   2193  1.151.6.1       mjf 	so = ((file_t *)kn->kn_obj)->f_data;
   2194  1.151.6.2       mjf 	solock(so);
   2195       1.72  jdolecek 	switch (kn->kn_filter) {
   2196       1.72  jdolecek 	case EVFILT_READ:
   2197       1.72  jdolecek 		if (so->so_options & SO_ACCEPTCONN)
   2198       1.72  jdolecek 			kn->kn_fop = &solisten_filtops;
   2199       1.72  jdolecek 		else
   2200       1.72  jdolecek 			kn->kn_fop = &soread_filtops;
   2201       1.72  jdolecek 		sb = &so->so_rcv;
   2202       1.72  jdolecek 		break;
   2203       1.72  jdolecek 	case EVFILT_WRITE:
   2204       1.72  jdolecek 		kn->kn_fop = &sowrite_filtops;
   2205       1.72  jdolecek 		sb = &so->so_snd;
   2206       1.72  jdolecek 		break;
   2207       1.72  jdolecek 	default:
   2208  1.151.6.2       mjf 		sounlock(so);
   2209      1.149     pooka 		return (EINVAL);
   2210       1.72  jdolecek 	}
   2211       1.73  christos 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
   2212       1.72  jdolecek 	sb->sb_flags |= SB_KNOTE;
   2213  1.151.6.2       mjf 	sounlock(so);
   2214       1.72  jdolecek 	return (0);
   2215       1.72  jdolecek }
   2216       1.72  jdolecek 
   2217  1.151.6.1       mjf static int
   2218  1.151.6.1       mjf sodopoll(struct socket *so, int events)
   2219  1.151.6.1       mjf {
   2220  1.151.6.1       mjf 	int revents;
   2221  1.151.6.1       mjf 
   2222  1.151.6.1       mjf 	revents = 0;
   2223  1.151.6.1       mjf 
   2224  1.151.6.1       mjf 	if (events & (POLLIN | POLLRDNORM))
   2225  1.151.6.1       mjf 		if (soreadable(so))
   2226  1.151.6.1       mjf 			revents |= events & (POLLIN | POLLRDNORM);
   2227  1.151.6.1       mjf 
   2228  1.151.6.1       mjf 	if (events & (POLLOUT | POLLWRNORM))
   2229  1.151.6.1       mjf 		if (sowritable(so))
   2230  1.151.6.1       mjf 			revents |= events & (POLLOUT | POLLWRNORM);
   2231  1.151.6.1       mjf 
   2232  1.151.6.1       mjf 	if (events & (POLLPRI | POLLRDBAND))
   2233  1.151.6.1       mjf 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
   2234  1.151.6.1       mjf 			revents |= events & (POLLPRI | POLLRDBAND);
   2235  1.151.6.1       mjf 
   2236  1.151.6.1       mjf 	return revents;
   2237  1.151.6.1       mjf }
   2238  1.151.6.1       mjf 
   2239  1.151.6.1       mjf int
   2240  1.151.6.1       mjf sopoll(struct socket *so, int events)
   2241  1.151.6.1       mjf {
   2242  1.151.6.1       mjf 	int revents = 0;
   2243  1.151.6.1       mjf 
   2244  1.151.6.2       mjf #ifndef DIAGNOSTIC
   2245  1.151.6.2       mjf 	/*
   2246  1.151.6.2       mjf 	 * Do a quick, unlocked check in expectation that the socket
   2247  1.151.6.2       mjf 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
   2248  1.151.6.2       mjf 	 * as the solocked() assertions will fail.
   2249  1.151.6.2       mjf 	 */
   2250  1.151.6.1       mjf 	if ((revents = sodopoll(so, events)) != 0)
   2251  1.151.6.1       mjf 		return revents;
   2252  1.151.6.2       mjf #endif
   2253  1.151.6.1       mjf 
   2254  1.151.6.2       mjf 	solock(so);
   2255  1.151.6.1       mjf 	if ((revents = sodopoll(so, events)) == 0) {
   2256  1.151.6.1       mjf 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
   2257  1.151.6.1       mjf 			selrecord(curlwp, &so->so_rcv.sb_sel);
   2258  1.151.6.2       mjf 			so->so_rcv.sb_flags |= SB_NOTIFY;
   2259  1.151.6.1       mjf 		}
   2260  1.151.6.1       mjf 
   2261  1.151.6.1       mjf 		if (events & (POLLOUT | POLLWRNORM)) {
   2262  1.151.6.1       mjf 			selrecord(curlwp, &so->so_snd.sb_sel);
   2263  1.151.6.2       mjf 			so->so_snd.sb_flags |= SB_NOTIFY;
   2264  1.151.6.1       mjf 		}
   2265  1.151.6.1       mjf 	}
   2266  1.151.6.2       mjf 	sounlock(so);
   2267  1.151.6.1       mjf 
   2268  1.151.6.1       mjf 	return revents;
   2269  1.151.6.1       mjf }
   2270  1.151.6.1       mjf 
   2271  1.151.6.1       mjf 
   2272       1.94      yamt #include <sys/sysctl.h>
   2273       1.94      yamt 
   2274       1.94      yamt static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
   2275       1.94      yamt 
   2276       1.94      yamt /*
   2277       1.94      yamt  * sysctl helper routine for kern.somaxkva.  ensures that the given
   2278       1.94      yamt  * value is not too small.
   2279       1.94      yamt  * (XXX should we maybe make sure it's not too large as well?)
   2280       1.94      yamt  */
   2281       1.94      yamt static int
   2282       1.94      yamt sysctl_kern_somaxkva(SYSCTLFN_ARGS)
   2283       1.94      yamt {
   2284       1.94      yamt 	int error, new_somaxkva;
   2285       1.94      yamt 	struct sysctlnode node;
   2286       1.94      yamt 
   2287       1.94      yamt 	new_somaxkva = somaxkva;
   2288       1.94      yamt 	node = *rnode;
   2289       1.94      yamt 	node.sysctl_data = &new_somaxkva;
   2290       1.94      yamt 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
   2291       1.94      yamt 	if (error || newp == NULL)
   2292       1.94      yamt 		return (error);
   2293       1.94      yamt 
   2294       1.94      yamt 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
   2295       1.94      yamt 		return (EINVAL);
   2296       1.94      yamt 
   2297      1.136        ad 	mutex_enter(&so_pendfree_lock);
   2298       1.94      yamt 	somaxkva = new_somaxkva;
   2299      1.136        ad 	cv_broadcast(&socurkva_cv);
   2300      1.136        ad 	mutex_exit(&so_pendfree_lock);
   2301       1.94      yamt 
   2302       1.94      yamt 	return (error);
   2303       1.94      yamt }
   2304       1.94      yamt 
   2305  1.151.6.5       mjf static void
   2306  1.151.6.5       mjf sysctl_kern_somaxkva_setup()
   2307       1.94      yamt {
   2308       1.94      yamt 
   2309  1.151.6.5       mjf 	KASSERT(socket_sysctllog == NULL);
   2310  1.151.6.5       mjf 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
   2311       1.97    atatat 		       CTLFLAG_PERMANENT,
   2312       1.97    atatat 		       CTLTYPE_NODE, "kern", NULL,
   2313       1.97    atatat 		       NULL, 0, NULL, 0,
   2314       1.97    atatat 		       CTL_KERN, CTL_EOL);
   2315       1.97    atatat 
   2316  1.151.6.5       mjf 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
   2317       1.97    atatat 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2318      1.103    atatat 		       CTLTYPE_INT, "somaxkva",
   2319      1.103    atatat 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
   2320      1.103    atatat 				    "used for socket buffers"),
   2321       1.94      yamt 		       sysctl_kern_somaxkva, 0, NULL, 0,
   2322       1.94      yamt 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
   2323       1.94      yamt }
   2324