X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/a3d08fcd5120d2aa8303b6349ca8b14e3f284af3..143464d58d2bd6378e74eec636961ceb0d32fb91:/bsd/kern/uipc_socket.c diff --git a/bsd/kern/uipc_socket.c b/bsd/kern/uipc_socket.c index ebeeec818..f7f3d5202 100644 --- a/bsd/kern/uipc_socket.c +++ b/bsd/kern/uipc_socket.c @@ -1,25 +1,30 @@ /* - * Copyright (c) 2000 Apple Computer, Inc. All rights reserved. + * Copyright (c) 1998-2013 Apple Inc. All rights reserved. * - * @APPLE_LICENSE_HEADER_START@ - * - * The contents of this file constitute Original Code as defined in and - * are subject to the Apple Public Source License Version 1.1 (the - * "License"). You may not use this file except in compliance with the - * License. Please obtain a copy of the License at - * http://www.apple.com/publicsource and read it before using this file. - * - * This Original Code and all software distributed under the License are - * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ + * + * This file contains Original Code and/or Modifications of Original Code + * as defined in and that are subject to the Apple Public Source License + * Version 2.0 (the 'License'). You may not use this file except in + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * unlawful or unlicensed copies of an Apple operating system, or to + * circumvent, violate, or enable the circumvention or violation of, any + * terms of an Apple operating system software license agreement. + * + * Please obtain a copy of the License at + * http://www.opensource.apple.com/apsl/ and read it before using this file. + * + * The Original Code and all software distributed under the License are + * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the - * License for the specific language governing rights and limitations - * under the License. - * - * @APPLE_LICENSE_HEADER_END@ + * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. + * Please see the License for the specific language governing rights and + * limitations under the License. + * + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ -/* Copyright (c) 1998, 1999 Apple Computer, Inc. All Rights Reserved */ /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */ /* * Copyright (c) 1982, 1986, 1988, 1990, 1993 @@ -54,14 +59,21 @@ * SUCH DAMAGE. * * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 - * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.16 2001/06/14 20:46:06 ume Exp $ + */ +/* + * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce + * support for mandatory and extensible security protections. This notice + * is included in support of clause 2.2 (b) of the Apple Public License, + * Version 2.0. */ #include #include #include #include -#include +#include +#include +#include #include #include #include @@ -75,77 +87,156 @@ #include #include #include +#include #include #include #include +#include +#include +#include +#include #include +#include +#include #include #include +#include +#include +#include #include +#include #include - -int so_cache_hw = 0; -int so_cache_timeouts = 0; -int so_cache_max_freed = 0; -int cached_sock_count = 0; -struct socket *socket_cache_head = 0; -struct socket *socket_cache_tail = 0; -u_long so_cache_time = 0; -int so_cache_init_done = 0; -struct zone *so_cache_zone; -extern int get_inpcb_str_size(); -extern int get_tcp_str_size(); +#include +#include +#include +#include +#include +#include + +#if CONFIG_MACF +#include +#include +#endif /* MAC */ + +#if MULTIPATH +#include +#endif /* MULTIPATH */ + +/* TODO: this should be in a header file somewhere */ +extern char *proc_name_address(void *p); + +static u_int32_t so_cache_hw; /* High water mark for socache */ +static u_int32_t so_cache_timeouts; /* number of timeouts */ +static u_int32_t so_cache_max_freed; /* max freed per timeout */ +static u_int32_t cached_sock_count = 0; +STAILQ_HEAD(, socket) so_cache_head; +int max_cached_sock_count = MAX_CACHED_SOCKETS; +static u_int32_t so_cache_time; +static int socketinit_done; +static struct zone *so_cache_zone; + +static lck_grp_t *so_cache_mtx_grp; +static lck_attr_t *so_cache_mtx_attr; +static lck_grp_attr_t *so_cache_mtx_grp_attr; +static lck_mtx_t *so_cache_mtx; #include -static void filt_sordetach(struct knote *kn); -static int filt_soread(struct knote *kn, long hint); -static void filt_sowdetach(struct knote *kn); -static int filt_sowrite(struct knote *kn, long hint); -static int filt_solisten(struct knote *kn, long hint); - -static struct filterops solisten_filtops = - { 1, NULL, filt_sordetach, filt_solisten }; -static struct filterops soread_filtops = - { 1, NULL, filt_sordetach, filt_soread }; -static struct filterops sowrite_filtops = - { 1, NULL, filt_sowdetach, filt_sowrite }; - +static void filt_sordetach(struct knote *kn); +static int filt_soread(struct knote *kn, long hint); +static void filt_sowdetach(struct knote *kn); +static int filt_sowrite(struct knote *kn, long hint); +static void filt_sockdetach(struct knote *kn); +static int filt_sockev(struct knote *kn, long hint); + +static int sooptcopyin_timeval(struct sockopt *, struct timeval *); +static int sooptcopyout_timeval(struct sockopt *, const struct timeval *); + +static struct filterops soread_filtops = { + .f_isfd = 1, + .f_detach = filt_sordetach, + .f_event = filt_soread, +}; + +static struct filterops sowrite_filtops = { + .f_isfd = 1, + .f_detach = filt_sowdetach, + .f_event = filt_sowrite, +}; + +static struct filterops sock_filtops = { + .f_isfd = 1, + .f_detach = filt_sockdetach, + .f_event = filt_sockev, +}; + +#define EVEN_MORE_LOCKING_DEBUG 0 int socket_debug = 0; -int socket_zone = M_SOCKET; +static int socket_zone = M_SOCKET; so_gen_t so_gencnt; /* generation count for sockets */ MALLOC_DEFINE(M_SONAME, "soname", "socket name"); MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); -#define DBG_LAYER_IN_BEG NETDBG_CODE(DBG_NETSOCK, 0) -#define DBG_LAYER_IN_END NETDBG_CODE(DBG_NETSOCK, 2) -#define DBG_LAYER_OUT_BEG NETDBG_CODE(DBG_NETSOCK, 1) -#define DBG_LAYER_OUT_END NETDBG_CODE(DBG_NETSOCK, 3) -#define DBG_FNC_SOSEND NETDBG_CODE(DBG_NETSOCK, (4 << 8) | 1) -#define DBG_FNC_SORECEIVE NETDBG_CODE(DBG_NETSOCK, (8 << 8)) -#define DBG_FNC_SOSHUTDOWN NETDBG_CODE(DBG_NETSOCK, (9 << 8)) +#define DBG_LAYER_IN_BEG NETDBG_CODE(DBG_NETSOCK, 0) +#define DBG_LAYER_IN_END NETDBG_CODE(DBG_NETSOCK, 2) +#define DBG_LAYER_OUT_BEG NETDBG_CODE(DBG_NETSOCK, 1) +#define DBG_LAYER_OUT_END NETDBG_CODE(DBG_NETSOCK, 3) +#define DBG_FNC_SOSEND NETDBG_CODE(DBG_NETSOCK, (4 << 8) | 1) +#define DBG_FNC_SORECEIVE NETDBG_CODE(DBG_NETSOCK, (8 << 8)) +#define DBG_FNC_SOSHUTDOWN NETDBG_CODE(DBG_NETSOCK, (9 << 8)) -#define MAX_SOOPTGETM_SIZE (128 * MCLBYTES) +#define MAX_SOOPTGETM_SIZE (128 * MCLBYTES) SYSCTL_DECL(_kern_ipc); -static int somaxconn = SOMAXCONN; -SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, &somaxconn, - 0, ""); +int somaxconn = SOMAXCONN; +SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, + CTLFLAG_RW | CTLFLAG_LOCKED, &somaxconn, 0, ""); /* Should we get a maximum also ??? */ static int sosendmaxchain = 65536; static int sosendminchain = 16384; static int sorecvmincopy = 16384; -SYSCTL_INT(_kern_ipc, OID_AUTO, sosendminchain, CTLFLAG_RW, &sosendminchain, - 0, ""); -SYSCTL_INT(_kern_ipc, OID_AUTO, sorecvmincopy, CTLFLAG_RW, &sorecvmincopy, - 0, ""); +SYSCTL_INT(_kern_ipc, OID_AUTO, sosendminchain, + CTLFLAG_RW | CTLFLAG_LOCKED, &sosendminchain, 0, ""); +SYSCTL_INT(_kern_ipc, OID_AUTO, sorecvmincopy, + CTLFLAG_RW | CTLFLAG_LOCKED, &sorecvmincopy, 0, ""); + +/* + * Set to enable jumbo clusters (if available) for large writes when + * the socket is marked with SOF_MULTIPAGES; see below. + */ +int sosendjcl = 1; +SYSCTL_INT(_kern_ipc, OID_AUTO, sosendjcl, + CTLFLAG_RW | CTLFLAG_LOCKED, &sosendjcl, 0, ""); + +/* + * Set this to ignore SOF_MULTIPAGES and use jumbo clusters for large + * writes on the socket for all protocols on any network interfaces, + * depending upon sosendjcl above. Be extra careful when setting this + * to 1, because sending down packets that cross physical pages down to + * broken drivers (those that falsely assume that the physical pages + * are contiguous) might lead to system panics or silent data corruption. + * When set to 0, the system will respect SOF_MULTIPAGES, which is set + * only for TCP sockets whose outgoing interface is IFNET_MULTIPAGES + * capable. Set this to 1 only for testing/debugging purposes. + */ +int sosendjcl_ignore_capab = 0; +SYSCTL_INT(_kern_ipc, OID_AUTO, sosendjcl_ignore_capab, + CTLFLAG_RW | CTLFLAG_LOCKED, &sosendjcl_ignore_capab, 0, ""); + +int sodefunctlog = 0; +SYSCTL_INT(_kern_ipc, OID_AUTO, sodefunctlog, CTLFLAG_RW | CTLFLAG_LOCKED, + &sodefunctlog, 0, ""); -void so_cache_timer(); -struct mbuf *m_getpackets(int, int, int); +int sothrottlelog = 0; +SYSCTL_INT(_kern_ipc, OID_AUTO, sothrottlelog, CTLFLAG_RW | CTLFLAG_LOCKED, + &sothrottlelog, 0, ""); +int sorestrictrecv = 1; +SYSCTL_INT(_kern_ipc, OID_AUTO, sorestrictrecv, CTLFLAG_RW | CTLFLAG_LOCKED, + &sorestrictrecv, 0, "Enable inbound interface restrictions"); /* * Socket operation routines. @@ -155,189 +246,249 @@ struct mbuf *m_getpackets(int, int, int); * switching out to the protocol specific routines. */ -#ifdef __APPLE__ -void socketinit() -{ - vm_size_t str_size; +/* sys_generic.c */ +extern void postevent(struct socket *, struct sockbuf *, int); +extern void evsofree(struct socket *); +extern int tcp_notsent_lowat_check(struct socket *so); +extern struct inpcbinfo tcbinfo; - so_cache_init_done = 1; +/* TODO: these should be in header file */ +extern int get_inpcb_str_size(void); +extern int get_tcp_str_size(void); - timeout(so_cache_timer, NULL, (SO_CACHE_FLUSH_INTERVAL * hz)); - str_size = (vm_size_t)( sizeof(struct socket) + 4 + - get_inpcb_str_size() + 4 + - get_tcp_str_size()); - so_cache_zone = zinit (str_size, 120000*str_size, 8192, "socache zone"); -#if TEMPDEBUG - kprintf("cached_sock_alloc -- so_cache_zone size is %x\n", str_size); -#endif +static unsigned int sl_zone_size; /* size of sockaddr_list */ +static struct zone *sl_zone; /* zone for sockaddr_list */ -} +static unsigned int se_zone_size; /* size of sockaddr_entry */ +static struct zone *se_zone; /* zone for sockaddr_entry */ + +vm_size_t so_cache_zone_element_size; -void cached_sock_alloc(so, waitok) -struct socket **so; -int waitok; +static int sodelayed_copy(struct socket *, struct uio *, struct mbuf **, user_ssize_t *); +static void cached_sock_alloc(struct socket **, int); +static void cached_sock_free(struct socket *); +/* + * SOTCDB_NO_DSCP is set by default, to prevent the networking stack from + * setting the DSCP code on the packet based on the service class; see + * for details. + */ +__private_extern__ u_int32_t sotcdb = SOTCDB_NO_DSCP; +SYSCTL_INT(_kern_ipc, OID_AUTO, sotcdb, CTLFLAG_RW | CTLFLAG_LOCKED, + &sotcdb, 0, ""); + +void +socketinit(void) { - caddr_t temp; - int s; - register u_long offset; + if (socketinit_done) { + printf("socketinit: already called...\n"); + return; + } + socketinit_done = 1; + PE_parse_boot_argn("socket_debug", &socket_debug, + sizeof (socket_debug)); - s = splnet(); - if (cached_sock_count) { - cached_sock_count--; - *so = socket_cache_head; - if (*so == 0) - panic("cached_sock_alloc: cached sock is null"); + /* + * allocate lock group attribute and group for socket cache mutex + */ + so_cache_mtx_grp_attr = lck_grp_attr_alloc_init(); + so_cache_mtx_grp = lck_grp_alloc_init("so_cache", + so_cache_mtx_grp_attr); - socket_cache_head = socket_cache_head->cache_next; - if (socket_cache_head) - socket_cache_head->cache_prev = 0; - else - socket_cache_tail = 0; - splx(s); + /* + * allocate the lock attribute for socket cache mutex + */ + so_cache_mtx_attr = lck_attr_alloc_init(); - temp = (*so)->so_saved_pcb; - bzero((caddr_t)*so, sizeof(struct socket)); -#if TEMPDEBUG - kprintf("cached_sock_alloc - retreiving cached sock %x - count == %d\n", *so, - cached_sock_count); -#endif - (*so)->so_saved_pcb = temp; - } - else { -#if TEMPDEBUG - kprintf("Allocating cached sock %x from memory\n", *so); -#endif + /* cached sockets mutex */ + so_cache_mtx = lck_mtx_alloc_init(so_cache_mtx_grp, so_cache_mtx_attr); + if (so_cache_mtx == NULL) { + panic("%s: unable to allocate so_cache_mtx\n", __func__); + /* NOTREACHED */ + } + STAILQ_INIT(&so_cache_head); - splx(s); - if (waitok) - *so = (struct socket *) zalloc(so_cache_zone); - else - *so = (struct socket *) zalloc_noblock(so_cache_zone); - - if (*so == 0) - return; - - bzero((caddr_t)*so, sizeof(struct socket)); - - /* - * Define offsets for extra structures into our single block of - * memory. Align extra structures on longword boundaries. - */ - - - offset = (u_long) *so; - offset += sizeof(struct socket); - if (offset & 0x3) { - offset += 4; - offset &= 0xfffffffc; - } - (*so)->so_saved_pcb = (caddr_t) offset; - offset += get_inpcb_str_size(); - if (offset & 0x3) { - offset += 4; - offset &= 0xfffffffc; - } - - ((struct inpcb *) (*so)->so_saved_pcb)->inp_saved_ppcb = (caddr_t) offset; -#if TEMPDEBUG - kprintf("Allocating cached socket - %x, pcb=%x tcpcb=%x\n", *so, - (*so)->so_saved_pcb, - ((struct inpcb *)(*so)->so_saved_pcb)->inp_saved_ppcb); -#endif - } + so_cache_zone_element_size = (vm_size_t)(sizeof (struct socket) + 4 + + get_inpcb_str_size() + 4 + get_tcp_str_size()); - (*so)->cached_in_sock_layer = 1; -} + so_cache_zone = zinit(so_cache_zone_element_size, + (120000 * so_cache_zone_element_size), 8192, "socache zone"); + zone_change(so_cache_zone, Z_CALLERACCT, FALSE); + zone_change(so_cache_zone, Z_NOENCRYPT, TRUE); + + sl_zone_size = sizeof (struct sockaddr_list); + if ((sl_zone = zinit(sl_zone_size, 1024 * sl_zone_size, 1024, + "sockaddr_list")) == NULL) { + panic("%s: unable to allocate sockaddr_list zone\n", __func__); + /* NOTREACHED */ + } + zone_change(sl_zone, Z_CALLERACCT, FALSE); + zone_change(sl_zone, Z_EXPAND, TRUE); + + se_zone_size = sizeof (struct sockaddr_entry); + if ((se_zone = zinit(se_zone_size, 1024 * se_zone_size, 1024, + "sockaddr_entry")) == NULL) { + panic("%s: unable to allocate sockaddr_entry zone\n", __func__); + /* NOTREACHED */ + } + zone_change(se_zone, Z_CALLERACCT, FALSE); + zone_change(se_zone, Z_EXPAND, TRUE); -void cached_sock_free(so) -struct socket *so; + in_pcbinit(); + sflt_init(); + socket_tclass_init(); +#if MULTIPATH + mp_pcbinit(); +#endif /* MULTIPATH */ +} + +static void +cached_sock_alloc(struct socket **so, int waitok) { - int s; + caddr_t temp; + uintptr_t offset; + lck_mtx_lock(so_cache_mtx); - s = splnet(); - if (++cached_sock_count > MAX_CACHED_SOCKETS) { - --cached_sock_count; - splx(s); -#if TEMPDEBUG - kprintf("Freeing overflowed cached socket %x\n", so); -#endif - zfree(so_cache_zone, (vm_offset_t) so); - } - else { -#if TEMPDEBUG - kprintf("Freeing socket %x into cache\n", so); -#endif - if (so_cache_hw < cached_sock_count) - so_cache_hw = cached_sock_count; + if (!STAILQ_EMPTY(&so_cache_head)) { + VERIFY(cached_sock_count > 0); + + *so = STAILQ_FIRST(&so_cache_head); + STAILQ_REMOVE_HEAD(&so_cache_head, so_cache_ent); + STAILQ_NEXT((*so), so_cache_ent) = NULL; - so->cache_next = socket_cache_head; - so->cache_prev = 0; - if (socket_cache_head) - socket_cache_head->cache_prev = so; + cached_sock_count--; + lck_mtx_unlock(so_cache_mtx); + + temp = (*so)->so_saved_pcb; + bzero((caddr_t)*so, sizeof (struct socket)); + + (*so)->so_saved_pcb = temp; + } else { + + lck_mtx_unlock(so_cache_mtx); + + if (waitok) + *so = (struct socket *)zalloc(so_cache_zone); else - socket_cache_tail = so; + *so = (struct socket *)zalloc_noblock(so_cache_zone); - so->cache_timestamp = so_cache_time; - socket_cache_head = so; - splx(s); - } + if (*so == NULL) + return; -#if TEMPDEBUG - kprintf("Freed cached sock %x into cache - count is %d\n", so, cached_sock_count); -#endif + bzero((caddr_t)*so, sizeof (struct socket)); + /* + * Define offsets for extra structures into our + * single block of memory. Align extra structures + * on longword boundaries. + */ -} + offset = (uintptr_t)*so; + offset += sizeof (struct socket); + + offset = ALIGN(offset); + + (*so)->so_saved_pcb = (caddr_t)offset; + offset += get_inpcb_str_size(); + + offset = ALIGN(offset); + + ((struct inpcb *)(void *)(*so)->so_saved_pcb)->inp_saved_ppcb = + (caddr_t)offset; + } + (*so)->cached_in_sock_layer = true; +} -void so_cache_timer() +static void +cached_sock_free(struct socket *so) { - register struct socket *p; - register int s; - register int n_freed = 0; - boolean_t funnel_state; - funnel_state = thread_funnel_set(network_flock, TRUE); + lck_mtx_lock(so_cache_mtx); + + so_cache_time = net_uptime(); + if (++cached_sock_count > max_cached_sock_count) { + --cached_sock_count; + lck_mtx_unlock(so_cache_mtx); + zfree(so_cache_zone, so); + } else { + if (so_cache_hw < cached_sock_count) + so_cache_hw = cached_sock_count; + + STAILQ_INSERT_TAIL(&so_cache_head, so, so_cache_ent); + + so->cache_timestamp = so_cache_time; + lck_mtx_unlock(so_cache_mtx); + } +} - ++so_cache_time; +void +so_update_last_owner_locked(struct socket *so, proc_t self) +{ + if (so->last_pid != 0) { + /* + * last_pid and last_upid should remain zero for sockets + * created using sock_socket. The check above achieves that + */ + if (self == PROC_NULL) + self = current_proc(); + + if (so->last_upid != proc_uniqueid(self) || + so->last_pid != proc_pid(self)) { + so->last_upid = proc_uniqueid(self); + so->last_pid = proc_pid(self); + proc_getexecutableuuid(self, so->last_uuid, + sizeof (so->last_uuid)); + } + } +} - s = splnet(); +void +so_update_policy(struct socket *so) +{ + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) + (void) inp_update_policy(sotoinpcb(so)); +} - while (p = socket_cache_tail) - { - if ((so_cache_time - p->cache_timestamp) < SO_CACHE_TIME_LIMIT) - break; +boolean_t +so_cache_timer(void) +{ + struct socket *p; + int n_freed = 0; + boolean_t rc = FALSE; + + lck_mtx_lock(so_cache_mtx); + so_cache_timeouts++; + so_cache_time = net_uptime(); + + while (!STAILQ_EMPTY(&so_cache_head)) { + VERIFY(cached_sock_count > 0); + p = STAILQ_FIRST(&so_cache_head); + if ((so_cache_time - p->cache_timestamp) < + SO_CACHE_TIME_LIMIT) + break; - so_cache_timeouts++; - - if (socket_cache_tail = p->cache_prev) - p->cache_prev->cache_next = 0; - if (--cached_sock_count == 0) - socket_cache_head = 0; + STAILQ_REMOVE_HEAD(&so_cache_head, so_cache_ent); + --cached_sock_count; - splx(s); + zfree(so_cache_zone, p); - zfree(so_cache_zone, (vm_offset_t) p); - - splnet(); - if (++n_freed >= SO_CACHE_MAX_FREE_BATCH) - { - so_cache_max_freed++; + if (++n_freed >= SO_CACHE_MAX_FREE_BATCH) { + so_cache_max_freed++; break; } } - splx(s); - timeout(so_cache_timer, NULL, (SO_CACHE_FLUSH_INTERVAL * hz)); - - (void) thread_funnel_set(network_flock, FALSE); + /* Schedule again if there is more to cleanup */ + if (!STAILQ_EMPTY(&so_cache_head)) + rc = TRUE; + lck_mtx_unlock(so_cache_mtx); + return (rc); } -#endif /* __APPLE__ */ /* * Get a socket structure from our zone, and initialize it. @@ -347,253 +498,415 @@ void so_cache_timer() * the protocols can be easily modified to do this. */ struct socket * -soalloc(waitok, dom, type) - int waitok; - int dom; - int type; +soalloc(int waitok, int dom, int type) { struct socket *so; - if ((dom == PF_INET) && (type == SOCK_STREAM)) - cached_sock_alloc(&so, waitok); - else - { - so = _MALLOC_ZONE(sizeof(*so), socket_zone, M_WAITOK); - if (so) - bzero(so, sizeof *so); + if ((dom == PF_INET) && (type == SOCK_STREAM)) { + cached_sock_alloc(&so, waitok); + } else { + MALLOC_ZONE(so, struct socket *, sizeof (*so), socket_zone, + M_WAITOK); + if (so != NULL) + bzero(so, sizeof (*so)); } - /* XXX race condition for reentrant kernel */ - - if (so) { - so->so_gencnt = ++so_gencnt; - so->so_zone = socket_zone; + if (so != NULL) { + so->so_gencnt = ++so_gencnt; + so->so_zone = socket_zone; +#if CONFIG_MACF_SOCKET + /* Convert waitok to M_WAITOK/M_NOWAIT for MAC Framework. */ + if (mac_socket_label_init(so, !waitok) != 0) { + sodealloc(so); + return (NULL); + } +#endif /* MAC_SOCKET */ } - return so; + return (so); } int -socreate(dom, aso, type, proto) - int dom; - struct socket **aso; - register int type; - int proto; +socreate_internal(int dom, struct socket **aso, int type, int proto, + struct proc *p, uint32_t flags, struct proc *ep) { - struct proc *p = current_proc(); - register struct protosw *prp; - register struct socket *so; - register int error = 0; + struct protosw *prp; + struct socket *so; + int error = 0; + #if TCPDEBUG extern int tcpconsdebug; #endif - if (proto) + + VERIFY(aso != NULL); + *aso = NULL; + + if (proto != 0) prp = pffindproto(dom, proto, type); else prp = pffindtype(dom, type); - if (prp == 0 || prp->pr_usrreqs->pru_attach == 0) - return (EPROTONOSUPPORT); -#ifndef __APPLE__ - - if (p->p_prison && jail_socket_unixiproute_only && - prp->pr_domain->dom_family != PF_LOCAL && - prp->pr_domain->dom_family != PF_INET && - prp->pr_domain->dom_family != PF_ROUTE) { + if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL) { + if (pffinddomain(dom) == NULL) + return (EAFNOSUPPORT); + if (proto != 0) { + if (pffindprotonotype(dom, proto) != NULL) + return (EPROTOTYPE); + } return (EPROTONOSUPPORT); } - -#endif if (prp->pr_type != type) return (EPROTOTYPE); - so = soalloc(p != 0, dom, type); - if (so == 0) + so = soalloc(1, dom, type); + if (so == NULL) return (ENOBUFS); + if (flags & SOCF_ASYNC) + so->so_state |= SS_NBIO; +#if MULTIPATH + if (flags & SOCF_MP_SUBFLOW) { + /* + * A multipath subflow socket is used internally in the kernel, + * therefore it does not have a file desciptor associated by + * default. + */ + so->so_state |= SS_NOFDREF; + so->so_flags |= SOF_MP_SUBFLOW; + } +#endif /* MULTIPATH */ + TAILQ_INIT(&so->so_incomp); TAILQ_INIT(&so->so_comp); so->so_type = type; + so->last_upid = proc_uniqueid(p); + so->last_pid = proc_pid(p); + proc_getexecutableuuid(p, so->last_uuid, sizeof (so->last_uuid)); + + if (ep != PROC_NULL && ep != p) { + so->e_upid = proc_uniqueid(ep); + so->e_pid = proc_pid(ep); + proc_getexecutableuuid(ep, so->e_uuid, sizeof (so->e_uuid)); + so->so_flags |= SOF_DELEGATED; + } -#ifdef __APPLE__ - if (p != 0) { - if (p->p_ucred->cr_uid == 0) - so->so_state = SS_PRIV; + so->so_cred = kauth_cred_proc_ref(p); + if (!suser(kauth_cred_get(), NULL)) + so->so_state |= SS_PRIV; - so->so_uid = p->p_ucred->cr_uid; - } -#else - so->so_cred = p->p_ucred; - crhold(so->so_cred); -#endif so->so_proto = prp; -#ifdef __APPLE__ - so->so_rcv.sb_flags |= SB_RECV; /* XXX */ - if (prp->pr_sfilter.tqh_first) - error = sfilter_init(so); - if (error == 0) -#endif - error = (*prp->pr_usrreqs->pru_attach)(so, proto, p); - if (error) { - /* - * Warning: - * If so_pcb is not zero, the socket will be leaked, - * so protocol attachment handler must be coded carefuly + so->so_rcv.sb_flags |= SB_RECV; + so->so_rcv.sb_so = so->so_snd.sb_so = so; + so->next_lock_lr = 0; + so->next_unlock_lr = 0; + +#if CONFIG_MACF_SOCKET + mac_socket_label_associate(kauth_cred_get(), so); +#endif /* MAC_SOCKET */ + + /* + * Attachment will create the per pcb lock if necessary and + * increase refcount for creation, make sure it's done before + * socket is inserted in lists. + */ + so->so_usecount++; + + error = (*prp->pr_usrreqs->pru_attach)(so, proto, p); + if (error != 0) { + /* + * Warning: + * If so_pcb is not zero, the socket will be leaked, + * so protocol attachment handler must be coded carefuly */ so->so_state |= SS_NOFDREF; - sofree(so); + so->so_usecount--; + sofreelastref(so, 1); /* will deallocate the socket */ return (error); } -#ifdef __APPLE__ - prp->pr_domain->dom_refs++; - so->so_rcv.sb_so = so->so_snd.sb_so = so; + + atomic_add_32(&prp->pr_domain->dom_refs, 1); TAILQ_INIT(&so->so_evlist); + + /* Attach socket filters for this protocol */ + sflt_initsock(so); #if TCPDEBUG if (tcpconsdebug == 2) so->so_options |= SO_DEBUG; #endif -#endif + so_set_default_traffic_class(so); + + /* + * If this thread or task is marked to create backgrounded sockets, + * mark the socket as background. + */ + if (proc_get_effective_thread_policy(current_thread(), TASK_POLICY_NEW_SOCKETS_BG)) { + socket_set_traffic_mgt_flags(so, TRAFFIC_MGT_SO_BACKGROUND); + so->so_background_thread = current_thread(); + } + + switch (dom) { + /* + * Don't mark Unix domain, system or multipath sockets as + * eligible for defunct by default. + */ + case PF_LOCAL: + case PF_SYSTEM: + case PF_MULTIPATH: + so->so_flags |= SOF_NODEFUNCT; + break; + default: + break; + } *aso = so; + return (0); } +/* + * Returns: 0 Success + * EAFNOSUPPORT + * EPROTOTYPE + * EPROTONOSUPPORT + * ENOBUFS + * :ENOBUFS[AF_UNIX] + * :ENOBUFS[TCP] + * :ENOMEM[TCP] + * :??? [other protocol families, IPSEC] + */ int -sobind(so, nam) - struct socket *so; - struct sockaddr *nam; +socreate(int dom, struct socket **aso, int type, int proto) +{ + return (socreate_internal(dom, aso, type, proto, current_proc(), 0, + PROC_NULL)); +} + +int +socreate_delegate(int dom, struct socket **aso, int type, int proto, pid_t epid) +{ + int error = 0; + struct proc *ep = PROC_NULL; + + if ((proc_selfpid() != epid) && ((ep = proc_find(epid)) == PROC_NULL)) { + error = ESRCH; + goto done; + } + error = socreate_internal(dom, aso, type, proto, current_proc(), 0, ep); + + /* + * It might not be wise to hold the proc reference when calling + * socreate_internal since it calls soalloc with M_WAITOK + */ +done: + if (ep != PROC_NULL) + proc_rele(ep); + + return (error); +} + +/* + * Returns: 0 Success + * :EINVAL Invalid argument [COMMON_START] + * :EAFNOSUPPORT Address family not supported + * :EADDRNOTAVAIL Address not available. + * :EINVAL Invalid argument + * :EAFNOSUPPORT Address family not supported [notdef] + * :EACCES Permission denied + * :EADDRINUSE Address in use + * :EAGAIN Resource unavailable, try again + * :EPERM Operation not permitted + * :??? + * :??? + * + * Notes: It's not possible to fully enumerate the return codes above, + * since socket filter authors and protocol family authors may + * not choose to limit their error returns to those listed, even + * though this may result in some software operating incorrectly. + * + * The error codes which are enumerated above are those known to + * be returned by the tcp_usr_bind function supplied. + */ +int +sobindlock(struct socket *so, struct sockaddr *nam, int dolock) { struct proc *p = current_proc(); - int error; - struct kextcb *kp; - int s = splnet(); + int error = 0; - error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, p); - if (error == 0) { - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_sobind) { - error = (*kp->e_soif->sf_sobind)(so, nam, kp); - if (error) { - if (error == EJUSTRETURN) { - error = 0; - break; - } - splx(s); - return(error); - } - } - kp = kp->e_next; - } + if (dolock) + socket_lock(so, 1); + VERIFY(so->so_usecount > 1); + + so_update_last_owner_locked(so, p); + so_update_policy(so); + + /* + * If this is a bind request on a socket that has been marked + * as inactive, reject it now before we go any further. + */ + if (so->so_flags & SOF_DEFUNCT) { + error = EINVAL; + SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] (%d)\n", + __func__, proc_pid(p), (uint64_t)VM_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error)); + goto out; } - splx(s); + + /* Socket filter */ + error = sflt_bind(so, nam); + + if (error == 0) + error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, p); +out: + if (dolock) + socket_unlock(so, 1); + + if (error == EJUSTRETURN) + error = 0; + return (error); } void -sodealloc(so) - struct socket *so; +sodealloc(struct socket *so) { + kauth_cred_unref(&so->so_cred); + + /* Remove any filters */ + sflt_termsock(so); + + /* Delete the state allocated for msg queues on a socket */ + if (so->so_flags & SOF_ENABLE_MSGS) { + FREE(so->so_msg_state, M_TEMP); + so->so_msg_state = NULL; + } + VERIFY(so->so_msg_state == NULL); + so->so_gencnt = ++so_gencnt; -#ifndef __APPLE__ - if (so->so_rcv.sb_hiwat) - (void)chgsbsize(so->so_cred->cr_uidinfo, - &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); - if (so->so_snd.sb_hiwat) - (void)chgsbsize(so->so_cred->cr_uidinfo, - &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); -#ifdef INET - if (so->so_accf != NULL) { - if (so->so_accf->so_accept_filter != NULL && - so->so_accf->so_accept_filter->accf_destroy != NULL) { - so->so_accf->so_accept_filter->accf_destroy(so); - } - if (so->so_accf->so_accept_filter_str != NULL) - FREE(so->so_accf->so_accept_filter_str, M_ACCF); - FREE(so->so_accf, M_ACCF); - } -#endif /* INET */ - crfree(so->so_cred); - zfreei(so->so_zone, so); -#else - if (so->cached_in_sock_layer == 1) - cached_sock_free(so); - else - _FREE_ZONE(so, sizeof(*so), so->so_zone); -#endif /* __APPLE__ */ +#if CONFIG_MACF_SOCKET + mac_socket_label_destroy(so); +#endif /* MAC_SOCKET */ + + if (so->cached_in_sock_layer) { + cached_sock_free(so); + } else { + FREE_ZONE(so, sizeof (*so), so->so_zone); + } } +/* + * Returns: 0 Success + * EINVAL + * EOPNOTSUPP + * :EINVAL[AF_UNIX] + * :EINVAL[TCP] + * :EADDRNOTAVAIL[TCP] Address not available. + * :EINVAL[TCP] Invalid argument + * :EAFNOSUPPORT[TCP] Address family not supported [notdef] + * :EACCES[TCP] Permission denied + * :EADDRINUSE[TCP] Address in use + * :EAGAIN[TCP] Resource unavailable, try again + * :EPERM[TCP] Operation not permitted + * :??? + * + * Notes: Other returns depend on the protocol family; all + * returns depend on what the filter author causes + * their filter to return. + */ int -solisten(so, backlog) - register struct socket *so; - int backlog; - +solisten(struct socket *so, int backlog) { - struct kextcb *kp; struct proc *p = current_proc(); - int s, error; + int error = 0; + + socket_lock(so, 1); + + so_update_last_owner_locked(so, p); + so_update_policy(so); + + if (so->so_proto == NULL) { + error = EINVAL; + goto out; + } + if ((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0) { + error = EOPNOTSUPP; + goto out; + } + + /* + * If the listen request is made on a socket that is not fully + * disconnected, or on a socket that has been marked as inactive, + * reject the request now. + */ + if ((so->so_state & + (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING)) || + (so->so_flags & SOF_DEFUNCT)) { + error = EINVAL; + if (so->so_flags & SOF_DEFUNCT) { + SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] " + "(%d)\n", __func__, proc_pid(p), + (uint64_t)VM_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error)); + } + goto out; + } + + if ((so->so_restrictions & SO_RESTRICT_DENY_IN) != 0) { + error = EPERM; + goto out; + } + + error = sflt_listen(so); + if (error == 0) + error = (*so->so_proto->pr_usrreqs->pru_listen)(so, p); - s = splnet(); - error = (*so->so_proto->pr_usrreqs->pru_listen)(so, p); if (error) { - splx(s); - return (error); + if (error == EJUSTRETURN) + error = 0; + goto out; } - if (TAILQ_EMPTY(&so->so_comp)) + + if (TAILQ_EMPTY(&so->so_comp)) so->so_options |= SO_ACCEPTCONN; - if (backlog < 0 || backlog > somaxconn) + /* + * POSIX: The implementation may have an upper limit on the length of + * the listen queue-either global or per accepting socket. If backlog + * exceeds this limit, the length of the listen queue is set to the + * limit. + * + * If listen() is called with a backlog argument value that is less + * than 0, the function behaves as if it had been called with a backlog + * argument value of 0. + * + * A backlog argument of 0 may allow the socket to accept connections, + * in which case the length of the listen queue may be set to an + * implementation-defined minimum value. + */ + if (backlog <= 0 || backlog > somaxconn) backlog = somaxconn; - so->so_qlimit = backlog; - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_solisten) { - error = (*kp->e_soif->sf_solisten)(so, kp); - if (error) { - if (error == EJUSTRETURN) { - error = 0; - break; - } - splx(s); - return(error); - } - } - kp = kp->e_next; - } - splx(s); - return (0); + so->so_qlimit = backlog; +out: + socket_unlock(so, 1); + return (error); } - void -sofree(so) - register struct socket *so; +sofreelastref(struct socket *so, int dealloc) { - int error; - struct kextcb *kp; struct socket *head = so->so_head; - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_sofree) { - error = (*kp->e_soif->sf_sofree)(so, kp); - if (error) { - selthreadclear(&so->so_snd.sb_sel); - selthreadclear(&so->so_rcv.sb_sel); - return; /* void fn */ - } - } - kp = kp->e_next; - } + /* Assume socket is locked */ - if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) { -#ifdef __APPLE__ + if (!(so->so_flags & SOF_PCBCLEARING) || !(so->so_state & SS_NOFDREF)) { selthreadclear(&so->so_snd.sb_sel); selthreadclear(&so->so_rcv.sb_sel); -#endif + so->so_rcv.sb_flags &= ~(SB_SEL|SB_UPCALL); + so->so_snd.sb_flags &= ~(SB_SEL|SB_UPCALL); + so->so_event = NULL; return; } if (head != NULL) { + socket_lock(head, 1); if (so->so_state & SS_INCOMP) { TAILQ_REMOVE(&head->so_incomp, so, so_list); head->so_incqlen--; @@ -604,10 +917,12 @@ sofree(so) * accept(2) may hang after select(2) indicated * that the listening socket was ready. */ -#ifdef __APPLE__ selthreadclear(&so->so_snd.sb_sel); selthreadclear(&so->so_rcv.sb_sel); -#endif + so->so_rcv.sb_flags &= ~(SB_SEL|SB_UPCALL); + so->so_snd.sb_flags &= ~(SB_SEL|SB_UPCALL); + so->so_event = NULL; + socket_unlock(head, 1); return; } else { panic("sofree: not queued"); @@ -615,68 +930,155 @@ sofree(so) head->so_qlen--; so->so_state &= ~SS_INCOMP; so->so_head = NULL; + socket_unlock(head, 1); } -#ifdef __APPLE__ - selthreadclear(&so->so_snd.sb_sel); - sbrelease(&so->so_snd); -#endif + sowflush(so); sorflush(so); - sfilter_term(so); - sodealloc(so); + +#if FLOW_DIVERT + if (so->so_flags & SOF_FLOW_DIVERT) { + flow_divert_detach(so); + } +#endif /* FLOW_DIVERT */ + + /* 3932268: disable upcall */ + so->so_rcv.sb_flags &= ~SB_UPCALL; + so->so_snd.sb_flags &= ~SB_UPCALL; + so->so_event = NULL; + + if (dealloc) + sodealloc(so); } -/* - * Close a socket on last file table reference removal. +void +soclose_wait_locked(struct socket *so) +{ + lck_mtx_t *mutex_held; + + if (so->so_proto->pr_getlock != NULL) + mutex_held = (*so->so_proto->pr_getlock)(so, 0); + else + mutex_held = so->so_proto->pr_domain->dom_mtx; + lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); + + /* + * Double check here and return if there's no outstanding upcall; + * otherwise proceed further only if SOF_UPCALLCLOSEWAIT is set. + */ + if (!so->so_upcallusecount || !(so->so_flags & SOF_UPCALLCLOSEWAIT)) + return; + so->so_rcv.sb_flags &= ~SB_UPCALL; + so->so_snd.sb_flags &= ~SB_UPCALL; + so->so_flags |= SOF_CLOSEWAIT; + (void) msleep((caddr_t)&so->so_upcallusecount, mutex_held, (PZERO - 1), + "soclose_wait_locked", NULL); + lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); + so->so_flags &= ~SOF_CLOSEWAIT; +} + +/* + * Close a socket on last file table reference removal. * Initiate disconnect if connected. * Free socket when disconnect complete. */ int -soclose(so) - register struct socket *so; +soclose_locked(struct socket *so) { - int s = splnet(); /* conservative */ int error = 0; - struct kextcb *kp; + lck_mtx_t *mutex_held; + struct timespec ts; -#ifndef __APPLE__ - funsetown(so->so_sigio); -#endif - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_soclose) { - error = (*kp->e_soif->sf_soclose)(so, kp); - if (error) { - splx(s); - return((error == EJUSTRETURN) ? 0 : error); - } - } - kp = kp->e_next; + if (so->so_usecount == 0) { + panic("soclose: so=%p refcount=0\n", so); + /* NOTREACHED */ } - if (so->so_options & SO_ACCEPTCONN) { + sflt_notify(so, sock_evt_closing, NULL); + + if (so->so_upcallusecount) + soclose_wait_locked(so); + + if ((so->so_options & SO_ACCEPTCONN)) { struct socket *sp, *sonext; + int socklock = 0; + + /* + * We do not want new connection to be added + * to the connection queues + */ + so->so_options &= ~SO_ACCEPTCONN; + + for (sp = TAILQ_FIRST(&so->so_incomp); + sp != NULL; sp = sonext) { + sonext = TAILQ_NEXT(sp, so_list); + + /* + * Radar 5350314 + * skip sockets thrown away by tcpdropdropblreq + * they will get cleanup by the garbage collection. + * otherwise, remove the incomp socket from the queue + * and let soabort trigger the appropriate cleanup. + */ + if (sp->so_flags & SOF_OVERFLOW) + continue; + + if (so->so_proto->pr_getlock != NULL) { + /* + * Lock ordering for consistency with the + * rest of the stack, we lock the socket + * first and then grabb the head. + */ + socket_unlock(so, 0); + socket_lock(sp, 1); + socket_lock(so, 0); + socklock = 1; + } + + TAILQ_REMOVE(&so->so_incomp, sp, so_list); + so->so_incqlen--; + + if (sp->so_state & SS_INCOMP) { + sp->so_state &= ~SS_INCOMP; + sp->so_head = NULL; + + (void) soabort(sp); + } - sp = TAILQ_FIRST(&so->so_incomp); - for (; sp != NULL; sp = sonext) { - sonext = TAILQ_NEXT(sp, so_list); - (void) soabort(sp); - } - for (sp = TAILQ_FIRST(&so->so_comp); sp != NULL; sp = sonext) { - sonext = TAILQ_NEXT(sp, so_list); - /* Dequeue from so_comp since sofree() won't do it */ - TAILQ_REMOVE(&so->so_comp, sp, so_list); - so->so_qlen--; - sp->so_state &= ~SS_COMP; - sp->so_head = NULL; - (void) soabort(sp); - } - - } - if (so->so_pcb == 0) + if (socklock) + socket_unlock(sp, 1); + } + + while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { + /* Dequeue from so_comp since sofree() won't do it */ + TAILQ_REMOVE(&so->so_comp, sp, so_list); + so->so_qlen--; + + if (so->so_proto->pr_getlock != NULL) { + socket_unlock(so, 0); + socket_lock(sp, 1); + } + + if (sp->so_state & SS_COMP) { + sp->so_state &= ~SS_COMP; + sp->so_head = NULL; + + (void) soabort(sp); + } + + if (so->so_proto->pr_getlock != NULL) { + socket_unlock(sp, 1); + socket_lock(so, 0); + } + } + } + if (so->so_pcb == NULL) { + /* 3915887: mark the socket as ready for dealloc */ + so->so_flags |= SOF_PCBCLEARING; goto discard; + } if (so->so_state & SS_ISCONNECTED) { if ((so->so_state & SS_ISDISCONNECTING) == 0) { - error = sodisconnect(so); + error = sodisconnectlocked(so); if (error) goto drop; } @@ -684,100 +1086,258 @@ soclose(so) if ((so->so_state & SS_ISDISCONNECTING) && (so->so_state & SS_NBIO)) goto drop; + if (so->so_proto->pr_getlock != NULL) + mutex_held = (*so->so_proto->pr_getlock)(so, 0); + else + mutex_held = so->so_proto->pr_domain->dom_mtx; while (so->so_state & SS_ISCONNECTED) { - error = tsleep((caddr_t)&so->so_timeo, - PSOCK | PCATCH, "soclos", so->so_linger); - if (error) + ts.tv_sec = (so->so_linger/100); + ts.tv_nsec = (so->so_linger % 100) * + NSEC_PER_USEC * 1000 * 10; + error = msleep((caddr_t)&so->so_timeo, + mutex_held, PSOCK | PCATCH, "soclose", &ts); + if (error) { + /* + * It's OK when the time fires, + * don't report an error + */ + if (error == EWOULDBLOCK) + error = 0; break; + } } } } drop: - if (so->so_pcb) { + if (so->so_usecount == 0) { + panic("soclose: usecount is zero so=%p\n", so); + /* NOTREACHED */ + } + if (so->so_pcb != NULL && !(so->so_flags & SOF_PCBCLEARING)) { + /* + * Let NetworkStatistics know this PCB is going away + * before we detach it. + */ + if (nstat_collect && + (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6)) + nstat_pcb_detach(so->so_pcb); + int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so); if (error == 0) error = error2; } + if (so->so_usecount <= 0) { + panic("soclose: usecount is zero so=%p\n", so); + /* NOTREACHED */ + } discard: - if (so->so_pcb && so->so_state & SS_NOFDREF) + if (so->so_pcb != NULL && !(so->so_flags & SOF_MP_SUBFLOW) && + (so->so_state & SS_NOFDREF)) { panic("soclose: NOFDREF"); + /* NOTREACHED */ + } so->so_state |= SS_NOFDREF; -#ifdef __APPLE__ - so->so_proto->pr_domain->dom_refs--; + + if (so->so_flags & SOF_MP_SUBFLOW) + so->so_flags &= ~SOF_MP_SUBFLOW; + + if ((so->so_flags & SOF_KNOTE) != 0) + KNOTE(&so->so_klist, SO_FILT_HINT_LOCKED); + + atomic_add_32(&so->so_proto->pr_domain->dom_refs, -1); evsofree(so); -#endif + + so->so_usecount--; sofree(so); - splx(s); + return (error); +} + +int +soclose(struct socket *so) +{ + int error = 0; + socket_lock(so, 1); + + if (so->so_retaincnt == 0) { + error = soclose_locked(so); + } else { + /* + * if the FD is going away, but socket is + * retained in kernel remove its reference + */ + so->so_usecount--; + if (so->so_usecount < 2) + panic("soclose: retaincnt non null and so=%p " + "usecount=%d\n", so, so->so_usecount); + } + socket_unlock(so, 1); return (error); } /* * Must be called at splnet... */ +/* Should already be locked */ int -soabort(so) - struct socket *so; +soabort(struct socket *so) { int error; - error = (*so->so_proto->pr_usrreqs->pru_abort)(so); - if (error) { - sofree(so); - return error; +#ifdef MORE_LOCKING_DEBUG + lck_mtx_t *mutex_held; + + if (so->so_proto->pr_getlock != NULL) + mutex_held = (*so->so_proto->pr_getlock)(so, 0); + else + mutex_held = so->so_proto->pr_domain->dom_mtx; + lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); +#endif + + if ((so->so_flags & SOF_ABORTED) == 0) { + so->so_flags |= SOF_ABORTED; + error = (*so->so_proto->pr_usrreqs->pru_abort)(so); + if (error) { + sofree(so); + return (error); + } } return (0); } int -soaccept(so, nam) - register struct socket *so; - struct sockaddr **nam; +soacceptlock(struct socket *so, struct sockaddr **nam, int dolock) { - int s = splnet(); int error; - struct kextcb *kp; + + if (dolock) + socket_lock(so, 1); + + so_update_last_owner_locked(so, PROC_NULL); + so_update_policy(so); if ((so->so_state & SS_NOFDREF) == 0) panic("soaccept: !NOFDREF"); so->so_state &= ~SS_NOFDREF; error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); - if (error == 0) { - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_soaccept) { - error = (*kp->e_soif->sf_soaccept)(so, nam, kp); - if (error) { - if (error == EJUSTRETURN) { - error = 0; - break; - } - splx(s); - return(error); - } - } - kp = kp->e_next; - } - } - - - splx(s); + + if (dolock) + socket_unlock(so, 1); return (error); } int -soconnect(so, nam) - register struct socket *so; - struct sockaddr *nam; +soaccept(struct socket *so, struct sockaddr **nam) +{ + return (soacceptlock(so, nam, 1)); +} + +int +soacceptfilter(struct socket *so) +{ + struct sockaddr *local = NULL, *remote = NULL; + int error = 0; + struct socket *head = so->so_head; + + /* + * Hold the lock even if this socket has not been made visible + * to the filter(s). For sockets with global locks, this protects + * against the head or peer going away + */ + socket_lock(so, 1); + if (sogetaddr_locked(so, &remote, 1) != 0 || + sogetaddr_locked(so, &local, 0) != 0) { + so->so_state &= ~(SS_NOFDREF | SS_COMP); + so->so_head = NULL; + socket_unlock(so, 1); + soclose(so); + /* Out of resources; try it again next time */ + error = ECONNABORTED; + goto done; + } + + error = sflt_accept(head, so, local, remote); + + /* + * If we get EJUSTRETURN from one of the filters, mark this socket + * as inactive and return it anyway. This newly accepted socket + * will be disconnected later before we hand it off to the caller. + */ + if (error == EJUSTRETURN) { + error = 0; + (void) sosetdefunct(current_proc(), so, + SHUTDOWN_SOCKET_LEVEL_DISCONNECT_INTERNAL, FALSE); + } + + if (error != 0) { + /* + * This may seem like a duplication to the above error + * handling part when we return ECONNABORTED, except + * the following is done while holding the lock since + * the socket has been exposed to the filter(s) earlier. + */ + so->so_state &= ~(SS_NOFDREF | SS_COMP); + so->so_head = NULL; + socket_unlock(so, 1); + soclose(so); + /* Propagate socket filter's error code to the caller */ + } else { + socket_unlock(so, 1); + } +done: + /* Callee checks for NULL pointer */ + sock_freeaddr(remote); + sock_freeaddr(local); + return (error); +} +/* + * Returns: 0 Success + * EOPNOTSUPP Operation not supported on socket + * EISCONN Socket is connected + * :EADDRNOTAVAIL Address not available. + * :EINVAL Invalid argument + * :EAFNOSUPPORT Address family not supported [notdef] + * :EACCES Permission denied + * :EADDRINUSE Address in use + * :EAGAIN Resource unavailable, try again + * :EPERM Operation not permitted + * :??? [anything a filter writer might set] + */ +int +soconnectlock(struct socket *so, struct sockaddr *nam, int dolock) { - int s; int error; struct proc *p = current_proc(); - struct kextcb *kp; - if (so->so_options & SO_ACCEPTCONN) - return (EOPNOTSUPP); - s = splnet(); + if (dolock) + socket_lock(so, 1); + + so_update_last_owner_locked(so, p); + so_update_policy(so); + + /* + * If this is a listening socket or if this is a previously-accepted + * socket that has been marked as inactive, reject the connect request. + */ + if ((so->so_options & SO_ACCEPTCONN) || (so->so_flags & SOF_DEFUNCT)) { + error = EOPNOTSUPP; + if (so->so_flags & SOF_DEFUNCT) { + SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] " + "(%d)\n", __func__, proc_pid(p), + (uint64_t)VM_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error)); + } + if (dolock) + socket_unlock(so, 1); + return (error); + } + + if ((so->so_restrictions & SO_RESTRICT_DENY_OUT) != 0) { + if (dolock) + socket_unlock(so, 1); + return (EPERM); + } + /* * If protocol is connection-based, can only connect once. * Otherwise, if connected, try to disconnect first. @@ -786,72 +1346,118 @@ soconnect(so, nam) */ if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && ((so->so_proto->pr_flags & PR_CONNREQUIRED) || - (error = sodisconnect(so)))) + (error = sodisconnectlocked(so)))) { error = EISCONN; - else { - /* - * Run connect filter before calling protocol: - * - non-blocking connect returns before completion; - * - allows filters to modify address. - */ - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_soconnect) { - error = (*kp->e_soif->sf_soconnect)(so, nam, kp); - if (error) { - if (error == EJUSTRETURN) { - error = 0; - } - splx(s); - return(error); - } - } - kp = kp->e_next; - } - error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, p); - } - splx(s); + } else { + /* + * Run connect filter before calling protocol: + * - non-blocking connect returns before completion; + */ + error = sflt_connectout(so, nam); + if (error != 0) { + if (error == EJUSTRETURN) + error = 0; + } else { + error = (*so->so_proto->pr_usrreqs->pru_connect) + (so, nam, p); + } + } + if (dolock) + socket_unlock(so, 1); return (error); } int -soconnect2(so1, so2) - register struct socket *so1; - struct socket *so2; +soconnect(struct socket *so, struct sockaddr *nam) +{ + return (soconnectlock(so, nam, 1)); +} + +/* + * Returns: 0 Success + * :EINVAL[AF_UNIX] + * :EPROTOTYPE[AF_UNIX] + * :??? [other protocol families] + * + * Notes: is not supported by [TCP]. + */ +int +soconnect2(struct socket *so1, struct socket *so2) { - int s = splnet(); int error; - struct kextcb *kp; + + socket_lock(so1, 1); + if (so2->so_proto->pr_lock) + socket_lock(so2, 1); error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); - if (error == 0) { - kp = sotokextcb(so1); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_soconnect2) { - error = (*kp->e_soif->sf_soconnect2)(so1, so2, kp); - if (error) { - if (error == EJUSTRETURN) { - return 0; - break; - } - splx(s); - return(error); - } - } - kp = kp->e_next; + + socket_unlock(so1, 1); + if (so2->so_proto->pr_lock) + socket_unlock(so2, 1); + return (error); +} + +int +soconnectxlocked(struct socket *so, struct sockaddr_list **src_sl, + struct sockaddr_list **dst_sl, struct proc *p, uint32_t ifscope, + associd_t aid, connid_t *pcid, uint32_t flags, void *arg, + uint32_t arglen) +{ + int error; + + /* + * If this is a listening socket or if this is a previously-accepted + * socket that has been marked as inactive, reject the connect request. + */ + if ((so->so_options & SO_ACCEPTCONN) || (so->so_flags & SOF_DEFUNCT)) { + error = EOPNOTSUPP; + if (so->so_flags & SOF_DEFUNCT) { + SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] " + "(%d)\n", __func__, proc_pid(p), + (uint64_t)VM_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error)); + } + return (error); + } + + if ((so->so_restrictions & SO_RESTRICT_DENY_OUT) != 0) + return (EPERM); + + /* + * If protocol is connection-based, can only connect once + * unless PR_MULTICONN is set. Otherwise, if connected, + * try to disconnect first. This allows user to disconnect + * by connecting to, e.g., a null address. + */ + if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) && + !(so->so_proto->pr_flags & PR_MULTICONN) && + ((so->so_proto->pr_flags & PR_CONNREQUIRED) || + (error = sodisconnectlocked(so)) != 0)) { + error = EISCONN; + } else { + /* + * Run connect filter before calling protocol: + * - non-blocking connect returns before completion; + */ + error = sflt_connectxout(so, dst_sl); + if (error != 0) { + if (error == EJUSTRETURN) + error = 0; + } else { + error = (*so->so_proto->pr_usrreqs->pru_connectx) + (so, src_sl, dst_sl, p, ifscope, aid, pcid, + flags, arg, arglen); } } - splx(s); + return (error); } int -sodisconnect(so) - register struct socket *so; +sodisconnectlocked(struct socket *so) { - int s = splnet(); int error; - struct kextcb *kp; if ((so->so_state & SS_ISCONNECTED) == 0) { error = ENOTCONN; @@ -861,31 +1467,174 @@ sodisconnect(so) error = EALREADY; goto bad; } + error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); + if (error == 0) + sflt_notify(so, sock_evt_disconnected, NULL); + +bad: + return (error); +} + +/* Locking version */ +int +sodisconnect(struct socket *so) +{ + int error; + + socket_lock(so, 1); + error = sodisconnectlocked(so); + socket_unlock(so, 1); + return (error); +} + +int +sodisconnectxlocked(struct socket *so, associd_t aid, connid_t cid) +{ + int error; + + /* + * Call the protocol disconnectx handler; let it handle all + * matters related to the connection state of this session. + */ + error = (*so->so_proto->pr_usrreqs->pru_disconnectx)(so, aid, cid); if (error == 0) { - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_sodisconnect) { - error = (*kp->e_soif->sf_sodisconnect)(so, kp); - if (error) { - if (error == EJUSTRETURN) { - error = 0; - break; - } - splx(s); - return(error); - } + /* + * The event applies only for the session, not for + * the disconnection of individual subflows. + */ + if (so->so_state & (SS_ISDISCONNECTING|SS_ISDISCONNECTED)) + sflt_notify(so, sock_evt_disconnected, NULL); + } + return (error); +} + +int +sodisconnectx(struct socket *so, associd_t aid, connid_t cid) +{ + int error; + + socket_lock(so, 1); + error = sodisconnectxlocked(so, aid, cid); + socket_unlock(so, 1); + return (error); +} + +int +sopeelofflocked(struct socket *so, associd_t aid, struct socket **psop) +{ + return ((*so->so_proto->pr_usrreqs->pru_peeloff)(so, aid, psop)); +} + +#define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) + +/* + * sosendcheck will lock the socket buffer if it isn't locked and + * verify that there is space for the data being inserted. + * + * Returns: 0 Success + * EPIPE + * sblock:EWOULDBLOCK + * sblock:EINTR + * sbwait:EBADF + * sbwait:EINTR + * [so_error]:??? + */ +int +sosendcheck(struct socket *so, struct sockaddr *addr, user_ssize_t resid, + int32_t clen, int32_t atomic, int flags, int *sblocked, + struct mbuf *control) +{ + int error = 0; + int32_t space; + int assumelock = 0; + +restart: + if (*sblocked == 0) { + if ((so->so_snd.sb_flags & SB_LOCK) != 0 && + so->so_send_filt_thread != 0 && + so->so_send_filt_thread == current_thread()) { + /* + * We're being called recursively from a filter, + * allow this to continue. Radar 4150520. + * Don't set sblocked because we don't want + * to perform an unlock later. + */ + assumelock = 1; + } else { + error = sblock(&so->so_snd, SBLOCKWAIT(flags)); + if (error) { + if (so->so_flags & SOF_DEFUNCT) + goto defunct; + return (error); } - kp = kp->e_next; + *sblocked = 1; } } -bad: - splx(s); - return (error); + /* + * If a send attempt is made on a socket that has been marked + * as inactive (disconnected), reject the request. + */ + if (so->so_flags & SOF_DEFUNCT) { +defunct: + error = EPIPE; + SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] (%d)\n", + __func__, proc_selfpid(), (uint64_t)VM_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error)); + return (error); + } + + if (so->so_state & SS_CANTSENDMORE) + return (EPIPE); + + if (so->so_error) { + error = so->so_error; + so->so_error = 0; + return (error); + } + + if ((so->so_state & SS_ISCONNECTED) == 0) { + if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { + if ((so->so_state & SS_ISCONFIRMING) == 0 && + !(resid == 0 && clen != 0)) + return (ENOTCONN); + } else if (addr == 0 && !(flags&MSG_HOLD)) { + return ((so->so_proto->pr_flags & PR_CONNREQUIRED) ? + ENOTCONN : EDESTADDRREQ); + } + } + if (so->so_flags & SOF_ENABLE_MSGS) + space = msgq_sbspace(so, control); + else + space = sbspace(&so->so_snd); + + if (flags & MSG_OOB) + space += 1024; + if ((atomic && resid > so->so_snd.sb_hiwat) || + clen > so->so_snd.sb_hiwat) + return (EMSGSIZE); + + if ((space < resid + clen && + (atomic || space < (int32_t)so->so_snd.sb_lowat || space < clen)) || + (so->so_type == SOCK_STREAM && so_wait_for_if_feedback(so))) { + if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO) || + assumelock) { + return (EWOULDBLOCK); + } + sbunlock(&so->so_snd, TRUE); /* keep socket locked */ + *sblocked = 0; + error = sbwait(&so->so_snd); + if (error) { + if (so->so_flags & SOF_DEFUNCT) + goto defunct; + return (error); + } + goto restart; + } + return (0); } -#define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_DONTWAIT : M_WAIT) /* * Send on a socket. * If send must go all at once and message is larger than @@ -906,309 +1655,403 @@ bad: * MSG_HOLD: go thru most of sosend(), but just enqueue the mbuf * MSG_SEND: go thru as for MSG_HOLD on current fragment, then * point at the mbuf chain being constructed and go from there. + * + * Returns: 0 Success + * EOPNOTSUPP + * EINVAL + * ENOBUFS + * uiomove:EFAULT + * sosendcheck:EPIPE + * sosendcheck:EWOULDBLOCK + * sosendcheck:EINTR + * sosendcheck:EBADF + * sosendcheck:EINTR + * sosendcheck:??? [value from so_error] + * :ECONNRESET[TCP] + * :EINVAL[TCP] + * :ENOBUFS[TCP] + * :EADDRINUSE[TCP] + * :EADDRNOTAVAIL[TCP] + * :EAFNOSUPPORT[TCP] + * :EACCES[TCP] + * :EAGAIN[TCP] + * :EPERM[TCP] + * :EMSGSIZE[TCP] + * :EHOSTUNREACH[TCP] + * :ENETUNREACH[TCP] + * :ENETDOWN[TCP] + * :ENOMEM[TCP] + * :ENOBUFS[TCP] + * :???[TCP] [ignorable: mostly IPSEC/firewall/DLIL] + * :EINVAL[AF_UNIX] + * :EOPNOTSUPP[AF_UNIX] + * :EPIPE[AF_UNIX] + * :ENOTCONN[AF_UNIX] + * :EISCONN[AF_UNIX] + * :???[AF_UNIX] [whatever a filter author chooses] + * :??? [whatever a filter author chooses] + * + * Notes: Other returns depend on the protocol family; all + * returns depend on what the filter author causes + * their filter to return. */ int -sosend(so, addr, uio, top, control, flags) - register struct socket *so; - struct sockaddr *addr; - struct uio *uio; - struct mbuf *top; - struct mbuf *control; - int flags; - +sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, + struct mbuf *top, struct mbuf *control, int flags) { struct mbuf **mp; - register struct mbuf *m, *freelist = NULL; - register long space, len, resid; - int clen = 0, error, s, dontroute, mlen, sendflags; + struct mbuf *m, *freelist = NULL; + user_ssize_t space, len, resid; + int clen = 0, error, dontroute, mlen, sendflags; int atomic = sosendallatonce(so) || top; + int sblocked = 0; struct proc *p = current_proc(); - struct kextcb *kp; + struct mbuf *control_copy = NULL; - if (uio) - resid = uio->uio_resid; + if (uio != NULL) + resid = uio_resid(uio); else resid = top->m_pkthdr.len; - KERNEL_DEBUG((DBG_FNC_SOSEND | DBG_FUNC_START), - so, - resid, - so->so_snd.sb_cc, - so->so_snd.sb_lowat, - so->so_snd.sb_hiwat); + KERNEL_DEBUG((DBG_FNC_SOSEND | DBG_FUNC_START), so, resid, + so->so_snd.sb_cc, so->so_snd.sb_lowat, so->so_snd.sb_hiwat); - /* + socket_lock(so, 1); + so_update_last_owner_locked(so, p); + so_update_policy(so); + + if (so->so_type != SOCK_STREAM && (flags & MSG_OOB) != 0) { + error = EOPNOTSUPP; + socket_unlock(so, 1); + goto out; + } + + /* * In theory resid should be unsigned. * However, space must be signed, as it might be less than 0 * if we over-committed, and we must use a signed comparison * of space and resid. On the other hand, a negative resid * causes us to loop sending 0-length segments to the protocol. * - * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM - * type sockets since that's an error. + * Usually, MSG_EOR isn't used on SOCK_STREAM type sockets. + * But it will be used by sockets doing message delivery. + * + * Note: We limit resid to be a positive 32 bits value as we use + * imin() to set bytes_to_copy -- radr://14558484 */ - if (resid < 0 || so->so_type == SOCK_STREAM && (flags & MSG_EOR)) { + if ((int32_t)resid < 0 || (so->so_type == SOCK_STREAM && + !(so->so_flags & SOF_ENABLE_MSGS) && (flags & MSG_EOR))) { error = EINVAL; + socket_unlock(so, 1); goto out; } - dontroute = - (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && + dontroute = (flags & MSG_DONTROUTE) && + (so->so_options & SO_DONTROUTE) == 0 && (so->so_proto->pr_flags & PR_ATOMIC); - if (p) - p->p_stats->p_ru.ru_msgsnd++; - if (control) + OSIncrementAtomicLong(&p->p_stats->p_ru.ru_msgsnd); + + if (control != NULL) clen = control->m_len; -#define snderr(errno) { error = errno; splx(s); goto release; } -restart: - error = sblock(&so->so_snd, SBLOCKWAIT(flags)); - if (error) - goto out; do { - s = splnet(); - if (so->so_state & SS_CANTSENDMORE) - snderr(EPIPE); - if (so->so_error) { - error = so->so_error; - so->so_error = 0; - splx(s); + error = sosendcheck(so, addr, resid, clen, atomic, flags, + &sblocked, control); + if (error) goto release; - } - if ((so->so_state & SS_ISCONNECTED) == 0) { - /* - * `sendto' and `sendmsg' is allowed on a connection- - * based socket if it supports implied connect. - * Return ENOTCONN if not connected and no address is - * supplied. - */ - if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && - (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { - if ((so->so_state & SS_ISCONFIRMING) == 0 && - !(resid == 0 && clen != 0)) - snderr(ENOTCONN); - } else if (addr == 0 && !(flags&MSG_HOLD)) - snderr(so->so_proto->pr_flags & PR_CONNREQUIRED ? - ENOTCONN : EDESTADDRREQ); - } - space = sbspace(&so->so_snd); - if (flags & MSG_OOB) - space += 1024; - if ((atomic && resid > so->so_snd.sb_hiwat) || - clen > so->so_snd.sb_hiwat) - snderr(EMSGSIZE); - if (space < resid + clen && - (atomic || space < so->so_snd.sb_lowat || space < clen)) { - if (so->so_state & SS_NBIO) - snderr(EWOULDBLOCK); - sbunlock(&so->so_snd); - error = sbwait(&so->so_snd); - splx(s); - if (error) - goto out; - goto restart; - } - splx(s); + mp = ⊤ - space -= clen; + if (so->so_flags & SOF_ENABLE_MSGS) + space = msgq_sbspace(so, control); + else + space = sbspace(&so->so_snd) - clen; + space += ((flags & MSG_OOB) ? 1024 : 0); do { - if (uio == NULL) { - /* - * Data is prepackaged in "top". - */ - resid = 0; - if (flags & MSG_EOR) - top->m_flags |= M_EOR; - } else { - boolean_t dropped_funnel = FALSE; - int chainlength; - int bytes_to_copy; - - bytes_to_copy = min(resid, space); - - if (sosendminchain > 0) { - if (bytes_to_copy >= sosendminchain) { - dropped_funnel = TRUE; - (void)thread_funnel_set(network_flock, FALSE); - } - chainlength = 0; - } else - chainlength = sosendmaxchain; + if (uio == NULL) { + /* + * Data is prepackaged in "top". + */ + resid = 0; + if (flags & MSG_EOR) + top->m_flags |= M_EOR; + } else { + int chainlength; + int bytes_to_copy; + boolean_t jumbocl; + + bytes_to_copy = imin(resid, space); + + if (sosendminchain > 0) + chainlength = 0; + else + chainlength = sosendmaxchain; + + /* + * Attempt to use larger than system page-size + * clusters for large writes only if there is + * a jumbo cluster pool and if the socket is + * marked accordingly. + */ + jumbocl = sosendjcl && njcl > 0 && + ((so->so_flags & SOF_MULTIPAGES) || + sosendjcl_ignore_capab); + + socket_unlock(so, 0); + + do { + int num_needed; + int hdrs_needed = (top == NULL) ? 1 : 0; + + /* + * try to maintain a local cache of mbuf + * clusters needed to complete this + * write the list is further limited to + * the number that are currently needed + * to fill the socket this mechanism + * allows a large number of mbufs/ + * clusters to be grabbed under a single + * mbuf lock... if we can't get any + * clusters, than fall back to trying + * for mbufs if we fail early (or + * miscalcluate the number needed) make + * sure to release any clusters we + * haven't yet consumed. + */ + if (freelist == NULL && + bytes_to_copy > MBIGCLBYTES && + jumbocl) { + num_needed = + bytes_to_copy / M16KCLBYTES; + + if ((bytes_to_copy - + (num_needed * M16KCLBYTES)) + >= MINCLSIZE) + num_needed++; + + freelist = + m_getpackets_internal( + (unsigned int *)&num_needed, + hdrs_needed, M_WAIT, 0, + M16KCLBYTES); + /* + * Fall back to 4K cluster size + * if allocation failed + */ + } - do { + if (freelist == NULL && + bytes_to_copy > MCLBYTES) { + num_needed = + bytes_to_copy / MBIGCLBYTES; + + if ((bytes_to_copy - + (num_needed * MBIGCLBYTES)) >= + MINCLSIZE) + num_needed++; + + freelist = + m_getpackets_internal( + (unsigned int *)&num_needed, + hdrs_needed, M_WAIT, 0, + MBIGCLBYTES); + /* + * Fall back to cluster size + * if allocation failed + */ + } - if (bytes_to_copy >= MINCLSIZE) { - /* - * try to maintain a local cache of mbuf clusters needed to complete this write - * the list is further limited to the number that are currently needed to fill the socket - * this mechanism allows a large number of mbufs/clusters to be grabbed under a single - * mbuf lock... if we can't get any clusters, than fall back to trying for mbufs - * if we fail early (or miscalcluate the number needed) make sure to release any clusters - * we haven't yet consumed. - */ - if ((m = freelist) == NULL) { - int num_needed; - int hdrs_needed = 0; - - if (top == 0) - hdrs_needed = 1; - num_needed = bytes_to_copy / MCLBYTES; - - if ((bytes_to_copy - (num_needed * MCLBYTES)) >= MINCLSIZE) - num_needed++; - - if ((freelist = m_getpackets(num_needed, hdrs_needed, M_WAIT)) == NULL) - goto getpackets_failed; - m = freelist; - } - freelist = m->m_next; - m->m_next = NULL; - - mlen = MCLBYTES; - len = min(mlen, bytes_to_copy); - } else { -getpackets_failed: - if (top == 0) { - MGETHDR(m, M_WAIT, MT_DATA); - mlen = MHLEN; - m->m_pkthdr.len = 0; - m->m_pkthdr.rcvif = (struct ifnet *)0; - } else { - MGET(m, M_WAIT, MT_DATA); - mlen = MLEN; - } - len = min(mlen, bytes_to_copy); - /* - * For datagram protocols, leave room - * for protocol headers in first mbuf. - */ - if (atomic && top == 0 && len < mlen) - MH_ALIGN(m, len); + if (freelist == NULL && + bytes_to_copy > MINCLSIZE) { + num_needed = + bytes_to_copy / MCLBYTES; + + if ((bytes_to_copy - + (num_needed * MCLBYTES)) >= + MINCLSIZE) + num_needed++; + + freelist = + m_getpackets_internal( + (unsigned int *)&num_needed, + hdrs_needed, M_WAIT, 0, + MCLBYTES); + /* + * Fall back to a single mbuf + * if allocation failed + */ + } + + if (freelist == NULL) { + if (top == NULL) + MGETHDR(freelist, + M_WAIT, MT_DATA); + else + MGET(freelist, + M_WAIT, MT_DATA); + + if (freelist == NULL) { + error = ENOBUFS; + socket_lock(so, 0); + goto release; + } + /* + * For datagram protocols, + * leave room for protocol + * headers in first mbuf. + */ + if (atomic && top == NULL && + bytes_to_copy < MHLEN) { + MH_ALIGN(freelist, + bytes_to_copy); + } + } + m = freelist; + freelist = m->m_next; + m->m_next = NULL; + + if ((m->m_flags & M_EXT)) + mlen = m->m_ext.ext_size; + else if ((m->m_flags & M_PKTHDR)) + mlen = + MHLEN - m_leadingspace(m); + else + mlen = MLEN; + len = imin(mlen, bytes_to_copy); + + chainlength += len; + + space -= len; + + error = uiomove(mtod(m, caddr_t), + len, uio); + + resid = uio_resid(uio); + + m->m_len = len; + *mp = m; + top->m_pkthdr.len += len; + if (error) + break; + mp = &m->m_next; + if (resid <= 0) { + if (flags & MSG_EOR) + top->m_flags |= M_EOR; + break; + } + bytes_to_copy = min(resid, space); + + } while (space > 0 && + (chainlength < sosendmaxchain || atomic || + resid < MINCLSIZE)); + + socket_lock(so, 0); + + if (error) + goto release; } - chainlength += len; - - space -= len; - - error = uiomove(mtod(m, caddr_t), (int)len, uio); - - resid = uio->uio_resid; - - m->m_len = len; - *mp = m; - top->m_pkthdr.len += len; - if (error) - break; - mp = &m->m_next; - if (resid <= 0) { - if (flags & MSG_EOR) - top->m_flags |= M_EOR; - break; + + if (flags & (MSG_HOLD|MSG_SEND)) { + /* Enqueue for later, go away if HOLD */ + struct mbuf *mb1; + if (so->so_temp && (flags & MSG_FLUSH)) { + m_freem(so->so_temp); + so->so_temp = NULL; + } + if (so->so_temp) + so->so_tail->m_next = top; + else + so->so_temp = top; + mb1 = top; + while (mb1->m_next) + mb1 = mb1->m_next; + so->so_tail = mb1; + if (flags & MSG_HOLD) { + top = NULL; + goto release; + } + top = so->so_temp; } - bytes_to_copy = min(resid, space); + if (dontroute) + so->so_options |= SO_DONTROUTE; - } while (space > 0 && (chainlength < sosendmaxchain || atomic || resid < MINCLSIZE)); + /* Compute flags here, for pru_send and NKEs */ + sendflags = (flags & MSG_OOB) ? PRUS_OOB : + /* + * If the user set MSG_EOF, the protocol + * understands this flag and nothing left to + * send then use PRU_SEND_EOF instead of PRU_SEND. + */ + ((flags & MSG_EOF) && + (so->so_proto->pr_flags & PR_IMPLOPCL) && + (resid <= 0)) ? PRUS_EOF : + /* If there is more to send set PRUS_MORETOCOME */ + (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; + + /* + * Socket filter processing + */ + error = sflt_data_out(so, addr, &top, + &control, (sendflags & MSG_OOB) ? + sock_data_filt_flag_oob : 0); + if (error) { + if (error == EJUSTRETURN) { + error = 0; + clen = 0; + control = NULL; + top = NULL; + } - if (dropped_funnel == TRUE) - (void)thread_funnel_set(network_flock, TRUE); - if (error) - goto release; - } - - if (flags & (MSG_HOLD|MSG_SEND)) - { /* Enqueue for later, go away if HOLD */ - register struct mbuf *mb1; - if (so->so_temp && (flags & MSG_FLUSH)) - { m_freem(so->so_temp); - so->so_temp = NULL; - } - if (so->so_temp) - so->so_tail->m_next = top; - else - so->so_temp = top; - mb1 = top; - while (mb1->m_next) - mb1 = mb1->m_next; - so->so_tail = mb1; - if (flags&MSG_HOLD) - { top = NULL; goto release; } - top = so->so_temp; - } - if (dontroute) - so->so_options |= SO_DONTROUTE; - s = splnet(); /* XXX */ - /* Compute flags here, for pru_send and NKEs */ - sendflags = (flags & MSG_OOB) ? PRUS_OOB : /* - * If the user set MSG_EOF, the protocol - * understands this flag and nothing left to - * send then use PRU_SEND_EOF instead of PRU_SEND. + * End Socket filter processing */ - ((flags & MSG_EOF) && - (so->so_proto->pr_flags & PR_IMPLOPCL) && - (resid <= 0)) ? - PRUS_EOF : - /* If there is more to send set PRUS_MORETOCOME */ - (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; - kp = sotokextcb(so); - while (kp) - { if (kp->e_soif && kp->e_soif->sf_sosend) { - error = (*kp->e_soif->sf_sosend)(so, &addr, - &uio, &top, - &control, - &sendflags, - kp); - if (error) { - splx(s); - if (error == EJUSTRETURN) { - sbunlock(&so->so_snd); - - if (freelist) - m_freem_list(freelist); - return(0); - } - goto release; - } + + if (so->so_flags & SOF_ENABLE_MSGS) { + /* + * Make a copy of control mbuf, + * so that msg priority can be + * passed to subsequent mbufs. + */ + control_copy = m_dup(control, M_NOWAIT); } - kp = kp->e_next; - } - - error = (*so->so_proto->pr_usrreqs->pru_send)(so, - sendflags, top, addr, control, p); - splx(s); -#ifdef __APPLE__ - if (flags & MSG_SEND) - so->so_temp = NULL; -#endif - if (dontroute) - so->so_options &= ~SO_DONTROUTE; - clen = 0; - control = 0; - top = 0; - mp = ⊤ - if (error) - goto release; + error = (*so->so_proto->pr_usrreqs->pru_send) + (so, sendflags, top, addr, control, p); + + if (flags & MSG_SEND) + so->so_temp = NULL; + + if (dontroute) + so->so_options &= ~SO_DONTROUTE; + + clen = 0; + control = control_copy; + control_copy = NULL; + top = NULL; + mp = ⊤ + if (error) + goto release; } while (resid && space > 0); } while (resid); release: - sbunlock(&so->so_snd); + if (sblocked) + sbunlock(&so->so_snd, FALSE); /* will unlock socket */ + else + socket_unlock(so, 1); out: - if (top) + if (top != NULL) m_freem(top); - if (control) + if (control != NULL) m_freem(control); - if (freelist) - m_freem_list(freelist); + if (freelist != NULL) + m_freem_list(freelist); + if (control_copy != NULL) + m_freem(control_copy); - KERNEL_DEBUG(DBG_FNC_SOSEND | DBG_FUNC_END, - so, - resid, - so->so_snd.sb_cc, - space, - error); + KERNEL_DEBUG(DBG_FNC_SOSEND | DBG_FUNC_END, so, resid, so->so_snd.sb_cc, + space, error); return (error); } @@ -1228,118 +2071,183 @@ out: * The caller may receive the data as a single mbuf chain by supplying * an mbuf **mp0 for use in returning the chain. The uio is then used * only for the count in uio_resid. + * + * Returns: 0 Success + * ENOBUFS + * ENOTCONN + * EWOULDBLOCK + * uiomove:EFAULT + * sblock:EWOULDBLOCK + * sblock:EINTR + * sbwait:EBADF + * sbwait:EINTR + * sodelayed_copy:EFAULT + * :EINVAL[TCP] + * :EWOULDBLOCK[TCP] + * :??? + * dom_externalize>:EMSGSIZE[AF_UNIX] + * dom_externalize>:ENOBUFS[AF_UNIX] + * dom_externalize>:??? + * + * Notes: Additional return values from calls through and + * dom_externalize> depend on protocols other than + * TCP or AF_UNIX, which are documented above. */ int -soreceive(so, psa, uio, mp0, controlp, flagsp) - register struct socket *so; - struct sockaddr **psa; - struct uio *uio; - struct mbuf **mp0; - struct mbuf **controlp; - int *flagsp; -{ - register struct mbuf *m, **mp, *ml; - register int flags, len, error, s, offset; +soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, + struct mbuf **mp0, struct mbuf **controlp, int *flagsp) +{ + struct mbuf *m, **mp, *ml = NULL; + struct mbuf *nextrecord, *free_list; + int flags, error, offset; + user_ssize_t len; struct protosw *pr = so->so_proto; - struct mbuf *nextrecord; - int moff, type = 0; - int orig_resid = uio->uio_resid; - struct kextcb *kp; - volatile struct mbuf *free_list; - volatile int delayed_copy_len; + int moff, type =0; + user_ssize_t orig_resid = uio_resid(uio); + user_ssize_t delayed_copy_len; int can_delay; int need_event; struct proc *p = current_proc(); + KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_START, so, uio_resid(uio), + so->so_rcv.sb_cc, so->so_rcv.sb_lowat, so->so_rcv.sb_hiwat); - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_START, - so, - uio->uio_resid, - so->so_rcv.sb_cc, - so->so_rcv.sb_lowat, - so->so_rcv.sb_hiwat); + socket_lock(so, 1); + so_update_last_owner_locked(so, p); + so_update_policy(so); - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_soreceive) { - error = (*kp->e_soif->sf_soreceive)(so, psa, &uio, - mp0, controlp, - flagsp, kp); - if (error) { - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error,0,0,0,0); - return((error == EJUSTRETURN) ? 0 : error); - } - } - kp = kp->e_next; +#ifdef MORE_LOCKING_DEBUG + if (so->so_usecount == 1) { + panic("%s: so=%x no other reference on socket\n", __func__, so); + /* NOTREACHED */ } - +#endif mp = mp0; - if (psa) - *psa = 0; - if (controlp) - *controlp = 0; - if (flagsp) + if (psa != NULL) + *psa = NULL; + if (controlp != NULL) + *controlp = NULL; + if (flagsp != NULL) flags = *flagsp &~ MSG_EOR; else flags = 0; - /* - * When SO_WANTOOBFLAG is set we try to get out-of-band data - * regardless of the flags argument. Here is the case were - * out-of-band data is not inline. - */ - if ((flags & MSG_OOB) || - ((so->so_options & SO_WANTOOBFLAG) != 0 && - (so->so_options & SO_OOBINLINE) == 0 && - (so->so_oobmark || (so->so_state & SS_RCVATMARK)))) { + + /* + * If a recv attempt is made on a previously-accepted socket + * that has been marked as inactive (disconnected), reject + * the request. + */ + if (so->so_flags & SOF_DEFUNCT) { + struct sockbuf *sb = &so->so_rcv; + + error = ENOTCONN; + SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] (%d)\n", + __func__, proc_pid(p), (uint64_t)VM_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error)); + /* + * This socket should have been disconnected and flushed + * prior to being returned from sodefunct(); there should + * be no data on its receive list, so panic otherwise. + */ + if (so->so_state & SS_DEFUNCT) + sb_empty_assert(sb, __func__); + socket_unlock(so, 1); + return (error); + } + + /* + * When SO_WANTOOBFLAG is set we try to get out-of-band data + * regardless of the flags argument. Here is the case were + * out-of-band data is not inline. + */ + if ((flags & MSG_OOB) || + ((so->so_options & SO_WANTOOBFLAG) != 0 && + (so->so_options & SO_OOBINLINE) == 0 && + (so->so_oobmark || (so->so_state & SS_RCVATMARK)))) { m = m_get(M_WAIT, MT_DATA); if (m == NULL) { - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, ENOBUFS,0,0,0,0); + socket_unlock(so, 1); + KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, + ENOBUFS, 0, 0, 0, 0); return (ENOBUFS); } error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); if (error) goto bad; + socket_unlock(so, 0); do { error = uiomove(mtod(m, caddr_t), - (int) min(uio->uio_resid, m->m_len), uio); + imin(uio_resid(uio), m->m_len), uio); m = m_free(m); - } while (uio->uio_resid && error == 0 && m); + } while (uio_resid(uio) && error == 0 && m != NULL); + socket_lock(so, 0); bad: - if (m) + if (m != NULL) m_freem(m); -#ifdef __APPLE__ + if ((so->so_options & SO_WANTOOBFLAG) != 0) { if (error == EWOULDBLOCK || error == EINVAL) { - /* + /* * Let's try to get normal data: - * EWOULDBLOCK: out-of-band data not receive yet; - * EINVAL: out-of-band data already read. + * EWOULDBLOCK: out-of-band data not + * receive yet. EINVAL: out-of-band data + * already read. */ error = 0; goto nooob; - } else if (error == 0 && flagsp) + } else if (error == 0 && flagsp != NULL) { *flagsp |= MSG_OOB; + } } - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error,0,0,0,0); -#endif + socket_unlock(so, 1); + KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error, + 0, 0, 0, 0); + return (error); } nooob: - if (mp) - *mp = (struct mbuf *)0; - if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) + if (mp != NULL) + *mp = NULL; + if (so->so_state & SS_ISCONFIRMING && uio_resid(uio)) (*pr->pr_usrreqs->pru_rcvd)(so, 0); - - free_list = (struct mbuf *)0; + free_list = NULL; delayed_copy_len = 0; restart: +#ifdef MORE_LOCKING_DEBUG + if (so->so_usecount <= 1) + printf("soreceive: sblock so=%p ref=%d on socket\n", + so, so->so_usecount); +#endif + /* + * See if the socket has been closed (SS_NOFDREF|SS_CANTRCVMORE) + * and if so just return to the caller. This could happen when + * soreceive() is called by a socket upcall function during the + * time the socket is freed. The socket buffer would have been + * locked across the upcall, therefore we cannot put this thread + * to sleep (else we will deadlock) or return EWOULDBLOCK (else + * we may livelock), because the lock on the socket buffer will + * only be released when the upcall routine returns to its caller. + * Because the socket has been officially closed, there can be + * no further read on it. + * + * A multipath subflow socket would have its SS_NOFDREF set by + * default, so check for SOF_MP_SUBFLOW socket flag; when the + * socket is closed for real, SOF_MP_SUBFLOW would be cleared. + */ + if ((so->so_state & (SS_NOFDREF | SS_CANTRCVMORE)) == + (SS_NOFDREF | SS_CANTRCVMORE) && !(so->so_flags & SOF_MP_SUBFLOW)) { + socket_unlock(so, 1); + return (0); + } + error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); if (error) { - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error,0,0,0,0); + socket_unlock(so, 1); + KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error, + 0, 0, 0, 0); return (error); } - s = splnet(); m = so->so_rcv.sb_mb; /* @@ -1353,15 +2261,22 @@ restart: * we have to do the receive in sections, and thus risk returning * a short count if a timeout or signal occurs after we start. */ - if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && - so->so_rcv.sb_cc < uio->uio_resid) && - (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || - ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && - m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { + if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && + so->so_rcv.sb_cc < uio_resid(uio)) && + (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || + ((flags & MSG_WAITALL) && uio_resid(uio) <= so->so_rcv.sb_hiwat)) && + m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { + /* + * Panic if we notice inconsistencies in the socket's + * receive list; both sb_mb and sb_cc should correctly + * reflect the contents of the list, otherwise we may + * end up with false positives during select() or poll() + * which could put the application in a bad state. + */ + SB_MB_CHECK(&so->so_rcv); - KASSERT(m != 0 || !so->so_rcv.sb_cc, ("receive 1")); if (so->so_error) { - if (m) + if (m != NULL) goto dontblock; error = so->so_error; if ((flags & MSG_PEEK) == 0) @@ -1369,13 +2284,13 @@ restart: goto release; } if (so->so_state & SS_CANTRCVMORE) { - if (m) + if (m != NULL) goto dontblock; else goto release; } - for (; m; m = m->m_next) - if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { + for (; m != NULL; m = m->m_next) + if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { m = so->so_rcv.sb_mb; goto dontblock; } @@ -1384,48 +2299,126 @@ restart: error = ENOTCONN; goto release; } - if (uio->uio_resid == 0) + if (uio_resid(uio) == 0) goto release; - if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { + if ((so->so_state & SS_NBIO) || + (flags & (MSG_DONTWAIT|MSG_NBIO))) { error = EWOULDBLOCK; goto release; } - sbunlock(&so->so_rcv); + SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); + sbunlock(&so->so_rcv, TRUE); /* keep socket locked */ +#if EVEN_MORE_LOCKING_DEBUG if (socket_debug) - printf("Waiting for socket data\n"); + printf("Waiting for socket data\n"); +#endif error = sbwait(&so->so_rcv); +#if EVEN_MORE_LOCKING_DEBUG if (socket_debug) - printf("SORECEIVE - sbwait returned %d\n", error); - splx(s); + printf("SORECEIVE - sbwait returned %d\n", error); +#endif + if (so->so_usecount < 1) { + panic("%s: after 2nd sblock so=%p ref=%d on socket\n", + __func__, so, so->so_usecount); + /* NOTREACHED */ + } if (error) { - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error,0,0,0,0); - return (error); + socket_unlock(so, 1); + KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error, + 0, 0, 0, 0); + return (error); } goto restart; } dontblock: -#ifndef __APPLE__ - if (uio->uio_procp) - uio->uio_procp->p_stats->p_ru.ru_msgrcv++; -#else /* __APPLE__ */ - /* - * 2207985 - * This should be uio->uio-procp; however, some callers of this - * function use auto variables with stack garbage, and fail to - * fill out the uio structure properly. - */ - if (p) - p->p_stats->p_ru.ru_msgrcv++; -#endif /* __APPLE__ */ + OSIncrementAtomicLong(&p->p_stats->p_ru.ru_msgrcv); + SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); nextrecord = m->m_nextpkt; if ((pr->pr_flags & PR_ADDR) && m->m_type == MT_SONAME) { KASSERT(m->m_type == MT_SONAME, ("receive 1a")); +#if CONFIG_MACF_SOCKET_SUBSET + /* + * Call the MAC framework for policy checking if we're in + * the user process context and the socket isn't connected. + */ + if (p != kernproc && !(so->so_state & SS_ISCONNECTED)) { + struct mbuf *m0 = m; + /* + * Dequeue this record (temporarily) from the receive + * list since we're about to drop the socket's lock + * where a new record may arrive and be appended to + * the list. Upon MAC policy failure, the record + * will be freed. Otherwise, we'll add it back to + * the head of the list. We cannot rely on SB_LOCK + * because append operation uses the socket's lock. + */ + do { + m->m_nextpkt = NULL; + sbfree(&so->so_rcv, m); + m = m->m_next; + } while (m != NULL); + m = m0; + so->so_rcv.sb_mb = nextrecord; + SB_EMPTY_FIXUP(&so->so_rcv); + SBLASTRECORDCHK(&so->so_rcv, "soreceive 1a"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 1a"); + socket_unlock(so, 0); + if (mac_socket_check_received(proc_ucred(p), so, + mtod(m, struct sockaddr *)) != 0) { + /* + * MAC policy failure; free this record and + * process the next record (or block until + * one is available). We have adjusted sb_cc + * and sb_mbcnt above so there is no need to + * call sbfree() again. + */ + do { + m = m_free(m); + } while (m != NULL); + /* + * Clear SB_LOCK but don't unlock the socket. + * Process the next record or wait for one. + */ + socket_lock(so, 0); + sbunlock(&so->so_rcv, TRUE); /* stay locked */ + goto restart; + } + socket_lock(so, 0); + /* + * If the socket has been defunct'd, drop it. + */ + if (so->so_flags & SOF_DEFUNCT) { + m_freem(m); + error = ENOTCONN; + goto release; + } + /* + * Re-adjust the socket receive list and re-enqueue + * the record in front of any packets which may have + * been appended while we dropped the lock. + */ + for (m = m0; m->m_next != NULL; m = m->m_next) + sballoc(&so->so_rcv, m); + sballoc(&so->so_rcv, m); + if (so->so_rcv.sb_mb == NULL) { + so->so_rcv.sb_lastrecord = m0; + so->so_rcv.sb_mbtail = m; + } + m = m0; + nextrecord = m->m_nextpkt = so->so_rcv.sb_mb; + so->so_rcv.sb_mb = m; + SBLASTRECORDCHK(&so->so_rcv, "soreceive 1b"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 1b"); + } +#endif /* CONFIG_MACF_SOCKET_SUBSET */ orig_resid = 0; - if (psa) { + if (psa != NULL) { *psa = dup_sockaddr(mtod(m, struct sockaddr *), - mp0 == 0); - if ((*psa == 0) && (flags & MSG_NEEDSA)) { + mp0 == NULL); + if ((*psa == NULL) && (flags & MSG_NEEDSA)) { error = EWOULDBLOCK; goto release; } @@ -1434,82 +2427,224 @@ dontblock: m = m->m_next; } else { sbfree(&so->so_rcv, m); + if (m->m_next == NULL && so->so_rcv.sb_cc != 0) { + panic("%s: about to create invalid socketbuf", + __func__); + /* NOTREACHED */ + } MFREE(m, so->so_rcv.sb_mb); m = so->so_rcv.sb_mb; + if (m != NULL) { + m->m_nextpkt = nextrecord; + } else { + so->so_rcv.sb_mb = nextrecord; + SB_EMPTY_FIXUP(&so->so_rcv); + } } } - while (m && m->m_type == MT_CONTROL && error == 0) { - if (flags & MSG_PEEK) { - if (controlp) - *controlp = m_copy(m, 0, m->m_len); - m = m->m_next; - } else { - sbfree(&so->so_rcv, m); - if (controlp) { - if (pr->pr_domain->dom_externalize && - mtod(m, struct cmsghdr *)->cmsg_type == - SCM_RIGHTS) - error = (*pr->pr_domain->dom_externalize)(m); - *controlp = m; - so->so_rcv.sb_mb = m->m_next; - m->m_next = 0; - m = so->so_rcv.sb_mb; + + /* + * Process one or more MT_CONTROL mbufs present before any data mbufs + * in the first mbuf chain on the socket buffer. If MSG_PEEK, we + * just copy the data; if !MSG_PEEK, we call into the protocol to + * perform externalization. + */ + if (m != NULL && m->m_type == MT_CONTROL) { + struct mbuf *cm = NULL, *cmn; + struct mbuf **cme = &cm; + struct sockbuf *sb_rcv = &so->so_rcv; + struct mbuf **msgpcm = NULL; + + /* + * Externalizing the control messages would require us to + * drop the socket's lock below. Once we re-acquire the + * lock, the mbuf chain might change. In order to preserve + * consistency, we unlink all control messages from the + * first mbuf chain in one shot and link them separately + * onto a different chain. + */ + do { + if (flags & MSG_PEEK) { + if (controlp != NULL) { + if (*controlp == NULL) { + msgpcm = controlp; + } + *controlp = m_copy(m, 0, m->m_len); + + /* + * If we failed to allocate an mbuf, + * release any previously allocated + * mbufs for control data. Return + * an error. Keep the mbufs in the + * socket as this is using + * MSG_PEEK flag. + */ + if (*controlp == NULL) { + m_freem(*msgpcm); + error = ENOBUFS; + goto release; + } + controlp = &(*controlp)->m_next; + } + m = m->m_next; } else { - MFREE(m, so->so_rcv.sb_mb); - m = so->so_rcv.sb_mb; + m->m_nextpkt = NULL; + sbfree(sb_rcv, m); + sb_rcv->sb_mb = m->m_next; + m->m_next = NULL; + *cme = m; + cme = &(*cme)->m_next; + m = sb_rcv->sb_mb; + } + } while (m != NULL && m->m_type == MT_CONTROL); + + if (!(flags & MSG_PEEK)) { + if (sb_rcv->sb_mb != NULL) { + sb_rcv->sb_mb->m_nextpkt = nextrecord; + } else { + sb_rcv->sb_mb = nextrecord; + SB_EMPTY_FIXUP(sb_rcv); + } + if (nextrecord == NULL) + sb_rcv->sb_lastrecord = m; + } + + SBLASTRECORDCHK(&so->so_rcv, "soreceive ctl"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive ctl"); + + while (cm != NULL) { + int cmsg_type; + + cmn = cm->m_next; + cm->m_next = NULL; + cmsg_type = mtod(cm, struct cmsghdr *)->cmsg_type; + + /* + * Call the protocol to externalize SCM_RIGHTS message + * and return the modified message to the caller upon + * success. Otherwise, all other control messages are + * returned unmodified to the caller. Note that we + * only get into this loop if MSG_PEEK is not set. + */ + if (pr->pr_domain->dom_externalize != NULL && + cmsg_type == SCM_RIGHTS) { + /* + * Release socket lock: see 3903171. This + * would also allow more records to be appended + * to the socket buffer. We still have SB_LOCK + * set on it, so we can be sure that the head + * of the mbuf chain won't change. + */ + socket_unlock(so, 0); + error = (*pr->pr_domain->dom_externalize)(cm); + socket_lock(so, 0); + } else { + error = 0; } + + if (controlp != NULL && error == 0) { + *controlp = cm; + controlp = &(*controlp)->m_next; + orig_resid = 0; + } else { + (void) m_free(cm); + } + cm = cmn; } - if (controlp) { - orig_resid = 0; - controlp = &(*controlp)->m_next; + /* + * Update the value of nextrecord in case we received new + * records when the socket was unlocked above for + * externalizing SCM_RIGHTS. + */ + if (m != NULL) + nextrecord = sb_rcv->sb_mb->m_nextpkt; + else + nextrecord = sb_rcv->sb_mb; + orig_resid = 0; + } + + /* + * If the socket is a TCP socket with message delivery + * enabled, then create a control msg to deliver the + * relative TCP sequence number for this data. Waiting + * until this point will protect against failures to + * allocate an mbuf for control msgs. + */ + if (so->so_type == SOCK_STREAM && SOCK_PROTO(so) == IPPROTO_TCP && + (so->so_flags & SOF_ENABLE_MSGS) && controlp != NULL) { + struct mbuf *seq_cm; + + seq_cm = sbcreatecontrol((caddr_t)&m->m_pkthdr.msg_seq, + sizeof (uint32_t), SCM_SEQNUM, SOL_SOCKET); + if (seq_cm == NULL) { + /* unable to allocate a control mbuf */ + error = ENOBUFS; + goto release; } + *controlp = seq_cm; + controlp = &seq_cm->m_next; } - if (m) { - if ((flags & MSG_PEEK) == 0) - m->m_nextpkt = nextrecord; + + if (m != NULL) { + if (!(flags & MSG_PEEK)) { + /* + * We get here because m points to an mbuf following + * any MT_SONAME or MT_CONTROL mbufs which have been + * processed above. In any case, m should be pointing + * to the head of the mbuf chain, and the nextrecord + * should be either NULL or equal to m->m_nextpkt. + * See comments above about SB_LOCK. + */ + if (m != so->so_rcv.sb_mb || + m->m_nextpkt != nextrecord) { + panic("%s: post-control !sync so=%p m=%p " + "nextrecord=%p\n", __func__, so, m, + nextrecord); + /* NOTREACHED */ + } + if (nextrecord == NULL) + so->so_rcv.sb_lastrecord = m; + } type = m->m_type; if (type == MT_OOBDATA) flags |= MSG_OOB; + } else { + if (!(flags & MSG_PEEK)) { + SB_EMPTY_FIXUP(&so->so_rcv); + } } + SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); + moff = 0; offset = 0; - if (!(flags & MSG_PEEK) && uio->uio_resid > sorecvmincopy) - can_delay = 1; + if (!(flags & MSG_PEEK) && uio_resid(uio) > sorecvmincopy) + can_delay = 1; else - can_delay = 0; + can_delay = 0; need_event = 0; - - while (m && (uio->uio_resid - delayed_copy_len) > 0 && error == 0) { + while (m != NULL && + (uio_resid(uio) - delayed_copy_len) > 0 && error == 0) { if (m->m_type == MT_OOBDATA) { if (type != MT_OOBDATA) break; - } else if (type == MT_OOBDATA) + } else if (type == MT_OOBDATA) { break; -#ifndef __APPLE__ -/* - * This assertion needs rework. The trouble is Appletalk is uses many - * mbuf types (NOT listed in mbuf.h!) which will trigger this panic. - * For now just remove the assertion... CSM 9/98 - */ - else - KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, - ("receive 3")); -#else + } /* - * Make sure to allways set MSG_OOB event when getting + * Make sure to allways set MSG_OOB event when getting * out of band data inline. */ if ((so->so_options & SO_WANTOOBFLAG) != 0 && - (so->so_options & SO_OOBINLINE) != 0 && - (so->so_state & SS_RCVATMARK) != 0) { + (so->so_options & SO_OOBINLINE) != 0 && + (so->so_state & SS_RCVATMARK) != 0) { flags |= MSG_OOB; } -#endif so->so_state &= ~SS_RCVATMARK; - len = uio->uio_resid - delayed_copy_len; + len = uio_resid(uio) - delayed_copy_len; if (so->so_oobmark && len > so->so_oobmark - offset) len = so->so_oobmark - offset; if (len > m->m_len - moff) @@ -1522,49 +2657,56 @@ dontblock: * we must note any additions to the sockbuf when we * block interrupts again. */ - if (mp == 0) { + if (mp == NULL) { + SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); if (can_delay && len == m->m_len) { - /* + /* * only delay the copy if we're consuming the * mbuf and we're NOT in MSG_PEEK mode * and we have enough data to make it worthwile - * to drop and retake the funnel... can_delay - * reflects the state of the 2 latter constraints - * moff should always be zero in these cases + * to drop and retake the lock... can_delay + * reflects the state of the 2 latter + * constraints moff should always be zero + * in these cases */ - delayed_copy_len += len; + delayed_copy_len += len; } else { - splx(s); - - if (delayed_copy_len) { - error = sodelayed_copy(uio, &free_list, &delayed_copy_len); + if (delayed_copy_len) { + error = sodelayed_copy(so, uio, + &free_list, &delayed_copy_len); if (error) { - s = splnet(); goto release; } + /* + * can only get here if MSG_PEEK is not + * set therefore, m should point at the + * head of the rcv queue; if it doesn't, + * it means something drastically + * changed while we were out from behind + * the lock in sodelayed_copy. perhaps + * a RST on the stream. in any event, + * the stream has been interrupted. it's + * probably best just to return whatever + * data we've moved and let the caller + * sort it out... + */ if (m != so->so_rcv.sb_mb) { - /* - * can only get here if MSG_PEEK is not set - * therefore, m should point at the head of the rcv queue... - * if it doesn't, it means something drastically changed - * while we were out from behind the funnel in sodelayed_copy... - * perhaps a RST on the stream... in any event, the stream has - * been interrupted... it's probably best just to return - * whatever data we've moved and let the caller sort it out... - */ - break; + break; } } - error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); + socket_unlock(so, 0); + error = uiomove(mtod(m, caddr_t) + moff, + (int)len, uio); + socket_lock(so, 0); - s = splnet(); if (error) - goto release; + goto release; } - } else - uio->uio_resid -= len; - + } else { + uio_setresid(uio, (uio_resid(uio) - len)); + } if (len == m->m_len - moff) { if (m->m_flags & M_EOR) flags |= MSG_EOR; @@ -1574,31 +2716,79 @@ dontblock: } else { nextrecord = m->m_nextpkt; sbfree(&so->so_rcv, m); + m->m_nextpkt = NULL; + + /* + * If this packet is an unordered packet + * (indicated by M_UNORDERED_DATA flag), remove + * the additional bytes added to the + * receive socket buffer size. + */ + if ((so->so_flags & SOF_ENABLE_MSGS) && + m->m_len && + (m->m_flags & M_UNORDERED_DATA) && + sbreserve(&so->so_rcv, + so->so_rcv.sb_hiwat - m->m_len)) { + if (so->so_msg_state->msg_uno_bytes > + m->m_len) { + so->so_msg_state-> + msg_uno_bytes -= m->m_len; + } else { + so->so_msg_state-> + msg_uno_bytes = 0; + } + m->m_flags &= ~M_UNORDERED_DATA; + } - if (mp) { + if (mp != NULL) { *mp = m; mp = &m->m_next; so->so_rcv.sb_mb = m = m->m_next; - *mp = (struct mbuf *)0; + *mp = NULL; } else { - m->m_nextpkt = 0; if (free_list == NULL) - free_list = m; - else - ml->m_next = m; - ml = m; + free_list = m; + else + ml->m_next = m; + ml = m; so->so_rcv.sb_mb = m = m->m_next; - ml->m_next = 0; + ml->m_next = NULL; } - if (m) + if (m != NULL) { m->m_nextpkt = nextrecord; + if (nextrecord == NULL) + so->so_rcv.sb_lastrecord = m; + } else { + so->so_rcv.sb_mb = nextrecord; + SB_EMPTY_FIXUP(&so->so_rcv); + } + SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); } } else { - if (flags & MSG_PEEK) + if (flags & MSG_PEEK) { moff += len; - else { - if (mp) - *mp = m_copym(m, 0, len, M_WAIT); + } else { + if (mp != NULL) { + int copy_flag; + + if (flags & MSG_DONTWAIT) + copy_flag = M_DONTWAIT; + else + copy_flag = M_WAIT; + *mp = m_copym(m, 0, len, copy_flag); + /* + * Failed to allocate an mbuf? + * Adjust uio_resid back, it was + * adjusted down by len bytes which + * we didn't copy over. + */ + if (*mp == NULL) { + uio_setresid(uio, + (uio_resid(uio) + len)); + break; + } + } m->m_data += len; m->m_len -= len; so->so_rcv.sb_cc -= len; @@ -1608,13 +2798,14 @@ dontblock: if ((flags & MSG_PEEK) == 0) { so->so_oobmark -= len; if (so->so_oobmark == 0) { - so->so_state |= SS_RCVATMARK; - /* - * delay posting the actual event until after - * any delayed copy processing has finished - */ - need_event = 1; - break; + so->so_state |= SS_RCVATMARK; + /* + * delay posting the actual event until + * after any delayed copy processing + * has finished + */ + need_event = 1; + break; } } else { offset += len; @@ -1625,220 +2816,404 @@ dontblock: if (flags & MSG_EOR) break; /* - * If the MSG_WAITALL or MSG_WAITSTREAM flag is set (for non-atomic socket), - * we must not quit until "uio->uio_resid == 0" or an error - * termination. If a signal/timeout occurs, return - * with a short count but without error. - * Keep sockbuf locked against other readers. + * If the MSG_WAITALL or MSG_WAITSTREAM flag is set + * (for non-atomic socket), we must not quit until + * "uio->uio_resid == 0" or an error termination. + * If a signal/timeout occurs, return with a short + * count but without error. Keep sockbuf locked + * against other readers. */ - while (flags & (MSG_WAITALL|MSG_WAITSTREAM) && m == 0 && (uio->uio_resid - delayed_copy_len) > 0 && + while (flags & (MSG_WAITALL|MSG_WAITSTREAM) && m == NULL && + (uio_resid(uio) - delayed_copy_len) > 0 && !sosendallatonce(so) && !nextrecord) { if (so->so_error || so->so_state & SS_CANTRCVMORE) - goto release; + goto release; + + /* + * Depending on the protocol (e.g. TCP), the following + * might cause the socket lock to be dropped and later + * be reacquired, and more data could have arrived and + * have been appended to the receive socket buffer by + * the time it returns. Therefore, we only sleep in + * sbwait() below if and only if the socket buffer is + * empty, in order to avoid a false sleep. + */ + if (pr->pr_flags & PR_WANTRCVD && so->so_pcb && + (((struct inpcb *)so->so_pcb)->inp_state != + INPCB_STATE_DEAD)) + (*pr->pr_usrreqs->pru_rcvd)(so, flags); + + SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); - if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) - (*pr->pr_usrreqs->pru_rcvd)(so, flags); - if (sbwait(&so->so_rcv)) { - error = 0; + if (so->so_rcv.sb_mb == NULL && sbwait(&so->so_rcv)) { + error = 0; goto release; } /* - * have to wait until after we get back from the sbwait to do the copy because - * we will drop the funnel if we have enough data that has been delayed... by dropping - * the funnel we open up a window allowing the netisr thread to process the incoming packets - * and to change the state of this socket... we're issuing the sbwait because - * the socket is empty and we're expecting the netisr thread to wake us up when more - * packets arrive... if we allow that processing to happen and then sbwait, we - * could stall forever with packets sitting in the socket if no further packets - * arrive from the remote side. + * have to wait until after we get back from the sbwait + * to do the copy because we will drop the lock if we + * have enough data that has been delayed... by dropping + * the lock we open up a window allowing the netisr + * thread to process the incoming packets and to change + * the state of this socket... we're issuing the sbwait + * because the socket is empty and we're expecting the + * netisr thread to wake us up when more packets arrive; + * if we allow that processing to happen and then sbwait + * we could stall forever with packets sitting in the + * socket if no further packets arrive from the remote + * side. * - * we want to copy before we've collected all the data to satisfy this request to - * allow the copy to overlap the incoming packet processing on an MP system + * we want to copy before we've collected all the data + * to satisfy this request to allow the copy to overlap + * the incoming packet processing on an MP system */ - if (delayed_copy_len > sorecvmincopy && (delayed_copy_len > (so->so_rcv.sb_hiwat / 2))) { - - error = sodelayed_copy(uio, &free_list, &delayed_copy_len); + if (delayed_copy_len > sorecvmincopy && + (delayed_copy_len > (so->so_rcv.sb_hiwat / 2))) { + error = sodelayed_copy(so, uio, + &free_list, &delayed_copy_len); if (error) - goto release; + goto release; } m = so->so_rcv.sb_mb; - if (m) { + if (m != NULL) { nextrecord = m->m_nextpkt; } + SB_MB_CHECK(&so->so_rcv); } } +#ifdef MORE_LOCKING_DEBUG + if (so->so_usecount <= 1) { + panic("%s: after big while so=%p ref=%d on socket\n", + __func__, so, so->so_usecount); + /* NOTREACHED */ + } +#endif - if (m && pr->pr_flags & PR_ATOMIC) { -#ifdef __APPLE__ - if (so->so_options & SO_DONTTRUNC) + if (m != NULL && pr->pr_flags & PR_ATOMIC) { + if (so->so_options & SO_DONTTRUNC) { flags |= MSG_RCVMORE; - else { -#endif + } else { flags |= MSG_TRUNC; if ((flags & MSG_PEEK) == 0) (void) sbdroprecord(&so->so_rcv); -#ifdef __APPLE__ } -#endif } - if ((flags & MSG_PEEK) == 0) { - if (m == 0) - so->so_rcv.sb_mb = nextrecord; - if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) - (*pr->pr_usrreqs->pru_rcvd)(so, flags); - } -#ifdef __APPLE__ + + /* + * pru_rcvd below (for TCP) may cause more data to be received + * if the socket lock is dropped prior to sending the ACK; some + * legacy OpenTransport applications don't handle this well + * (if it receives less data than requested while MSG_HAVEMORE + * is set), and so we set the flag now based on what we know + * prior to calling pru_rcvd. + */ if ((so->so_options & SO_WANTMORE) && so->so_rcv.sb_cc > 0) flags |= MSG_HAVEMORE; - if (delayed_copy_len) { - error = sodelayed_copy(uio, &free_list, &delayed_copy_len); + if ((flags & MSG_PEEK) == 0) { + if (m == NULL) { + so->so_rcv.sb_mb = nextrecord; + /* + * First part is an inline SB_EMPTY_FIXUP(). Second + * part makes sure sb_lastrecord is up-to-date if + * there is still data in the socket buffer. + */ + if (so->so_rcv.sb_mb == NULL) { + so->so_rcv.sb_mbtail = NULL; + so->so_rcv.sb_lastrecord = NULL; + } else if (nextrecord->m_nextpkt == NULL) { + so->so_rcv.sb_lastrecord = nextrecord; + } + SB_MB_CHECK(&so->so_rcv); + } + SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); + if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) + (*pr->pr_usrreqs->pru_rcvd)(so, flags); + } + if (delayed_copy_len) { + error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len); if (error) - goto release; + goto release; } - if (free_list) { - m_freem_list((struct mbuf *)free_list); - free_list = (struct mbuf *)0; + if (free_list != NULL) { + m_freem_list(free_list); + free_list = NULL; } if (need_event) - postevent(so, 0, EV_OOB); -#endif - if (orig_resid == uio->uio_resid && orig_resid && + postevent(so, 0, EV_OOB); + + if (orig_resid == uio_resid(uio) && orig_resid && (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { - sbunlock(&so->so_rcv); - splx(s); + sbunlock(&so->so_rcv, TRUE); /* keep socket locked */ goto restart; } - if (flagsp) + if (flagsp != NULL) *flagsp |= flags; release: - if (delayed_copy_len) { - error = sodelayed_copy(uio, &free_list, &delayed_copy_len); +#ifdef MORE_LOCKING_DEBUG + if (so->so_usecount <= 1) { + panic("%s: release so=%p ref=%d on socket\n", __func__, + so, so->so_usecount); + /* NOTREACHED */ } - if (free_list) { - m_freem_list((struct mbuf *)free_list); - } - sbunlock(&so->so_rcv); - splx(s); +#endif + if (delayed_copy_len) + error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len); + + if (free_list != NULL) + m_freem_list(free_list); + + sbunlock(&so->so_rcv, FALSE); /* will unlock socket */ - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, - so, - uio->uio_resid, - so->so_rcv.sb_cc, - 0, - error); + KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, so, uio_resid(uio), + so->so_rcv.sb_cc, 0, error); return (error); } - -int sodelayed_copy(struct uio *uio, struct mbuf **free_list, int *resid) +/* + * Returns: 0 Success + * uiomove:EFAULT + */ +static int +sodelayed_copy(struct socket *so, struct uio *uio, struct mbuf **free_list, + user_ssize_t *resid) { - int error = 0; - boolean_t dropped_funnel = FALSE; + int error = 0; struct mbuf *m; m = *free_list; - if (*resid >= sorecvmincopy) { - dropped_funnel = TRUE; - - (void)thread_funnel_set(network_flock, FALSE); - } - while (m && error == 0) { - - error = uiomove(mtod(m, caddr_t), (int)m->m_len, uio); + socket_unlock(so, 0); + while (m != NULL && error == 0) { + error = uiomove(mtod(m, caddr_t), (int)m->m_len, uio); m = m->m_next; } m_freem_list(*free_list); - *free_list = (struct mbuf *)NULL; + *free_list = NULL; *resid = 0; - if (dropped_funnel == TRUE) - (void)thread_funnel_set(network_flock, TRUE); + socket_lock(so, 0); return (error); } - +/* + * Returns: 0 Success + * EINVAL + * ENOTCONN + * :EINVAL + * :EADDRNOTAVAIL[TCP] + * :ENOBUFS[TCP] + * :EMSGSIZE[TCP] + * :EHOSTUNREACH[TCP] + * :ENETUNREACH[TCP] + * :ENETDOWN[TCP] + * :ENOMEM[TCP] + * :EACCES[TCP] + * :EMSGSIZE[TCP] + * :ENOBUFS[TCP] + * :???[TCP] [ignorable: mostly IPSEC/firewall/DLIL] + * :??? [other protocol families] + */ int -soshutdown(so, how) - register struct socket *so; - register int how; -{ - register struct protosw *pr = so->so_proto; - struct kextcb *kp; - int ret; - - - KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_START, 0,0,0,0,0); - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_soshutdown) { - ret = (*kp->e_soif->sf_soshutdown)(so, how, kp); - if (ret) { - KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_END, 0,0,0,0,0); - return((ret == EJUSTRETURN) ? 0 : ret); - } +soshutdown(struct socket *so, int how) +{ + int error; + + switch (how) { + case SHUT_RD: + case SHUT_WR: + case SHUT_RDWR: + socket_lock(so, 1); + if ((so->so_state & + (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING)) == 0) { + error = ENOTCONN; + } else { + error = soshutdownlock(so, how); } - kp = kp->e_next; + socket_unlock(so, 1); + break; + default: + error = EINVAL; + break; } + return (error); +} + +int +soshutdownlock(struct socket *so, int how) +{ + struct protosw *pr = so->so_proto; + int error = 0; + + sflt_notify(so, sock_evt_shutdown, &how); + if (how != SHUT_WR) { + if ((so->so_state & SS_CANTRCVMORE) != 0) { + /* read already shut down */ + error = ENOTCONN; + goto done; + } sorflush(so); postevent(so, 0, EV_RCLOSED); } if (how != SHUT_RD) { - ret = ((*pr->pr_usrreqs->pru_shutdown)(so)); - postevent(so, 0, EV_WCLOSED); - KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_END, 0,0,0,0,0); - return(ret); + if ((so->so_state & SS_CANTSENDMORE) != 0) { + /* write already shut down */ + error = ENOTCONN; + goto done; + } + error = (*pr->pr_usrreqs->pru_shutdown)(so); + postevent(so, 0, EV_WCLOSED); } +done: + KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_END, 0, 0, 0, 0, 0); + return (error); +} - KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_END, 0,0,0,0,0); - return (0); +void +sowflush(struct socket *so) +{ + struct sockbuf *sb = &so->so_snd; +#ifdef notyet + lck_mtx_t *mutex_held; + /* + * XXX: This code is currently commented out, because we may get here + * as part of sofreelastref(), and at that time, pr_getlock() may no + * longer be able to return us the lock; this will be fixed in future. + */ + if (so->so_proto->pr_getlock != NULL) + mutex_held = (*so->so_proto->pr_getlock)(so, 0); + else + mutex_held = so->so_proto->pr_domain->dom_mtx; + + lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); +#endif /* notyet */ + + /* + * Obtain lock on the socket buffer (SB_LOCK). This is required + * to prevent the socket buffer from being unexpectedly altered + * while it is used by another thread in socket send/receive. + * + * sblock() must not fail here, hence the assertion. + */ + (void) sblock(sb, SBL_WAIT | SBL_NOINTR | SBL_IGNDEFUNCT); + VERIFY(sb->sb_flags & SB_LOCK); + + sb->sb_flags &= ~(SB_SEL|SB_UPCALL); + sb->sb_flags |= SB_DROP; + sb->sb_upcall = NULL; + sb->sb_upcallarg = NULL; + + sbunlock(sb, TRUE); /* keep socket locked */ + + selthreadclear(&sb->sb_sel); + sbrelease(sb); } void -sorflush(so) - register struct socket *so; +sorflush(struct socket *so) { - register struct sockbuf *sb = &so->so_rcv; - register struct protosw *pr = so->so_proto; - register int s, error; + struct sockbuf *sb = &so->so_rcv; + struct protosw *pr = so->so_proto; struct sockbuf asb; - struct kextcb *kp; +#ifdef notyet + lck_mtx_t *mutex_held; + /* + * XXX: This code is currently commented out, because we may get here + * as part of sofreelastref(), and at that time, pr_getlock() may no + * longer be able to return us the lock; this will be fixed in future. + */ + if (so->so_proto->pr_getlock != NULL) + mutex_held = (*so->so_proto->pr_getlock)(so, 0); + else + mutex_held = so->so_proto->pr_domain->dom_mtx; - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_sorflush) { - if ((*kp->e_soif->sf_sorflush)(so, kp)) - return; - } - kp = kp->e_next; - } + lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); +#endif /* notyet */ + + sflt_notify(so, sock_evt_flush_read, NULL); - sb->sb_flags |= SB_NOINTR; - (void) sblock(sb, M_WAIT); - s = splimp(); socantrcvmore(so); - sbunlock(sb); -#ifdef __APPLE__ + + /* + * Obtain lock on the socket buffer (SB_LOCK). This is required + * to prevent the socket buffer from being unexpectedly altered + * while it is used by another thread in socket send/receive. + * + * sblock() must not fail here, hence the assertion. + */ + (void) sblock(sb, SBL_WAIT | SBL_NOINTR | SBL_IGNDEFUNCT); + VERIFY(sb->sb_flags & SB_LOCK); + + /* + * Copy only the relevant fields from "sb" to "asb" which we + * need for sbrelease() to function. In particular, skip + * sb_sel as it contains the wait queue linkage, which would + * wreak havoc if we were to issue selthreadclear() on "asb". + * Make sure to not carry over SB_LOCK in "asb", as we need + * to acquire it later as part of sbrelease(). + */ + bzero(&asb, sizeof (asb)); + asb.sb_cc = sb->sb_cc; + asb.sb_hiwat = sb->sb_hiwat; + asb.sb_mbcnt = sb->sb_mbcnt; + asb.sb_mbmax = sb->sb_mbmax; + asb.sb_ctl = sb->sb_ctl; + asb.sb_lowat = sb->sb_lowat; + asb.sb_mb = sb->sb_mb; + asb.sb_mbtail = sb->sb_mbtail; + asb.sb_lastrecord = sb->sb_lastrecord; + asb.sb_so = sb->sb_so; + asb.sb_flags = sb->sb_flags; + asb.sb_flags &= ~(SB_LOCK|SB_SEL|SB_KNOTE|SB_UPCALL); + asb.sb_flags |= SB_DROP; + + /* + * Ideally we'd bzero() these and preserve the ones we need; + * but to do that we'd need to shuffle things around in the + * sockbuf, and we can't do it now because there are KEXTS + * that are directly referring to the socket structure. + * + * Setting SB_DROP acts as a barrier to prevent further appends. + * Clearing SB_SEL is done for selthreadclear() below. + */ + sb->sb_cc = 0; + sb->sb_hiwat = 0; + sb->sb_mbcnt = 0; + sb->sb_mbmax = 0; + sb->sb_ctl = 0; + sb->sb_lowat = 0; + sb->sb_mb = NULL; + sb->sb_mbtail = NULL; + sb->sb_lastrecord = NULL; + sb->sb_timeo.tv_sec = 0; + sb->sb_timeo.tv_usec = 0; + sb->sb_upcall = NULL; + sb->sb_upcallarg = NULL; + sb->sb_flags &= ~(SB_SEL|SB_UPCALL); + sb->sb_flags |= SB_DROP; + + sbunlock(sb, TRUE); /* keep socket locked */ + + /* + * Note that selthreadclear() is called on the original "sb" and + * not the local "asb" because of the way wait queue linkage is + * implemented. Given that selwakeup() may be triggered, SB_SEL + * should no longer be set (cleared above.) + */ selthreadclear(&sb->sb_sel); -#endif - asb = *sb; - bzero((caddr_t)sb, sizeof (*sb)); - if (asb.sb_flags & SB_KNOTE) { - sb->sb_sel.si_note = asb.sb_sel.si_note; - sb->sb_flags = SB_KNOTE; - } - splx(s); - if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) + + if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose) (*pr->pr_domain->dom_dispose)(asb.sb_mb); sbrelease(&asb); @@ -1850,13 +3225,13 @@ sorflush(so) * to be some kind of integer, but not a specific size. * In addition to their use here, these functions are also called by the * protocol-level pr_ctloutput() routines. + * + * Returns: 0 Success + * EINVAL + * copyin:EFAULT */ int -sooptcopyin(sopt, buf, len, minlen) - struct sockopt *sopt; - void *buf; - size_t len; - size_t minlen; +sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) { size_t valsize; @@ -1867,57 +3242,162 @@ sooptcopyin(sopt, buf, len, minlen) * is set to however much we actually retrieved. */ if ((valsize = sopt->sopt_valsize) < minlen) - return EINVAL; + return (EINVAL); if (valsize > len) sopt->sopt_valsize = valsize = len; - if (sopt->sopt_p != 0) + if (sopt->sopt_p != kernproc) return (copyin(sopt->sopt_val, buf, valsize)); - bcopy(sopt->sopt_val, buf, valsize); - return 0; + bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), buf, valsize); + return (0); +} + +/* + * sooptcopyin_timeval + * Copy in a timeval value into tv_p, and take into account whether the + * the calling process is 64-bit or 32-bit. Moved the sanity checking + * code here so that we can verify the 64-bit tv_sec value before we lose + * the top 32-bits assigning tv64.tv_sec to tv_p->tv_sec. + */ +static int +sooptcopyin_timeval(struct sockopt *sopt, struct timeval *tv_p) +{ + int error; + + if (proc_is64bit(sopt->sopt_p)) { + struct user64_timeval tv64; + + if (sopt->sopt_valsize < sizeof (tv64)) + return (EINVAL); + + sopt->sopt_valsize = sizeof (tv64); + if (sopt->sopt_p != kernproc) { + error = copyin(sopt->sopt_val, &tv64, sizeof (tv64)); + if (error != 0) + return (error); + } else { + bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), &tv64, + sizeof (tv64)); + } + if (tv64.tv_sec < 0 || tv64.tv_sec > LONG_MAX || + tv64.tv_usec < 0 || tv64.tv_usec >= 1000000) + return (EDOM); + + tv_p->tv_sec = tv64.tv_sec; + tv_p->tv_usec = tv64.tv_usec; + } else { + struct user32_timeval tv32; + + if (sopt->sopt_valsize < sizeof (tv32)) + return (EINVAL); + + sopt->sopt_valsize = sizeof (tv32); + if (sopt->sopt_p != kernproc) { + error = copyin(sopt->sopt_val, &tv32, sizeof (tv32)); + if (error != 0) { + return (error); + } + } else { + bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), &tv32, + sizeof (tv32)); + } +#ifndef __LP64__ + /* + * K64todo "comparison is always false due to + * limited range of data type" + */ + if (tv32.tv_sec < 0 || tv32.tv_sec > LONG_MAX || + tv32.tv_usec < 0 || tv32.tv_usec >= 1000000) + return (EDOM); +#endif + tv_p->tv_sec = tv32.tv_sec; + tv_p->tv_usec = tv32.tv_usec; + } + return (0); } +/* + * Returns: 0 Success + * EINVAL + * ENOPROTOOPT + * ENOBUFS + * EDOM + * sooptcopyin:EINVAL + * sooptcopyin:EFAULT + * sooptcopyin_timeval:EINVAL + * sooptcopyin_timeval:EFAULT + * sooptcopyin_timeval:EDOM + * :EOPNOTSUPP[AF_UNIX] + * :???w + * sflt_attach_private:??? [whatever a filter author chooses] + * :??? [whatever a filter author chooses] + * + * Notes: Other returns depend on the protocol family; all + * returns depend on what the filter author causes + * their filter to return. + */ int -sosetopt(so, sopt) - struct socket *so; - struct sockopt *sopt; +sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) { int error, optval; struct linger l; struct timeval tv; - short val; - struct kextcb *kp; +#if CONFIG_MACF_SOCKET + struct mac extmac; +#endif /* MAC_SOCKET */ - if (sopt->sopt_dir != SOPT_SET) { + if (sopt->sopt_dir != SOPT_SET) sopt->sopt_dir = SOPT_SET; + + if (dolock) + socket_lock(so, 1); + + if ((so->so_state & (SS_CANTRCVMORE | SS_CANTSENDMORE)) == + (SS_CANTRCVMORE | SS_CANTSENDMORE) && + (so->so_flags & SOF_NPX_SETOPTSHUT) == 0) { + /* the socket has been shutdown, no more sockopt's */ + error = EINVAL; + goto out; } - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_socontrol) { - error = (*kp->e_soif->sf_socontrol)(so, sopt, kp); - if (error) - return((error == EJUSTRETURN) ? 0 : error); - } - kp = kp->e_next; + error = sflt_setsockopt(so, sopt); + if (error != 0) { + if (error == EJUSTRETURN) + error = 0; + goto out; } - error = 0; if (sopt->sopt_level != SOL_SOCKET) { - if (so->so_proto && so->so_proto->pr_ctloutput) - return ((*so->so_proto->pr_ctloutput) - (so, sopt)); + if (so->so_proto != NULL && + so->so_proto->pr_ctloutput != NULL) { + error = (*so->so_proto->pr_ctloutput)(so, sopt); + goto out; + } error = ENOPROTOOPT; } else { + /* + * Allow socket-level (SOL_SOCKET) options to be filtered by + * the protocol layer, if needed. A zero value returned from + * the handler means use default socket-level processing as + * done by the rest of this routine. Otherwise, any other + * return value indicates that the option is unsupported. + */ + if (so->so_proto != NULL && (error = so->so_proto->pr_usrreqs-> + pru_socheckopt(so, sopt)) != 0) + goto out; + + error = 0; switch (sopt->sopt_name) { case SO_LINGER: - error = sooptcopyin(sopt, &l, sizeof l, sizeof l); - if (error) - goto bad; + case SO_LINGER_SEC: + error = sooptcopyin(sopt, &l, sizeof (l), sizeof (l)); + if (error != 0) + goto out; - so->so_linger = l.l_linger; - if (l.l_onoff) + so->so_linger = (sopt->sopt_name == SO_LINGER) ? + l.l_linger : l.l_linger * hz; + if (l.l_onoff != 0) so->so_options |= SO_LINGER; else so->so_options &= ~SO_LINGER; @@ -1932,15 +3412,14 @@ sosetopt(so, sopt) case SO_REUSEPORT: case SO_OOBINLINE: case SO_TIMESTAMP: -#ifdef __APPLE__ + case SO_TIMESTAMP_MONOTONIC: case SO_DONTTRUNC: case SO_WANTMORE: case SO_WANTOOBFLAG: -#endif - error = sooptcopyin(sopt, &optval, sizeof optval, - sizeof optval); - if (error) - goto bad; + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; if (optval) so->so_options |= sopt->sopt_name; else @@ -1951,10 +3430,10 @@ sosetopt(so, sopt) case SO_RCVBUF: case SO_SNDLOWAT: case SO_RCVLOWAT: - error = sooptcopyin(sopt, &optval, sizeof optval, - sizeof optval); - if (error) - goto bad; + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; /* * Values < 1 make no sense for any of these @@ -1962,20 +3441,24 @@ sosetopt(so, sopt) */ if (optval < 1) { error = EINVAL; - goto bad; + goto out; } switch (sopt->sopt_name) { case SO_SNDBUF: - case SO_RCVBUF: - if (sbreserve(sopt->sopt_name == SO_SNDBUF ? - &so->so_snd : &so->so_rcv, - (u_long) optval) == 0) { + case SO_RCVBUF: { + struct sockbuf *sb = + (sopt->sopt_name == SO_SNDBUF) ? + &so->so_snd : &so->so_rcv; + if (sbreserve(sb, (u_int32_t)optval) == 0) { error = ENOBUFS; - goto bad; + goto out; } + sb->sb_flags |= SB_USRSIZE; + sb->sb_flags &= ~SB_AUTOSIZE; + sb->sb_idealsize = (u_int32_t)optval; break; - + } /* * Make sure the low-water is never greater than * the high-water. @@ -1995,95 +3478,329 @@ sosetopt(so, sopt) case SO_SNDTIMEO: case SO_RCVTIMEO: - error = sooptcopyin(sopt, &tv, sizeof tv, - sizeof tv); - if (error) - goto bad; - - /* assert(hz > 0); */ - if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz || - tv.tv_usec < 0 || tv.tv_usec >= 1000000) { - error = EDOM; - goto bad; - } - /* assert(tick > 0); */ - /* assert(ULONG_MAX - SHRT_MAX >= 1000000); */ - { - long tmp = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick; - if (tmp > SHRT_MAX) { - error = EDOM; - goto bad; - } - val = tmp; - } + error = sooptcopyin_timeval(sopt, &tv); + if (error != 0) + goto out; switch (sopt->sopt_name) { case SO_SNDTIMEO: - so->so_snd.sb_timeo = val; + so->so_snd.sb_timeo = tv; break; case SO_RCVTIMEO: - so->so_rcv.sb_timeo = val; + so->so_rcv.sb_timeo = tv; break; } break; - case SO_NKE: - { + case SO_NKE: { struct so_nke nke; - struct NFDescriptor *nf1, *nf2 = NULL; - error = sooptcopyin(sopt, &nke, - sizeof nke, sizeof nke); - if (error) - goto bad; + error = sooptcopyin(sopt, &nke, sizeof (nke), + sizeof (nke)); + if (error != 0) + goto out; - error = nke_insert(so, &nke); + error = sflt_attach_internal(so, nke.nke_handle); break; } case SO_NOSIGPIPE: - error = sooptcopyin(sopt, &optval, sizeof optval, - sizeof optval); - if (error) - goto bad; - if (optval) - so->so_flags |= SOF_NOSIGPIPE; - else - so->so_flags &= ~SOF_NOSIGPIPE; - + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + if (optval != 0) + so->so_flags |= SOF_NOSIGPIPE; + else + so->so_flags &= ~SOF_NOSIGPIPE; break; case SO_NOADDRERR: - error = sooptcopyin(sopt, &optval, sizeof optval, - sizeof optval); - if (error) - goto bad; - if (optval) - so->so_flags |= SOF_NOADDRAVAIL; - else - so->so_flags &= ~SOF_NOADDRAVAIL; - + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + if (optval != 0) + so->so_flags |= SOF_NOADDRAVAIL; + else + so->so_flags &= ~SOF_NOADDRAVAIL; + break; + + case SO_REUSESHAREUID: + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + if (optval != 0) + so->so_flags |= SOF_REUSESHAREUID; + else + so->so_flags &= ~SOF_REUSESHAREUID; + break; + + case SO_NOTIFYCONFLICT: + if (kauth_cred_issuser(kauth_cred_get()) == 0) { + error = EPERM; + goto out; + } + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + if (optval != 0) + so->so_flags |= SOF_NOTIFYCONFLICT; + else + so->so_flags &= ~SOF_NOTIFYCONFLICT; + break; + + case SO_RESTRICTIONS: + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + + error = so_set_restrictions(so, optval); + break; + + case SO_LABEL: +#if CONFIG_MACF_SOCKET + if ((error = sooptcopyin(sopt, &extmac, sizeof (extmac), + sizeof (extmac))) != 0) + goto out; + + error = mac_setsockopt_label(proc_ucred(sopt->sopt_p), + so, &extmac); +#else + error = EOPNOTSUPP; +#endif /* MAC_SOCKET */ + break; + + case SO_UPCALLCLOSEWAIT: + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + if (optval != 0) + so->so_flags |= SOF_UPCALLCLOSEWAIT; + else + so->so_flags &= ~SOF_UPCALLCLOSEWAIT; + break; + + case SO_RANDOMPORT: + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + if (optval != 0) + so->so_flags |= SOF_BINDRANDOMPORT; + else + so->so_flags &= ~SOF_BINDRANDOMPORT; + break; + + case SO_NP_EXTENSIONS: { + struct so_np_extensions sonpx; + + error = sooptcopyin(sopt, &sonpx, sizeof (sonpx), + sizeof (sonpx)); + if (error != 0) + goto out; + if (sonpx.npx_mask & ~SONPX_MASK_VALID) { + error = EINVAL; + goto out; + } + /* + * Only one bit defined for now + */ + if ((sonpx.npx_mask & SONPX_SETOPTSHUT)) { + if ((sonpx.npx_flags & SONPX_SETOPTSHUT)) + so->so_flags |= SOF_NPX_SETOPTSHUT; + else + so->so_flags &= ~SOF_NPX_SETOPTSHUT; + } + break; + } + + case SO_TRAFFIC_CLASS: { + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + error = so_set_traffic_class(so, optval); + if (error != 0) + goto out; + break; + } + + case SO_RECV_TRAFFIC_CLASS: { + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + if (optval == 0) + so->so_flags &= ~SOF_RECV_TRAFFIC_CLASS; + else + so->so_flags |= SOF_RECV_TRAFFIC_CLASS; + break; + } + + case SO_TRAFFIC_CLASS_DBG: { + struct so_tcdbg so_tcdbg; + + error = sooptcopyin(sopt, &so_tcdbg, + sizeof (struct so_tcdbg), sizeof (struct so_tcdbg)); + if (error != 0) + goto out; + error = so_set_tcdbg(so, &so_tcdbg); + if (error != 0) + goto out; + break; + } + + case SO_PRIVILEGED_TRAFFIC_CLASS: + error = priv_check_cred(kauth_cred_get(), + PRIV_NET_PRIVILEGED_TRAFFIC_CLASS, 0); + if (error != 0) + goto out; + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + if (optval == 0) + so->so_flags &= ~SOF_PRIVILEGED_TRAFFIC_CLASS; + else + so->so_flags |= SOF_PRIVILEGED_TRAFFIC_CLASS; + break; + + case SO_DEFUNCTOK: + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0 || (so->so_flags & SOF_DEFUNCT)) { + if (error == 0) + error = EBADF; + goto out; + } + /* + * Any process can set SO_DEFUNCTOK (clear + * SOF_NODEFUNCT), but only root can clear + * SO_DEFUNCTOK (set SOF_NODEFUNCT). + */ + if (optval == 0 && + kauth_cred_issuser(kauth_cred_get()) == 0) { + error = EPERM; + goto out; + } + if (optval) + so->so_flags &= ~SOF_NODEFUNCT; + else + so->so_flags |= SOF_NODEFUNCT; + + if (SOCK_DOM(so) == PF_INET || + SOCK_DOM(so) == PF_INET6) { + char s[MAX_IPv6_STR_LEN]; + char d[MAX_IPv6_STR_LEN]; + struct inpcb *inp = sotoinpcb(so); + + SODEFUNCTLOG(("%s[%d]: so 0x%llx [%s %s:%d -> " + "%s:%d] is now marked as %seligible for " + "defunct\n", __func__, proc_selfpid(), + (uint64_t)VM_KERNEL_ADDRPERM(so), + (SOCK_TYPE(so) == SOCK_STREAM) ? + "TCP" : "UDP", inet_ntop(SOCK_DOM(so), + ((SOCK_DOM(so) == PF_INET) ? + (void *)&inp->inp_laddr.s_addr : + (void *)&inp->in6p_laddr), s, sizeof (s)), + ntohs(inp->in6p_lport), + inet_ntop(SOCK_DOM(so), + (SOCK_DOM(so) == PF_INET) ? + (void *)&inp->inp_faddr.s_addr : + (void *)&inp->in6p_faddr, d, sizeof (d)), + ntohs(inp->in6p_fport), + (so->so_flags & SOF_NODEFUNCT) ? + "not " : "")); + } else { + SODEFUNCTLOG(("%s[%d]: so 0x%llx [%d,%d] is " + "now marked as %seligible for defunct\n", + __func__, proc_selfpid(), + (uint64_t)VM_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), + (so->so_flags & SOF_NODEFUNCT) ? + "not " : "")); + } + break; + + case SO_ISDEFUNCT: + /* This option is not settable */ + error = EINVAL; + break; + + case SO_OPPORTUNISTIC: + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error == 0) + error = so_set_opportunistic(so, optval); + break; + + case SO_FLUSH: + /* This option is handled by lower layer(s) */ + error = 0; + break; + + case SO_RECV_ANYIF: + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error == 0) + error = so_set_recv_anyif(so, optval); + break; + + case SO_TRAFFIC_MGT_BACKGROUND: { + /* This option is handled by lower layer(s) */ + error = 0; + break; + } + +#if FLOW_DIVERT + case SO_FLOW_DIVERT_TOKEN: + error = flow_divert_token_set(so, sopt); + break; +#endif /* FLOW_DIVERT */ + + + case SO_DELEGATED: + if ((error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval))) != 0) + break; + + error = so_set_effective_pid(so, optval, sopt->sopt_p); + break; + + case SO_DELEGATED_UUID: { + uuid_t euuid; + + if ((error = sooptcopyin(sopt, &euuid, sizeof (euuid), + sizeof (euuid))) != 0) + break; + + error = so_set_effective_uuid(so, euuid, sopt->sopt_p); break; + } default: error = ENOPROTOOPT; break; } - if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { - (void) ((*so->so_proto->pr_ctloutput) - (so, sopt)); + if (error == 0 && so->so_proto != NULL && + so->so_proto->pr_ctloutput != NULL) { + (void) so->so_proto->pr_ctloutput(so, sopt); } } -bad: +out: + if (dolock) + socket_unlock(so, 1); return (error); } -/* Helper routine for getsockopt */ +/* Helper routines for getsockopt */ int -sooptcopyout(sopt, buf, len) - struct sockopt *sopt; - void *buf; - size_t len; +sooptcopyout(struct sockopt *sopt, void *buf, size_t len) { int error; size_t valsize; @@ -2101,53 +3818,105 @@ sooptcopyout(sopt, buf, len) */ valsize = min(len, sopt->sopt_valsize); sopt->sopt_valsize = valsize; - if (sopt->sopt_val != 0) { - if (sopt->sopt_p != 0) + if (sopt->sopt_val != USER_ADDR_NULL) { + if (sopt->sopt_p != kernproc) error = copyout(buf, sopt->sopt_val, valsize); else - bcopy(buf, sopt->sopt_val, valsize); + bcopy(buf, CAST_DOWN(caddr_t, sopt->sopt_val), valsize); } - return error; + return (error); } +static int +sooptcopyout_timeval(struct sockopt *sopt, const struct timeval *tv_p) +{ + int error; + size_t len; + struct user64_timeval tv64; + struct user32_timeval tv32; + const void * val; + size_t valsize; + + error = 0; + if (proc_is64bit(sopt->sopt_p)) { + len = sizeof (tv64); + tv64.tv_sec = tv_p->tv_sec; + tv64.tv_usec = tv_p->tv_usec; + val = &tv64; + } else { + len = sizeof (tv32); + tv32.tv_sec = tv_p->tv_sec; + tv32.tv_usec = tv_p->tv_usec; + val = &tv32; + } + valsize = min(len, sopt->sopt_valsize); + sopt->sopt_valsize = valsize; + if (sopt->sopt_val != USER_ADDR_NULL) { + if (sopt->sopt_p != kernproc) + error = copyout(val, sopt->sopt_val, valsize); + else + bcopy(val, CAST_DOWN(caddr_t, sopt->sopt_val), valsize); + } + return (error); +} + +/* + * Return: 0 Success + * ENOPROTOOPT + * :EOPNOTSUPP[AF_UNIX] + * :??? + * :??? + */ int -sogetopt(so, sopt) - struct socket *so; - struct sockopt *sopt; +sogetoptlock(struct socket *so, struct sockopt *sopt, int dolock) { int error, optval; struct linger l; struct timeval tv; - struct mbuf *m; - struct kextcb *kp; +#if CONFIG_MACF_SOCKET + struct mac extmac; +#endif /* MAC_SOCKET */ - if (sopt->sopt_dir != SOPT_GET) { - sopt->sopt_dir = SOPT_GET; - } + if (sopt->sopt_dir != SOPT_GET) + sopt->sopt_dir = SOPT_GET; - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_socontrol) { - error = (*kp->e_soif->sf_socontrol)(so, sopt, kp); - if (error) - return((error == EJUSTRETURN) ? 0 : error); - } - kp = kp->e_next; + if (dolock) + socket_lock(so, 1); + + error = sflt_getsockopt(so, sopt); + if (error != 0) { + if (error == EJUSTRETURN) + error = 0; + goto out; } - error = 0; if (sopt->sopt_level != SOL_SOCKET) { - if (so->so_proto && so->so_proto->pr_ctloutput) { - return ((*so->so_proto->pr_ctloutput) - (so, sopt)); - } else - return (ENOPROTOOPT); + if (so->so_proto != NULL && + so->so_proto->pr_ctloutput != NULL) { + error = (*so->so_proto->pr_ctloutput)(so, sopt); + goto out; + } + error = ENOPROTOOPT; } else { + /* + * Allow socket-level (SOL_SOCKET) options to be filtered by + * the protocol layer, if needed. A zero value returned from + * the handler means use default socket-level processing as + * done by the rest of this routine. Otherwise, any other + * return value indicates that the option is unsupported. + */ + if (so->so_proto != NULL && (error = so->so_proto->pr_usrreqs-> + pru_socheckopt(so, sopt)) != 0) + goto out; + + error = 0; switch (sopt->sopt_name) { case SO_LINGER: - l.l_onoff = so->so_options & SO_LINGER; - l.l_linger = so->so_linger; - error = sooptcopyout(sopt, &l, sizeof l); + case SO_LINGER_SEC: + l.l_onoff = ((so->so_options & SO_LINGER) ? 1 : 0); + l.l_linger = (sopt->sopt_name == SO_LINGER) ? + so->so_linger : so->so_linger / hz; + error = sooptcopyout(sopt, &l, sizeof (l)); break; case SO_USELOOPBACK: @@ -2159,50 +3928,43 @@ sogetopt(so, sopt) case SO_BROADCAST: case SO_OOBINLINE: case SO_TIMESTAMP: -#ifdef __APPLE__ + case SO_TIMESTAMP_MONOTONIC: case SO_DONTTRUNC: case SO_WANTMORE: case SO_WANTOOBFLAG: -#endif optval = so->so_options & sopt->sopt_name; integer: - error = sooptcopyout(sopt, &optval, sizeof optval); + error = sooptcopyout(sopt, &optval, sizeof (optval)); break; case SO_TYPE: optval = so->so_type; goto integer; -#ifdef __APPLE__ case SO_NREAD: - { - int pkt_total; - struct mbuf *m1; - - pkt_total = 0; - m1 = so->so_rcv.sb_mb; - if (so->so_proto->pr_flags & PR_ATOMIC) - { -#if 0 - kprintf("SKT CC: %d\n", so->so_rcv.sb_cc); -#endif - while (m1) { - if (m1->m_type == MT_DATA) + if (so->so_proto->pr_flags & PR_ATOMIC) { + int pkt_total; + struct mbuf *m1; + + pkt_total = 0; + m1 = so->so_rcv.sb_mb; + while (m1 != NULL) { + if (m1->m_type == MT_DATA || + m1->m_type == MT_HEADER || + m1->m_type == MT_OOBDATA) pkt_total += m1->m_len; -#if 0 - kprintf("CNT: %d/%d\n", m1->m_len, pkt_total); -#endif m1 = m1->m_next; } optval = pkt_total; - } else - optval = so->so_rcv.sb_cc; -#if 0 - kprintf("RTN: %d\n", optval); -#endif + } else { + optval = so->so_rcv.sb_cc - so->so_rcv.sb_ctl; + } goto integer; - } -#endif + + case SO_NWRITE: + optval = so->so_snd.sb_cc; + goto integer; + case SO_ERROR: optval = so->so_error; so->so_error = 0; @@ -2226,108 +3988,173 @@ integer: case SO_SNDTIMEO: case SO_RCVTIMEO: - optval = (sopt->sopt_name == SO_SNDTIMEO ? - so->so_snd.sb_timeo : so->so_rcv.sb_timeo); + tv = (sopt->sopt_name == SO_SNDTIMEO ? + so->so_snd.sb_timeo : so->so_rcv.sb_timeo); - tv.tv_sec = optval / hz; - tv.tv_usec = (optval % hz) * tick; - error = sooptcopyout(sopt, &tv, sizeof tv); - break; + error = sooptcopyout_timeval(sopt, &tv); + break; - case SO_NOSIGPIPE: - optval = (so->so_flags & SOF_NOSIGPIPE); - goto integer; + case SO_NOSIGPIPE: + optval = (so->so_flags & SOF_NOSIGPIPE); + goto integer; case SO_NOADDRERR: - optval = (so->so_flags & SOF_NOADDRAVAIL); - goto integer; + optval = (so->so_flags & SOF_NOADDRAVAIL); + goto integer; + + case SO_REUSESHAREUID: + optval = (so->so_flags & SOF_REUSESHAREUID); + goto integer; + + + case SO_NOTIFYCONFLICT: + optval = (so->so_flags & SOF_NOTIFYCONFLICT); + goto integer; + + case SO_RESTRICTIONS: + optval = so_get_restrictions(so); + goto integer; + + case SO_LABEL: +#if CONFIG_MACF_SOCKET + if ((error = sooptcopyin(sopt, &extmac, sizeof (extmac), + sizeof (extmac))) != 0 || + (error = mac_socket_label_get(proc_ucred( + sopt->sopt_p), so, &extmac)) != 0) + break; + + error = sooptcopyout(sopt, &extmac, sizeof (extmac)); +#else + error = EOPNOTSUPP; +#endif /* MAC_SOCKET */ + break; + + case SO_PEERLABEL: +#if CONFIG_MACF_SOCKET + if ((error = sooptcopyin(sopt, &extmac, sizeof (extmac), + sizeof (extmac))) != 0 || + (error = mac_socketpeer_label_get(proc_ucred( + sopt->sopt_p), so, &extmac)) != 0) + break; + + error = sooptcopyout(sopt, &extmac, sizeof (extmac)); +#else + error = EOPNOTSUPP; +#endif /* MAC_SOCKET */ + break; + +#ifdef __APPLE_API_PRIVATE + case SO_UPCALLCLOSEWAIT: + optval = (so->so_flags & SOF_UPCALLCLOSEWAIT); + goto integer; +#endif + case SO_RANDOMPORT: + optval = (so->so_flags & SOF_BINDRANDOMPORT); + goto integer; + + case SO_NP_EXTENSIONS: { + struct so_np_extensions sonpx; + + sonpx.npx_flags = (so->so_flags & SOF_NPX_SETOPTSHUT) ? + SONPX_SETOPTSHUT : 0; + sonpx.npx_mask = SONPX_MASK_VALID; + + error = sooptcopyout(sopt, &sonpx, + sizeof (struct so_np_extensions)); + break; + } + + case SO_TRAFFIC_CLASS: + optval = so->so_traffic_class; + goto integer; + + case SO_RECV_TRAFFIC_CLASS: + optval = (so->so_flags & SOF_RECV_TRAFFIC_CLASS); + goto integer; + + case SO_TRAFFIC_CLASS_STATS: + error = sooptcopyout(sopt, &so->so_tc_stats, + sizeof (so->so_tc_stats)); + break; + + case SO_TRAFFIC_CLASS_DBG: + error = sogetopt_tcdbg(so, sopt); + break; + + case SO_PRIVILEGED_TRAFFIC_CLASS: + optval = (so->so_flags & SOF_PRIVILEGED_TRAFFIC_CLASS); + goto integer; + + case SO_DEFUNCTOK: + optval = !(so->so_flags & SOF_NODEFUNCT); + goto integer; + + case SO_ISDEFUNCT: + optval = (so->so_flags & SOF_DEFUNCT); + goto integer; + + case SO_OPPORTUNISTIC: + optval = so_get_opportunistic(so); + goto integer; + + case SO_FLUSH: + /* This option is not gettable */ + error = EINVAL; + break; + + case SO_RECV_ANYIF: + optval = so_get_recv_anyif(so); + goto integer; + + case SO_TRAFFIC_MGT_BACKGROUND: + /* This option is handled by lower layer(s) */ + if (so->so_proto != NULL && + so->so_proto->pr_ctloutput != NULL) { + (void) so->so_proto->pr_ctloutput(so, sopt); + } + break; + +#if FLOW_DIVERT + case SO_FLOW_DIVERT_TOKEN: + error = flow_divert_token_get(so, sopt); + break; +#endif /* FLOW_DIVERT */ default: error = ENOPROTOOPT; break; } - return (error); } -} - -#ifdef __APPLE__ -/* - * Network filter support - */ -/* Run the list of filters, creating extension control blocks */ -sfilter_init(register struct socket *so) -{ struct kextcb *kp, **kpp; - struct protosw *prp; - struct NFDescriptor *nfp; - - prp = so->so_proto; - nfp = prp->pr_sfilter.tqh_first; /* non-null */ - kpp = &so->so_ext; - kp = NULL; - while (nfp) - { MALLOC(kp, struct kextcb *, sizeof(*kp), - M_TEMP, M_WAITOK); - if (kp == NULL) - return(ENOBUFS); /* so_free will clean up */ - *kpp = kp; - kpp = &kp->e_next; - kp->e_next = NULL; - kp->e_fcb = NULL; - kp->e_nfd = nfp; - kp->e_soif = nfp->nf_soif; - kp->e_sout = nfp->nf_soutil; - /* - * Ignore return value for create - * Everyone gets a chance at startup - */ - if (kp->e_soif && kp->e_soif->sf_socreate) - (*kp->e_soif->sf_socreate)(so, prp, kp); - nfp = nfp->nf_next.tqe_next; - } - return(0); +out: + if (dolock) + socket_unlock(so, 1); + return (error); } /* - * Run the list of filters, freeing extension control blocks - * Assumes the soif/soutil blocks have been handled. + * The size limits on our soopt_getm is different from that on FreeBSD. + * We limit the size of options to MCLBYTES. This will have to change + * if we need to define options that need more space than MCLBYTES. */ -sfilter_term(struct socket *so) -{ struct kextcb *kp, *kp1; - - kp = so->so_ext; - while (kp) - { kp1 = kp->e_next; - /* - * Ignore return code on termination; everyone must - * get terminated. - */ - if (kp->e_soif && kp->e_soif->sf_sofree) - kp->e_soif->sf_sofree(so, kp); - FREE(kp, M_TEMP); - kp = kp1; - } - return(0); -} -#endif __APPLE__ - -/* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ int soopt_getm(struct sockopt *sopt, struct mbuf **mp) { struct mbuf *m, *m_prev; int sopt_size = sopt->sopt_valsize; + int how; - if (sopt_size > MAX_SOOPTGETM_SIZE) - return EMSGSIZE; + if (sopt_size <= 0 || sopt_size > MCLBYTES) + return (EMSGSIZE); - MGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT, MT_DATA); - if (m == 0) - return ENOBUFS; + how = sopt->sopt_p != kernproc ? M_WAIT : M_DONTWAIT; + MGET(m, how, MT_DATA); + if (m == NULL) + return (ENOBUFS); if (sopt_size > MLEN) { - MCLGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT); + MCLGET(m, how); if ((m->m_flags & M_EXT) == 0) { m_free(m); - return ENOBUFS; + return (ENOBUFS); } m->m_len = min(MCLBYTES, sopt_size); } else { @@ -2337,17 +4164,18 @@ soopt_getm(struct sockopt *sopt, struct mbuf **mp) *mp = m; m_prev = m; - while (sopt_size) { - MGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT, MT_DATA); - if (m == 0) { + while (sopt_size > 0) { + MGET(m, how, MT_DATA); + if (m == NULL) { m_freem(*mp); - return ENOBUFS; + return (ENOBUFS); } if (sopt_size > MLEN) { - MCLGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT); + MCLGET(m, how); if ((m->m_flags & M_EXT) == 0) { m_freem(*mp); - return ENOBUFS; + m_freem(m); + return (ENOBUFS); } m->m_len = min(MCLBYTES, sopt_size); } else { @@ -2357,101 +4185,100 @@ soopt_getm(struct sockopt *sopt, struct mbuf **mp) m_prev->m_next = m; m_prev = m; } - return 0; + return (0); } -/* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ +/* copyin sopt data into mbuf chain */ int soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) { struct mbuf *m0 = m; - if (sopt->sopt_val == NULL) - return 0; + if (sopt->sopt_val == USER_ADDR_NULL) + return (0); while (m != NULL && sopt->sopt_valsize >= m->m_len) { - if (sopt->sopt_p != NULL) { + if (sopt->sopt_p != kernproc) { int error; error = copyin(sopt->sopt_val, mtod(m, char *), - m->m_len); + m->m_len); if (error != 0) { m_freem(m0); - return(error); + return (error); } - } else - bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); + } else { + bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), + mtod(m, char *), m->m_len); + } sopt->sopt_valsize -= m->m_len; - (caddr_t)sopt->sopt_val += m->m_len; + sopt->sopt_val += m->m_len; m = m->m_next; } - if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ + /* should be allocated enoughly at ip6_sooptmcopyin() */ + if (m != NULL) { panic("soopt_mcopyin"); - return 0; + /* NOTREACHED */ + } + return (0); } -/* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ +/* copyout mbuf chain data into soopt */ int soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) { struct mbuf *m0 = m; size_t valsize = 0; - if (sopt->sopt_val == NULL) - return 0; + if (sopt->sopt_val == USER_ADDR_NULL) + return (0); while (m != NULL && sopt->sopt_valsize >= m->m_len) { - if (sopt->sopt_p != NULL) { + if (sopt->sopt_p != kernproc) { int error; error = copyout(mtod(m, char *), sopt->sopt_val, - m->m_len); + m->m_len); if (error != 0) { m_freem(m0); - return(error); + return (error); } - } else - bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); - sopt->sopt_valsize -= m->m_len; - (caddr_t)sopt->sopt_val += m->m_len; - valsize += m->m_len; - m = m->m_next; + } else { + bcopy(mtod(m, char *), + CAST_DOWN(caddr_t, sopt->sopt_val), m->m_len); + } + sopt->sopt_valsize -= m->m_len; + sopt->sopt_val += m->m_len; + valsize += m->m_len; + m = m->m_next; } if (m != NULL) { /* enough soopt buffer should be given from user-land */ m_freem(m0); - return(EINVAL); + return (EINVAL); } sopt->sopt_valsize = valsize; - return 0; + return (0); } void -sohasoutofband(so) - register struct socket *so; +sohasoutofband(struct socket *so) { - struct proc *p; - struct kextcb *kp; - - kp = sotokextcb(so); - while (kp) { - if (kp->e_soif && kp->e_soif->sf_sohasoutofband) { - if ((*kp->e_soif->sf_sohasoutofband)(so, kp)) - return; - } - kp = kp->e_next; - } if (so->so_pgid < 0) gsignal(-so->so_pgid, SIGURG); - else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0) - psignal(p, SIGURG); + else if (so->so_pgid > 0) + proc_signal(so->so_pgid, SIGURG); selwakeup(&so->so_rcv.sb_sel); } int -sopoll(struct socket *so, int events, struct ucred *cred, void * wql) +sopoll(struct socket *so, int events, kauth_cred_t cred, void * wql) { +#pragma unused(cred) struct proc *p = current_proc(); int revents = 0; - int s = splnet(); + + socket_lock(so, 1); + so_update_last_owner_locked(so, PROC_NULL); + so_update_policy(so); if (events & (POLLIN | POLLRDNORM)) if (soreadable(so)) @@ -2467,130 +4294,1073 @@ sopoll(struct socket *so, int events, struct ucred *cred, void * wql) if (revents == 0) { if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { - /* Darwin sets the flag first, BSD calls selrecord first */ + /* + * Darwin sets the flag first, + * BSD calls selrecord first + */ so->so_rcv.sb_flags |= SB_SEL; selrecord(p, &so->so_rcv.sb_sel, wql); } if (events & (POLLOUT | POLLWRNORM)) { - /* Darwin sets the flag first, BSD calls selrecord first */ + /* + * Darwin sets the flag first, + * BSD calls selrecord first + */ so->so_snd.sb_flags |= SB_SEL; selrecord(p, &so->so_snd.sb_sel, wql); } } - splx(s); + socket_unlock(so, 1); return (revents); } - int -soo_kqfilter(struct file *fp, struct knote *kn, struct proc *p) +soo_kqfilter(struct fileproc *fp, struct knote *kn, vfs_context_t ctx) { - struct socket *so = (struct socket *)kn->kn_fp->f_data; - struct sockbuf *sb; - int s; +#pragma unused(fp) +#if !CONFIG_MACF_SOCKET +#pragma unused(ctx) +#endif /* MAC_SOCKET */ + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + struct klist *skl; + + socket_lock(so, 1); + so_update_last_owner_locked(so, PROC_NULL); + so_update_policy(so); + +#if CONFIG_MACF_SOCKET + if (mac_socket_check_kqfilter(proc_ucred(vfs_context_proc(ctx)), + kn, so) != 0) { + socket_unlock(so, 1); + return (1); + } +#endif /* MAC_SOCKET */ switch (kn->kn_filter) { case EVFILT_READ: - if (so->so_options & SO_ACCEPTCONN) - kn->kn_fop = &solisten_filtops; - else - kn->kn_fop = &soread_filtops; - sb = &so->so_rcv; + kn->kn_fop = &soread_filtops; + skl = &so->so_rcv.sb_sel.si_note; break; case EVFILT_WRITE: kn->kn_fop = &sowrite_filtops; - sb = &so->so_snd; + skl = &so->so_snd.sb_sel.si_note; + break; + case EVFILT_SOCK: + kn->kn_fop = &sock_filtops; + skl = &so->so_klist; break; default: + socket_unlock(so, 1); return (1); } - if (sb->sb_sel.si_flags & SI_INITED) - return (1); - - s = splnet(); - if (KNOTE_ATTACH(&sb->sb_sel.si_note, kn)) - sb->sb_flags |= SB_KNOTE; - splx(s); + if (KNOTE_ATTACH(skl, kn)) { + switch (kn->kn_filter) { + case EVFILT_READ: + so->so_rcv.sb_flags |= SB_KNOTE; + break; + case EVFILT_WRITE: + so->so_snd.sb_flags |= SB_KNOTE; + break; + case EVFILT_SOCK: + so->so_flags |= SOF_KNOTE; + break; + default: + socket_unlock(so, 1); + return (1); + } + } + socket_unlock(so, 1); return (0); } static void filt_sordetach(struct knote *kn) { - struct socket *so = (struct socket *)kn->kn_fp->f_data; - int s = splnet(); + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; - if (so->so_rcv.sb_flags & SB_KNOTE && - !(so->so_rcv.sb_sel.si_flags & SI_INITED)) + socket_lock(so, 1); + if (so->so_rcv.sb_flags & SB_KNOTE) if (KNOTE_DETACH(&so->so_rcv.sb_sel.si_note, kn)) so->so_rcv.sb_flags &= ~SB_KNOTE; - splx(s); + socket_unlock(so, 1); } /*ARGSUSED*/ static int filt_soread(struct knote *kn, long hint) { - struct socket *so = (struct socket *)kn->kn_fp->f_data; + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; - kn->kn_data = so->so_rcv.sb_cc; - if (so->so_state & SS_CANTRCVMORE) { - kn->kn_flags |= EV_EOF; - kn->kn_fflags = so->so_error; - return (1); + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_lock(so, 1); + + if (so->so_options & SO_ACCEPTCONN) { + int isempty; + + /* + * Radar 6615193 handle the listen case dynamically + * for kqueue read filter. This allows to call listen() + * after registering the kqueue EVFILT_READ. + */ + + kn->kn_data = so->so_qlen; + isempty = ! TAILQ_EMPTY(&so->so_comp); + + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_unlock(so, 1); + + return (isempty); + } + + /* socket isn't a listener */ + + kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl; + + if (so->so_oobmark) { + if (kn->kn_flags & EV_OOBAND) { + kn->kn_data -= so->so_oobmark; + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_unlock(so, 1); + return (1); + } + kn->kn_data = so->so_oobmark; + kn->kn_flags |= EV_OOBAND; + } else { + if (so->so_state & SS_CANTRCVMORE) { + kn->kn_flags |= EV_EOF; + kn->kn_fflags = so->so_error; + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_unlock(so, 1); + return (1); + } + } + + if (so->so_state & SS_RCVATMARK) { + if (kn->kn_flags & EV_OOBAND) { + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_unlock(so, 1); + return (1); + } + kn->kn_flags |= EV_OOBAND; + } else if (kn->kn_flags & EV_OOBAND) { + kn->kn_data = 0; + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_unlock(so, 1); + return (0); } - if (so->so_error) /* temporary udp error */ + + if (so->so_error) { /* temporary udp error */ + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_unlock(so, 1); return (1); - if (kn->kn_sfflags & NOTE_LOWAT) - return (kn->kn_data >= kn->kn_sdata); - return (kn->kn_data >= so->so_rcv.sb_lowat); + } + + int64_t lowwat = so->so_rcv.sb_lowat; + if (kn->kn_sfflags & NOTE_LOWAT) { + if (kn->kn_sdata > so->so_rcv.sb_hiwat) + lowwat = so->so_rcv.sb_hiwat; + else if (kn->kn_sdata > lowwat) + lowwat = kn->kn_sdata; + } + + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_unlock(so, 1); + + return ((kn->kn_flags & EV_OOBAND) || kn->kn_data >= lowwat); } static void filt_sowdetach(struct knote *kn) { - struct socket *so = (struct socket *)kn->kn_fp->f_data; - int s = splnet(); + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + socket_lock(so, 1); - if(so->so_snd.sb_flags & SB_KNOTE && - !(so->so_snd.sb_sel.si_flags & SI_INITED)) + if (so->so_snd.sb_flags & SB_KNOTE) if (KNOTE_DETACH(&so->so_snd.sb_sel.si_note, kn)) so->so_snd.sb_flags &= ~SB_KNOTE; - splx(s); + socket_unlock(so, 1); +} + +int +so_wait_for_if_feedback(struct socket *so) +{ + if ((SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) && + (so->so_state & SS_ISCONNECTED)) { + struct inpcb *inp = sotoinpcb(so); + if (INP_WAIT_FOR_IF_FEEDBACK(inp)) + return (1); + } + return (0); } /*ARGSUSED*/ static int filt_sowrite(struct knote *kn, long hint) { - struct socket *so = (struct socket *)kn->kn_fp->f_data; + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + int ret = 0; + + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_lock(so, 1); kn->kn_data = sbspace(&so->so_snd); if (so->so_state & SS_CANTSENDMORE) { - kn->kn_flags |= EV_EOF; + kn->kn_flags |= EV_EOF; kn->kn_fflags = so->so_error; - return (1); + ret = 1; + goto out; + } + if (so->so_error) { /* temporary udp error */ + ret = 1; + goto out; } - if (so->so_error) /* temporary udp error */ - return (1); if (((so->so_state & SS_ISCONNECTED) == 0) && - (so->so_proto->pr_flags & PR_CONNREQUIRED)) - return (0); - if (kn->kn_sfflags & NOTE_LOWAT) - return (kn->kn_data >= kn->kn_sdata); - return (kn->kn_data >= so->so_snd.sb_lowat); + (so->so_proto->pr_flags & PR_CONNREQUIRED)) { + ret = 0; + goto out; + } + int64_t lowwat = so->so_snd.sb_lowat; + if (kn->kn_sfflags & NOTE_LOWAT) { + if (kn->kn_sdata > so->so_snd.sb_hiwat) + lowwat = so->so_snd.sb_hiwat; + else if (kn->kn_sdata > lowwat) + lowwat = kn->kn_sdata; + } + if (kn->kn_data >= lowwat) { + if ((so->so_flags & SOF_NOTSENT_LOWAT) != 0) { + ret = tcp_notsent_lowat_check(so); + } else { + ret = 1; + } + } + if (so_wait_for_if_feedback(so)) + ret = 0; +out: + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_unlock(so, 1); + return (ret); +} + +static void +filt_sockdetach(struct knote *kn) +{ + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + socket_lock(so, 1); + + if ((so->so_flags & SOF_KNOTE) != 0) + if (KNOTE_DETACH(&so->so_klist, kn)) + so->so_flags &= ~SOF_KNOTE; + socket_unlock(so, 1); } -/*ARGSUSED*/ static int -filt_solisten(struct knote *kn, long hint) +filt_sockev(struct knote *kn, long hint) +{ + int ret = 0, locked = 0; + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + long ev_hint = (hint & SO_FILT_HINT_EV); + + if ((hint & SO_FILT_HINT_LOCKED) == 0) { + socket_lock(so, 1); + locked = 1; + } + + if (ev_hint & SO_FILT_HINT_CONNRESET) { + if (kn->kn_sfflags & NOTE_CONNRESET) + kn->kn_fflags |= NOTE_CONNRESET; + } + if (ev_hint & SO_FILT_HINT_TIMEOUT) { + if (kn->kn_sfflags & NOTE_TIMEOUT) + kn->kn_fflags |= NOTE_TIMEOUT; + } + if (ev_hint & SO_FILT_HINT_NOSRCADDR) { + if (kn->kn_sfflags & NOTE_NOSRCADDR) + kn->kn_fflags |= NOTE_NOSRCADDR; + } + if (ev_hint & SO_FILT_HINT_IFDENIED) { + if ((kn->kn_sfflags & NOTE_IFDENIED)) + kn->kn_fflags |= NOTE_IFDENIED; + } + if (ev_hint & SO_FILT_HINT_KEEPALIVE) { + if (kn->kn_sfflags & NOTE_KEEPALIVE) + kn->kn_fflags |= NOTE_KEEPALIVE; + } + if (ev_hint & SO_FILT_HINT_ADAPTIVE_WTIMO) { + if (kn->kn_sfflags & NOTE_ADAPTIVE_WTIMO) + kn->kn_fflags |= NOTE_ADAPTIVE_WTIMO; + } + if (ev_hint & SO_FILT_HINT_ADAPTIVE_RTIMO) { + if (kn->kn_sfflags & NOTE_ADAPTIVE_RTIMO) + kn->kn_fflags |= NOTE_ADAPTIVE_RTIMO; + } + if (ev_hint & SO_FILT_HINT_CONNECTED) { + if (kn->kn_sfflags & NOTE_CONNECTED) + kn->kn_fflags |= NOTE_CONNECTED; + } + if (ev_hint & SO_FILT_HINT_DISCONNECTED) { + if (kn->kn_sfflags & NOTE_DISCONNECTED) + kn->kn_fflags |= NOTE_DISCONNECTED; + } + if (ev_hint & SO_FILT_HINT_CONNINFO_UPDATED) { + if (so->so_proto != NULL && + (so->so_proto->pr_flags & PR_EVCONNINFO) && + (kn->kn_sfflags & NOTE_CONNINFO_UPDATED)) + kn->kn_fflags |= NOTE_CONNINFO_UPDATED; + } + + if ((kn->kn_sfflags & NOTE_READCLOSED) && + (so->so_state & SS_CANTRCVMORE)) + kn->kn_fflags |= NOTE_READCLOSED; + + if ((kn->kn_sfflags & NOTE_WRITECLOSED) && + (so->so_state & SS_CANTSENDMORE)) + kn->kn_fflags |= NOTE_WRITECLOSED; + + if ((kn->kn_sfflags & NOTE_SUSPEND) && + ((ev_hint & SO_FILT_HINT_SUSPEND) || + (so->so_flags & SOF_SUSPENDED))) { + kn->kn_fflags &= ~(NOTE_SUSPEND | NOTE_RESUME); + kn->kn_fflags |= NOTE_SUSPEND; + } + + if ((kn->kn_sfflags & NOTE_RESUME) && + ((ev_hint & SO_FILT_HINT_RESUME) || + (so->so_flags & SOF_SUSPENDED) == 0)) { + kn->kn_fflags &= ~(NOTE_SUSPEND | NOTE_RESUME); + kn->kn_fflags |= NOTE_RESUME; + } + + if (so->so_error != 0) { + ret = 1; + kn->kn_data = so->so_error; + kn->kn_flags |= EV_EOF; + } else { + get_sockev_state(so, (u_int32_t *)&(kn->kn_data)); + } + + if (kn->kn_fflags != 0) + ret = 1; + + if (locked) + socket_unlock(so, 1); + + return (ret); +} + +void +get_sockev_state(struct socket *so, u_int32_t *statep) { - struct socket *so = (struct socket *)kn->kn_fp->f_data; + u_int32_t state = *(statep); + + if (so->so_state & SS_ISCONNECTED) + state |= SOCKEV_CONNECTED; + else + state &= ~(SOCKEV_CONNECTED); + state |= ((so->so_state & SS_ISDISCONNECTED) ? SOCKEV_DISCONNECTED : 0); + *(statep) = state; +} + +#define SO_LOCK_HISTORY_STR_LEN \ + (2 * SO_LCKDBG_MAX * (2 + (2 * sizeof (void *)) + 1) + 1) - kn->kn_data = so->so_qlen; - return (! TAILQ_EMPTY(&so->so_comp)); +__private_extern__ const char * +solockhistory_nr(struct socket *so) +{ + size_t n = 0; + int i; + static char lock_history_str[SO_LOCK_HISTORY_STR_LEN]; + + bzero(lock_history_str, sizeof (lock_history_str)); + for (i = SO_LCKDBG_MAX - 1; i >= 0; i--) { + n += snprintf(lock_history_str + n, + SO_LOCK_HISTORY_STR_LEN - n, "%p:%p ", + so->lock_lr[(so->next_lock_lr + i) % SO_LCKDBG_MAX], + so->unlock_lr[(so->next_unlock_lr + i) % SO_LCKDBG_MAX]); + } + return (lock_history_str); } +int +socket_lock(struct socket *so, int refcount) +{ + int error = 0; + void *lr_saved; + + lr_saved = __builtin_return_address(0); + + if (so->so_proto->pr_lock) { + error = (*so->so_proto->pr_lock)(so, refcount, lr_saved); + } else { +#ifdef MORE_LOCKING_DEBUG + lck_mtx_assert(so->so_proto->pr_domain->dom_mtx, + LCK_MTX_ASSERT_NOTOWNED); +#endif + lck_mtx_lock(so->so_proto->pr_domain->dom_mtx); + if (refcount) + so->so_usecount++; + so->lock_lr[so->next_lock_lr] = lr_saved; + so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX; + } + + return (error); +} + +int +socket_unlock(struct socket *so, int refcount) +{ + int error = 0; + void *lr_saved; + lck_mtx_t *mutex_held; + + lr_saved = __builtin_return_address(0); + + if (so->so_proto == NULL) { + panic("%s: null so_proto so=%p\n", __func__, so); + /* NOTREACHED */ + } + + if (so && so->so_proto->pr_unlock) { + error = (*so->so_proto->pr_unlock)(so, refcount, lr_saved); + } else { + mutex_held = so->so_proto->pr_domain->dom_mtx; +#ifdef MORE_LOCKING_DEBUG + lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); +#endif + so->unlock_lr[so->next_unlock_lr] = lr_saved; + so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; + + if (refcount) { + if (so->so_usecount <= 0) { + panic("%s: bad refcount=%d so=%p (%d, %d, %d) " + "lrh=%s", __func__, so->so_usecount, so, + SOCK_DOM(so), so->so_type, + SOCK_PROTO(so), solockhistory_nr(so)); + /* NOTREACHED */ + } + + so->so_usecount--; + if (so->so_usecount == 0) + sofreelastref(so, 1); + } + lck_mtx_unlock(mutex_held); + } + + return (error); +} + +/* Called with socket locked, will unlock socket */ +void +sofree(struct socket *so) +{ + lck_mtx_t *mutex_held; + + if (so->so_proto->pr_getlock != NULL) + mutex_held = (*so->so_proto->pr_getlock)(so, 0); + else + mutex_held = so->so_proto->pr_domain->dom_mtx; + lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); + + sofreelastref(so, 0); +} + +void +soreference(struct socket *so) +{ + socket_lock(so, 1); /* locks & take one reference on socket */ + socket_unlock(so, 0); /* unlock only */ +} + +void +sodereference(struct socket *so) +{ + socket_lock(so, 0); + socket_unlock(so, 1); +} + +/* + * Set or clear SOF_MULTIPAGES on the socket to enable or disable the + * possibility of using jumbo clusters. Caller must ensure to hold + * the socket lock. + */ +void +somultipages(struct socket *so, boolean_t set) +{ + if (set) + so->so_flags |= SOF_MULTIPAGES; + else + so->so_flags &= ~SOF_MULTIPAGES; +} + +int +so_isdstlocal(struct socket *so) { + + struct inpcb *inp = (struct inpcb *)so->so_pcb; + + if (SOCK_DOM(so) == PF_INET) + return (inaddr_local(inp->inp_faddr)); + else if (SOCK_DOM(so) == PF_INET6) + return (in6addr_local(&inp->in6p_faddr)); + + return (0); +} + +int +sosetdefunct(struct proc *p, struct socket *so, int level, boolean_t noforce) +{ + struct sockbuf *rcv, *snd; + int err = 0, defunct; + + rcv = &so->so_rcv; + snd = &so->so_snd; + + defunct = (so->so_flags & SOF_DEFUNCT); + if (defunct) { + if (!(snd->sb_flags & rcv->sb_flags & SB_DROP)) { + panic("%s: SB_DROP not set", __func__); + /* NOTREACHED */ + } + goto done; + } + + if (so->so_flags & SOF_NODEFUNCT) { + if (noforce) { + err = EOPNOTSUPP; + SODEFUNCTLOG(("%s[%d]: (target pid %d level %d) " + "so 0x%llx [%d,%d] is not eligible for defunct " + "(%d)\n", __func__, proc_selfpid(), proc_pid(p), + level, (uint64_t)VM_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), err)); + return (err); + } + so->so_flags &= ~SOF_NODEFUNCT; + SODEFUNCTLOG(("%s[%d]: (target pid %d level %d) so 0x%llx " + "[%d,%d] defunct by force\n", __func__, proc_selfpid(), + proc_pid(p), level, (uint64_t)VM_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so))); + } + + so->so_flags |= SOF_DEFUNCT; + + /* Prevent further data from being appended to the socket buffers */ + snd->sb_flags |= SB_DROP; + rcv->sb_flags |= SB_DROP; + + /* Flush any existing data in the socket buffers */ + if (rcv->sb_cc != 0) { + rcv->sb_flags &= ~SB_SEL; + selthreadclear(&rcv->sb_sel); + sbrelease(rcv); + } + if (snd->sb_cc != 0) { + snd->sb_flags &= ~SB_SEL; + selthreadclear(&snd->sb_sel); + sbrelease(snd); + } + +done: + SODEFUNCTLOG(("%s[%d]: (target pid %d level %d) so 0x%llx [%d,%d] %s " + "defunct\n", __func__, proc_selfpid(), proc_pid(p), level, + (uint64_t)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), SOCK_TYPE(so), + defunct ? "is already" : "marked as")); + + return (err); +} + +int +sodefunct(struct proc *p, struct socket *so, int level) +{ + struct sockbuf *rcv, *snd; + + if (!(so->so_flags & SOF_DEFUNCT)) { + panic("%s improperly called", __func__); + /* NOTREACHED */ + } + if (so->so_state & SS_DEFUNCT) + goto done; + + rcv = &so->so_rcv; + snd = &so->so_snd; + + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) { + char s[MAX_IPv6_STR_LEN]; + char d[MAX_IPv6_STR_LEN]; + struct inpcb *inp = sotoinpcb(so); + + SODEFUNCTLOG(("%s[%d]: (target pid %d level %d) so 0x%llx [%s " + "%s:%d -> %s:%d] is now defunct [rcv_si 0x%x, snd_si 0x%x, " + "rcv_fl 0x%x, snd_fl 0x%x]\n", __func__, proc_selfpid(), + proc_pid(p), level, (uint64_t)VM_KERNEL_ADDRPERM(so), + (SOCK_TYPE(so) == SOCK_STREAM) ? "TCP" : "UDP", + inet_ntop(SOCK_DOM(so), ((SOCK_DOM(so) == PF_INET) ? + (void *)&inp->inp_laddr.s_addr : (void *)&inp->in6p_laddr), + s, sizeof (s)), ntohs(inp->in6p_lport), + inet_ntop(SOCK_DOM(so), (SOCK_DOM(so) == PF_INET) ? + (void *)&inp->inp_faddr.s_addr : (void *)&inp->in6p_faddr, + d, sizeof (d)), ntohs(inp->in6p_fport), + (uint32_t)rcv->sb_sel.si_flags, + (uint32_t)snd->sb_sel.si_flags, + rcv->sb_flags, snd->sb_flags)); + } else { + SODEFUNCTLOG(("%s[%d]: (target pid %d level %d) so 0x%llx " + "[%d,%d] is now defunct [rcv_si 0x%x, snd_si 0x%x, " + "rcv_fl 0x%x, snd_fl 0x%x]\n", __func__, proc_selfpid(), + proc_pid(p), level, (uint64_t)VM_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), (uint32_t)rcv->sb_sel.si_flags, + (uint32_t)snd->sb_sel.si_flags, rcv->sb_flags, + snd->sb_flags)); + } + + /* + * Unwedge threads blocked on sbwait() and sb_lock(). + */ + sbwakeup(rcv); + sbwakeup(snd); + + if (rcv->sb_flags & SB_LOCK) + sbunlock(rcv, TRUE); /* keep socket locked */ + if (snd->sb_flags & SB_LOCK) + sbunlock(snd, TRUE); /* keep socket locked */ + + /* + * Flush the buffers and disconnect. We explicitly call shutdown + * on both data directions to ensure that SS_CANT{RCV,SEND}MORE + * states are set for the socket. This would also flush out data + * hanging off the receive list of this socket. + */ + (void) soshutdownlock(so, SHUT_RD); + (void) soshutdownlock(so, SHUT_WR); + (void) sodisconnectlocked(so); + + /* + * Explicitly handle connectionless-protocol disconnection + * and release any remaining data in the socket buffers. + */ + if (!(so->so_flags & SS_ISDISCONNECTED)) + (void) soisdisconnected(so); + + if (so->so_error == 0) + so->so_error = EBADF; + + if (rcv->sb_cc != 0) { + rcv->sb_flags &= ~SB_SEL; + selthreadclear(&rcv->sb_sel); + sbrelease(rcv); + } + if (snd->sb_cc != 0) { + snd->sb_flags &= ~SB_SEL; + selthreadclear(&snd->sb_sel); + sbrelease(snd); + } + so->so_state |= SS_DEFUNCT; + +done: + return (0); +} + +__private_extern__ int +so_set_recv_anyif(struct socket *so, int optval) +{ + int ret = 0; + +#if INET6 + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) { +#else + if (SOCK_DOM(so) == PF_INET) { +#endif /* !INET6 */ + if (optval) + sotoinpcb(so)->inp_flags |= INP_RECV_ANYIF; + else + sotoinpcb(so)->inp_flags &= ~INP_RECV_ANYIF; + } + + return (ret); +} + +__private_extern__ int +so_get_recv_anyif(struct socket *so) +{ + int ret = 0; + +#if INET6 + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) { +#else + if (SOCK_DOM(so) == PF_INET) { +#endif /* !INET6 */ + ret = (sotoinpcb(so)->inp_flags & INP_RECV_ANYIF) ? 1 : 0; + } + + return (ret); +} + +int +so_set_restrictions(struct socket *so, uint32_t vals) +{ + int nocell_old, nocell_new; + int ret = 0; + + /* + * Deny-type restrictions are trapdoors; once set they cannot be + * unset for the lifetime of the socket. This allows them to be + * issued by a framework on behalf of the application without + * having to worry that they can be undone. + * + * Note here that socket-level restrictions overrides any protocol + * level restrictions. For instance, SO_RESTRICT_DENY_CELLULAR + * socket restriction issued on the socket has a higher precendence + * than INP_NO_IFT_CELLULAR. The latter is affected by the UUID + * policy PROC_UUID_NO_CELLULAR for unrestricted sockets only, + * i.e. when SO_RESTRICT_DENY_CELLULAR has not been issued. + */ + nocell_old = (so->so_restrictions & SO_RESTRICT_DENY_CELLULAR); + so->so_restrictions |= (vals & (SO_RESTRICT_DENY_IN | + SO_RESTRICT_DENY_OUT | SO_RESTRICT_DENY_CELLULAR)); + nocell_new = (so->so_restrictions & SO_RESTRICT_DENY_CELLULAR); + + /* other than deny cellular, there's nothing more to do */ + if ((nocell_new - nocell_old) == 0) + return (ret); + + /* we can only set, not clear restrictions */ + VERIFY((nocell_new - nocell_old) > 0); + +#if INET6 + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) { +#else + if (SOCK_DOM(so) == PF_INET) { +#endif /* !INET6 */ + /* if deny cellular is now set, do what's needed for INPCB */ + inp_set_nocellular(sotoinpcb(so)); + } + + return (ret); +} + +uint32_t +so_get_restrictions(struct socket *so) +{ + return (so->so_restrictions & (SO_RESTRICT_DENY_IN | + SO_RESTRICT_DENY_OUT | SO_RESTRICT_DENY_CELLULAR)); +} + +struct sockaddr_entry * +sockaddrentry_alloc(int how) +{ + struct sockaddr_entry *se; + + se = (how == M_WAITOK) ? zalloc(se_zone) : zalloc_noblock(se_zone); + if (se != NULL) + bzero(se, se_zone_size); + + return (se); +} + +void +sockaddrentry_free(struct sockaddr_entry *se) +{ + if (se->se_addr != NULL) { + FREE(se->se_addr, M_SONAME); + se->se_addr = NULL; + } + zfree(se_zone, se); +} + +struct sockaddr_entry * +sockaddrentry_dup(const struct sockaddr_entry *src_se, int how) +{ + struct sockaddr_entry *dst_se; + + dst_se = sockaddrentry_alloc(how); + if (dst_se != NULL) { + int len = src_se->se_addr->sa_len; + + MALLOC(dst_se->se_addr, struct sockaddr *, + len, M_SONAME, how | M_ZERO); + if (dst_se->se_addr != NULL) { + bcopy(src_se->se_addr, dst_se->se_addr, len); + } else { + sockaddrentry_free(dst_se); + dst_se = NULL; + } + } + + return (dst_se); +} + +struct sockaddr_list * +sockaddrlist_alloc(int how) +{ + struct sockaddr_list *sl; + + sl = (how == M_WAITOK) ? zalloc(sl_zone) : zalloc_noblock(sl_zone); + if (sl != NULL) { + bzero(sl, sl_zone_size); + TAILQ_INIT(&sl->sl_head); + } + return (sl); +} + +void +sockaddrlist_free(struct sockaddr_list *sl) +{ + struct sockaddr_entry *se, *tse; + + TAILQ_FOREACH_SAFE(se, &sl->sl_head, se_link, tse) { + sockaddrlist_remove(sl, se); + sockaddrentry_free(se); + } + VERIFY(sl->sl_cnt == 0 && TAILQ_EMPTY(&sl->sl_head)); + zfree(sl_zone, sl); +} + +void +sockaddrlist_insert(struct sockaddr_list *sl, struct sockaddr_entry *se) +{ + VERIFY(!(se->se_flags & SEF_ATTACHED)); + se->se_flags |= SEF_ATTACHED; + TAILQ_INSERT_TAIL(&sl->sl_head, se, se_link); + sl->sl_cnt++; + VERIFY(sl->sl_cnt != 0); +} + +void +sockaddrlist_remove(struct sockaddr_list *sl, struct sockaddr_entry *se) +{ + VERIFY(se->se_flags & SEF_ATTACHED); + se->se_flags &= ~SEF_ATTACHED; + VERIFY(sl->sl_cnt != 0); + sl->sl_cnt--; + TAILQ_REMOVE(&sl->sl_head, se, se_link); +} + +struct sockaddr_list * +sockaddrlist_dup(const struct sockaddr_list *src_sl, int how) +{ + struct sockaddr_entry *src_se, *tse; + struct sockaddr_list *dst_sl; + + dst_sl = sockaddrlist_alloc(how); + if (dst_sl == NULL) + return (NULL); + + TAILQ_FOREACH_SAFE(src_se, &src_sl->sl_head, se_link, tse) { + struct sockaddr_entry *dst_se; + + if (src_se->se_addr == NULL) + continue; + + dst_se = sockaddrentry_dup(src_se, how); + if (dst_se == NULL) { + sockaddrlist_free(dst_sl); + return (NULL); + } + + sockaddrlist_insert(dst_sl, dst_se); + } + VERIFY(src_sl->sl_cnt == dst_sl->sl_cnt); + + return (dst_sl); +} + +int +so_set_effective_pid(struct socket *so, int epid, struct proc *p) +{ + struct proc *ep = PROC_NULL; + int error = 0; + + /* pid 0 is reserved for kernel */ + if (epid == 0) { + error = EINVAL; + goto done; + } + + /* + * If this is an in-kernel socket, prevent its delegate + * association from changing unless the socket option is + * coming from within the kernel itself. + */ + if (so->last_pid == 0 && p != kernproc) { + error = EACCES; + goto done; + } + + /* + * If this is issued by a process that's recorded as the + * real owner of the socket, or if the pid is the same as + * the process's own pid, then proceed. Otherwise ensure + * that the issuing process has the necessary privileges. + */ + if (epid != so->last_pid || epid != proc_pid(p)) { + if ((error = priv_check_cred(kauth_cred_get(), + PRIV_NET_PRIVILEGED_SOCKET_DELEGATE, 0))) { + error = EACCES; + goto done; + } + } + + /* Find the process that corresponds to the effective pid */ + if ((ep = proc_find(epid)) == PROC_NULL) { + error = ESRCH; + goto done; + } + + /* + * If a process tries to delegate the socket to itself, then + * there's really nothing to do; treat it as a way for the + * delegate association to be cleared. Note that we check + * the passed-in proc rather than calling proc_selfpid(), + * as we need to check the process issuing the socket option + * which could be kernproc. Given that we don't allow 0 for + * effective pid, it means that a delegated in-kernel socket + * stays delegated during its lifetime (which is probably OK.) + */ + if (epid == proc_pid(p)) { + so->so_flags &= ~SOF_DELEGATED; + so->e_upid = 0; + so->e_pid = 0; + uuid_clear(so->e_uuid); + } else { + so->so_flags |= SOF_DELEGATED; + so->e_upid = proc_uniqueid(ep); + so->e_pid = proc_pid(ep); + proc_getexecutableuuid(ep, so->e_uuid, sizeof (so->e_uuid)); + } + +done: + if (error == 0 && net_io_policy_log) { + uuid_string_t buf; + + uuid_unparse(so->e_uuid, buf); + log(LOG_DEBUG, "%s[%s,%d]: so 0x%llx [%d,%d] epid %d (%s) " + "euuid %s%s\n", __func__, proc_name_address(p), + proc_pid(p), (uint64_t)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), + SOCK_TYPE(so), so->e_pid, proc_name_address(ep), buf, + ((so->so_flags & SOF_DELEGATED) ? " [delegated]" : "")); + } else if (error != 0 && net_io_policy_log) { + log(LOG_ERR, "%s[%s,%d]: so 0x%llx [%d,%d] epid %d (%s) " + "ERROR (%d)\n", __func__, proc_name_address(p), + proc_pid(p), (uint64_t)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), + SOCK_TYPE(so), epid, (ep == PROC_NULL) ? "PROC_NULL" : + proc_name_address(ep), error); + } + + if (ep != PROC_NULL) + proc_rele(ep); + + return (error); +} + +int +so_set_effective_uuid(struct socket *so, uuid_t euuid, struct proc *p) +{ + uuid_string_t buf; + uuid_t uuid; + int error = 0; + + /* UUID must not be all-zeroes (reserved for kernel) */ + if (uuid_is_null(euuid)) { + error = EINVAL; + goto done;; + } + + /* + * If this is an in-kernel socket, prevent its delegate + * association from changing unless the socket option is + * coming from within the kernel itself. + */ + if (so->last_pid == 0 && p != kernproc) { + error = EACCES; + goto done; + } + + /* Get the UUID of the issuing process */ + proc_getexecutableuuid(p, uuid, sizeof (uuid)); + + /* + * If this is issued by a process that's recorded as the + * real owner of the socket, or if the uuid is the same as + * the process's own uuid, then proceed. Otherwise ensure + * that the issuing process has the necessary privileges. + */ + if (uuid_compare(euuid, so->last_uuid) != 0 || + uuid_compare(euuid, uuid) != 0) { + if ((error = priv_check_cred(kauth_cred_get(), + PRIV_NET_PRIVILEGED_SOCKET_DELEGATE, 0))) { + error = EACCES; + goto done; + } + } + + /* + * If a process tries to delegate the socket to itself, then + * there's really nothing to do; treat it as a way for the + * delegate association to be cleared. Note that we check + * the uuid of the passed-in proc rather than that of the + * current process, as we need to check the process issuing + * the socket option which could be kernproc itself. Given + * that we don't allow 0 for effective uuid, it means that + * a delegated in-kernel socket stays delegated during its + * lifetime (which is okay.) + */ + if (uuid_compare(euuid, uuid) == 0) { + so->so_flags &= ~SOF_DELEGATED; + so->e_upid = 0; + so->e_pid = 0; + uuid_clear(so->e_uuid); + } else { + so->so_flags |= SOF_DELEGATED; + /* + * Unlike so_set_effective_pid(), we only have the UUID + * here and the process ID is not known. Inherit the + * real {pid,upid} of the socket. + */ + so->e_upid = so->last_upid; + so->e_pid = so->last_pid; + uuid_copy(so->e_uuid, euuid); + } + +done: + if (error == 0 && net_io_policy_log) { + uuid_unparse(so->e_uuid, buf); + log(LOG_DEBUG, "%s[%s,%d]: so 0x%llx [%d,%d] epid %d " + "euuid %s%s\n", __func__, proc_name_address(p), proc_pid(p), + (uint64_t)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), + SOCK_TYPE(so), so->e_pid, buf, + ((so->so_flags & SOF_DELEGATED) ? " [delegated]" : "")); + } else if (error != 0 && net_io_policy_log) { + uuid_unparse(euuid, buf); + log(LOG_DEBUG, "%s[%s,%d]: so 0x%llx [%d,%d] euuid %s " + "ERROR (%d)\n", __func__, proc_name_address(p), proc_pid(p), + (uint64_t)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), + SOCK_TYPE(so), buf, error); + } + + return (error); +} + +void +netpolicy_post_msg(uint32_t ev_code, struct netpolicy_event_data *ev_data, + uint32_t ev_datalen) +{ + struct kev_msg ev_msg; + + /* + * A netpolicy event always starts with a netpolicy_event_data + * structure, but the caller can provide for a longer event + * structure to post, depending on the event code. + */ + VERIFY(ev_data != NULL && ev_datalen >= sizeof (*ev_data)); + + bzero(&ev_msg, sizeof (ev_msg)); + ev_msg.vendor_code = KEV_VENDOR_APPLE; + ev_msg.kev_class = KEV_NETWORK_CLASS; + ev_msg.kev_subclass = KEV_NETPOLICY_SUBCLASS; + ev_msg.event_code = ev_code; + + ev_msg.dv[0].data_ptr = ev_data; + ev_msg.dv[0].data_length = ev_datalen; + + kev_post_msg(&ev_msg); +}