X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/d1ecb069dfe24481e4a83f44cb5217a2b06746d7..a1c7dba18ef36983396c282fe85292db066e39db:/bsd/kern/uipc_socket.c diff --git a/bsd/kern/uipc_socket.c b/bsd/kern/uipc_socket.c index 666f99c46..6b57b3cf8 100644 --- a/bsd/kern/uipc_socket.c +++ b/bsd/kern/uipc_socket.c @@ -1,8 +1,8 @@ /* - * Copyright (c) 1998-2008 Apple Inc. All rights reserved. + * Copyright (c) 1998-2014 Apple Inc. All rights reserved. * * @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 @@ -11,10 +11,10 @@ * 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, @@ -22,7 +22,7 @@ * 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) 1995 NeXT Computer, Inc. All Rights Reserved */ @@ -59,7 +59,6 @@ * 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 @@ -88,41 +87,62 @@ #include #include #include +#include #include +#include #include #include #include #include +#include +#include #include +#include +#include +#include #include #include +#include +#include +#include #include #include #include #include #include #include +#include +#include +#include +#include #if CONFIG_MACF #include #include #endif /* MAC */ -int so_cache_hw = 0; -int so_cache_timeouts = 0; -int so_cache_max_freed = 0; -int cached_sock_count = 0; -__private_extern__ int max_cached_sock_count = MAX_CACHED_SOCKETS; -struct socket *socket_cache_head = 0; -struct socket *socket_cache_tail = 0; -u_int32_t so_cache_time = 0; -int so_cache_init_done = 0; -struct zone *so_cache_zone; - -static lck_grp_t *so_cache_mtx_grp; -static lck_attr_t *so_cache_mtx_attr; +#if MULTIPATH +#include +#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; -lck_mtx_t *so_cache_mtx; +static lck_mtx_t *so_cache_mtx; #include @@ -130,27 +150,39 @@ 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 *sopt, struct timeval * tv_p); - -static int -sooptcopyout_timeval(struct sockopt *sopt, const struct timeval * tv_p); +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, + .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, + .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, +}; + +SYSCTL_DECL(_kern_ipc); + #define EVEN_MORE_LOCKING_DEBUG 0 + int socket_debug = 0; -int socket_zone = M_SOCKET; +SYSCTL_INT(_kern_ipc, OID_AUTO, socket_debug, + CTLFLAG_RW | CTLFLAG_LOCKED, &socket_debug, 0, ""); + +static int socket_zone = M_SOCKET; so_gen_t so_gencnt; /* generation count for sockets */ MALLOC_DEFINE(M_SONAME, "soname", "socket name"); @@ -161,32 +193,33 @@ MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); #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_SOSEND_LIST NETDBG_CODE(DBG_NETSOCK, (4 << 8) | 3) #define DBG_FNC_SORECEIVE NETDBG_CODE(DBG_NETSOCK, (8 << 8)) +#define DBG_FNC_SORECEIVE_LIST NETDBG_CODE(DBG_NETSOCK, (8 << 8) | 3) #define DBG_FNC_SOSHUTDOWN NETDBG_CODE(DBG_NETSOCK, (9 << 8)) #define MAX_SOOPTGETM_SIZE (128 * MCLBYTES) - -SYSCTL_DECL(_kern_ipc); - int somaxconn = SOMAXCONN; -SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, &somaxconn, 0, ""); +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, &sosendjcl, 0, ""); +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 @@ -200,61 +233,75 @@ SYSCTL_INT(_kern_ipc, OID_AUTO, sosendjcl, CTLFLAG_RW, &sosendjcl, 0, ""); * 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, - &sosendjcl_ignore_capab, 0, ""); +SYSCTL_INT(_kern_ipc, OID_AUTO, sosendjcl_ignore_capab, + CTLFLAG_RW | CTLFLAG_LOCKED, &sosendjcl_ignore_capab, 0, ""); -/* - * Socket operation routines. - * These routines are called by the routines in - * sys_socket.c or from a system process, and - * implement the semantics of socket operations by - * switching out to the protocol specific routines. - */ +int sosendbigcl_ignore_capab = 0; +SYSCTL_INT(_kern_ipc, OID_AUTO, sosendbigcl_ignore_capab, + CTLFLAG_RW | CTLFLAG_LOCKED, &sosendbigcl_ignore_capab, 0, ""); + +int sodefunctlog = 0; +SYSCTL_INT(_kern_ipc, OID_AUTO, sodefunctlog, CTLFLAG_RW | CTLFLAG_LOCKED, + &sodefunctlog, 0, ""); -/* sys_generic.c */ -extern void postevent(struct socket *, struct sockbuf *, int); -extern void evsofree(struct socket *); +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"); + +int sorestrictsend = 1; +SYSCTL_INT(_kern_ipc, OID_AUTO, sorestrictsend, CTLFLAG_RW | CTLFLAG_LOCKED, + &sorestrictsend, 0, "Enable outbound interface restrictions"); + +extern struct inpcbinfo tcbinfo; /* TODO: these should be in header file */ extern int get_inpcb_str_size(void); extern int get_tcp_str_size(void); -extern struct domain *pffinddomain(int); -extern struct protosw *pffindprotonotype(int, int); -extern int soclose_locked(struct socket *); -extern int soo_kqfilter(struct fileproc *, struct knote *, struct proc *); -extern int uthread_get_background_state(uthread_t); +static unsigned int sl_zone_size; /* size of sockaddr_list */ +static struct zone *sl_zone; /* zone for sockaddr_list */ -#ifdef __APPLE__ +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; -static int sodelayed_copy(struct socket *, struct uio *, struct mbuf **, int *); +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 *); -static void so_cache_timer(void *); - -void soclose_wait_locked(struct socket *so); -int so_isdstlocal(struct socket *so); +/* + * 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) { - vm_size_t str_size; + _CASSERT(sizeof(so_gencnt) == sizeof(uint64_t)); + VERIFY(IS_P2ALIGNED(&so_gencnt, sizeof(uint32_t))); - if (so_cache_init_done) { + if (socketinit_done) { printf("socketinit: already called...\n"); return; } + socketinit_done = 1; - PE_parse_boot_argn("socket_debug", &socket_debug, sizeof (socket_debug)); + PE_parse_boot_argn("socket_debug", &socket_debug, + sizeof (socket_debug)); /* * 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); @@ -263,62 +310,72 @@ socketinit(void) */ so_cache_mtx_attr = lck_attr_alloc_init(); - so_cache_init_done = 1; - /* 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); - if (so_cache_mtx == NULL) - return; /* we're hosed... */ + so_cache_zone_element_size = (vm_size_t)(sizeof (struct socket) + 4 + + get_inpcb_str_size() + 4 + get_tcp_str_size()); - str_size = (vm_size_t)(sizeof (struct socket) + 4 + - get_inpcb_str_size() + 4 + get_tcp_str_size()); + 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); - so_cache_zone = zinit(str_size, 120000*str_size, 8192, "socache zone"); -#if TEMPDEBUG - printf("cached_sock_alloc -- so_cache_zone size is %x\n", str_size); -#endif - timeout(so_cache_timer, NULL, (SO_CACHE_FLUSH_INTERVAL * hz)); + 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); - so_cache_zone_element_size = str_size; + in_pcbinit(); sflt_init(); + socket_tclass_init(); +#if MULTIPATH + mp_pcbinit(); +#endif /* MULTIPATH */ } static void cached_sock_alloc(struct socket **so, int waitok) { caddr_t temp; - register uintptr_t offset; + uintptr_t offset; lck_mtx_lock(so_cache_mtx); - if (cached_sock_count) { - cached_sock_count--; - *so = socket_cache_head; - if (*so == 0) - panic("cached_sock_alloc: cached sock is null"); + if (!STAILQ_EMPTY(&so_cache_head)) { + VERIFY(cached_sock_count > 0); - socket_cache_head = socket_cache_head->cache_next; - if (socket_cache_head) - socket_cache_head->cache_prev = 0; - else - socket_cache_tail = 0; + *so = STAILQ_FIRST(&so_cache_head); + STAILQ_REMOVE_HEAD(&so_cache_head, so_cache_ent); + STAILQ_NEXT((*so), so_cache_ent) = NULL; + cached_sock_count--; lck_mtx_unlock(so_cache_mtx); temp = (*so)->so_saved_pcb; bzero((caddr_t)*so, sizeof (struct socket)); -#if TEMPDEBUG - kprintf("cached_sock_alloc - retreiving cached sock %p - " - "count == %d\n", *so, cached_sock_count); -#endif + (*so)->so_saved_pcb = temp; - (*so)->cached_in_sock_layer = 1; } else { -#if TEMPDEBUG - kprintf("Allocating cached sock %p from memory\n", *so); -#endif lck_mtx_unlock(so_cache_mtx); @@ -327,17 +384,18 @@ cached_sock_alloc(struct socket **so, int waitok) else *so = (struct socket *)zalloc_noblock(so_cache_zone); - if (*so == 0) + if (*so == NULL) 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. + * Define offsets for extra structures into our + * single block of memory. Align extra structures + * on longword boundaries. */ - offset = (uintptr_t) *so; + offset = (uintptr_t)*so; offset += sizeof (struct socket); offset = ALIGN(offset); @@ -347,16 +405,11 @@ cached_sock_alloc(struct socket **so, int waitok) offset = ALIGN(offset); - ((struct inpcb *)(*so)->so_saved_pcb)->inp_saved_ppcb = + ((struct inpcb *)(void *)(*so)->so_saved_pcb)->inp_saved_ppcb = (caddr_t)offset; -#if TEMPDEBUG - kprintf("Allocating cached socket - %p, pcb=%p tcpcb=%p\n", - *so, (*so)->so_saved_pcb, - ((struct inpcb *)(*so)->so_saved_pcb)->inp_saved_ppcb); -#endif } - (*so)->cached_in_sock_layer = 1; + (*so)->cached_in_sock_layer = true; } static void @@ -365,58 +418,80 @@ cached_sock_free(struct socket *so) 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); -#if TEMPDEBUG - kprintf("Freeing overflowed cached socket %p\n", so); -#endif zfree(so_cache_zone, so); } else { -#if TEMPDEBUG - kprintf("Freeing socket %p into cache\n", so); -#endif if (so_cache_hw < cached_sock_count) so_cache_hw = cached_sock_count; - so->cache_next = socket_cache_head; - so->cache_prev = 0; - if (socket_cache_head) - socket_cache_head->cache_prev = so; - else - socket_cache_tail = so; + STAILQ_INSERT_TAIL(&so_cache_head, so, so_cache_ent); so->cache_timestamp = so_cache_time; - socket_cache_head = so; lck_mtx_unlock(so_cache_mtx); } +} -#if TEMPDEBUG - kprintf("Freed cached sock %p into cache - count is %d\n", - so, cached_sock_count); -#endif +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)); + } + proc_pidoriginatoruuid(so->so_vuuid, sizeof(so->so_vuuid)); + } } -static void -so_cache_timer(__unused void *dummy) +void +so_update_policy(struct socket *so) { - register struct socket *p; - register int n_freed = 0; + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) + (void) inp_update_policy(sotoinpcb(so)); +} - lck_mtx_lock(so_cache_mtx); +#if NECP +static void +so_update_necp_policy(struct socket *so, struct sockaddr *override_local_addr, struct sockaddr *override_remote_addr) +{ + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) + inp_update_necp_policy(sotoinpcb(so), override_local_addr, override_remote_addr, 0); +} +#endif /* NECP */ - ++so_cache_time; +boolean_t +so_cache_timer(void) +{ + struct socket *p; + int n_freed = 0; + boolean_t rc = FALSE; - while ((p = socket_cache_tail)) { - if ((so_cache_time - p->cache_timestamp) < SO_CACHE_TIME_LIMIT) + 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; zfree(so_cache_zone, p); @@ -425,11 +500,14 @@ so_cache_timer(__unused void *dummy) break; } } - lck_mtx_unlock(so_cache_mtx); - timeout(so_cache_timer, NULL, (SO_CACHE_FLUSH_INTERVAL * hz)); + /* 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. @@ -451,14 +529,12 @@ soalloc(int waitok, int dom, int type) if (so != NULL) bzero(so, sizeof (*so)); } - /* XXX race condition for reentrant kernel */ -//###LD Atomic add for so_gencnt if (so != NULL) { - so->so_gencnt = ++so_gencnt; + so->so_gencnt = OSIncrementAtomic64((SInt64 *)&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) { + /* Convert waitok to M_WAITOK/M_NOWAIT for MAC Framework. */ + if (mac_socket_label_init(so, !waitok) != 0) { sodealloc(so); return (NULL); } @@ -468,66 +544,77 @@ soalloc(int waitok, int dom, int type) return (so); } -/* - * Returns: 0 Success - * EAFNOSUPPORT - * EPROTOTYPE - * EPROTONOSUPPORT - * ENOBUFS - * :ENOBUFS[AF_UNIX] - * :ENOBUFS[TCP] - * :ENOMEM[TCP] - * :EISCONN[TCP] - * :??? [other protocol families, IPSEC] - */ int -socreate(int dom, struct socket **aso, 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; - thread_t thread; - struct uthread *ut; + 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) { - if (pffinddomain(dom) == NULL) { + if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL) { + if (pffinddomain(dom) == NULL) return (EAFNOSUPPORT); - } if (proto != 0) { - if (pffindprotonotype(dom, proto) != NULL) { + if (pffindprotonotype(dom, proto) != NULL) return (EPROTOTYPE); - } } return (EPROTONOSUPPORT); } if (prp->pr_type != type) return (EPROTOTYPE); so = soalloc(1, dom, type); - if (so == 0) + 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)); + proc_pidoriginatoruuid(so->so_vuuid, sizeof(so->so_vuuid)); + + 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; + } - so->so_uid = kauth_cred_getuid(kauth_cred_get()); + so->so_cred = kauth_cred_proc_ref(p); if (!suser(kauth_cred_get(), NULL)) - so->so_state = SS_PRIV; + so->so_state |= SS_PRIV; so->so_proto = prp; -#ifdef __APPLE__ - so->so_rcv.sb_flags |= SB_RECV; /* XXX */ + so->so_rcv.sb_flags |= SB_RECV; so->so_rcv.sb_so = so->so_snd.sb_so = so; -#endif so->next_lock_lr = 0; so->next_unlock_lr = 0; @@ -535,15 +622,15 @@ socreate(int dom, struct socket **aso, int type, int proto) mac_socket_label_associate(kauth_cred_get(), so); #endif /* MAC_SOCKET */ -//### Attachement will create the per pcb lock if necessary and increase refcount /* - * for creation, make sure it's done before - * socket is inserted in lists + * 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) { + if (error != 0) { /* * Warning: * If so_pcb is not zero, the socket will be leaked, @@ -554,8 +641,8 @@ socreate(int dom, struct socket **aso, int type, int proto) sofreelastref(so, 1); /* will deallocate the socket */ return (error); } -#ifdef __APPLE__ - prp->pr_domain->dom_refs++; + + atomic_add_32(&prp->pr_domain->dom_refs, 1); TAILQ_INIT(&so->so_evlist); /* Attach socket filters for this protocol */ @@ -564,29 +651,89 @@ socreate(int dom, struct socket **aso, int type, int proto) if (tcpconsdebug == 2) so->so_options |= SO_DEBUG; #endif -#endif + so_set_default_traffic_class(so); + /* - * If this is a background thread/task, mark the socket as such. + * If this thread or task is marked to create backgrounded sockets, + * mark the socket as background. */ - thread = current_thread(); - ut = get_bsdthread_info(thread); - if (uthread_get_background_state(ut)) { + 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 = thread; - /* - * In case setpriority(PRIO_DARWIN_THREAD) was called - * on this thread, regulate network (TCP) traffics. - */ - if (ut->uu_flag & UT_BACKGROUND_TRAFFIC_MGT) { - socket_set_traffic_mgt_flags(so, - TRAFFIC_MGT_SO_BG_REGULATE); - } + 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; } + /* + * Entitlements can't be checked at socket creation time except if the + * application requested a feature guarded by a privilege (c.f., socket + * delegation). + * The priv(9) and the Sandboxing APIs are designed with the idea that + * a privilege check should only be triggered by a userland request. + * A privilege check at socket creation time is time consuming and + * could trigger many authorisation error messages from the security + * APIs. + */ + *aso = so; + return (0); } +/* + * Returns: 0 Success + * EAFNOSUPPORT + * EPROTOTYPE + * EPROTONOSUPPORT + * ENOBUFS + * :ENOBUFS[AF_UNIX] + * :ENOBUFS[TCP] + * :ENOMEM[TCP] + * :??? [other protocol families, IPSEC] + */ +int +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] @@ -610,49 +757,42 @@ socreate(int dom, struct socket **aso, int type, int proto) * be returned by the tcp_usr_bind function supplied. */ int -sobind(struct socket *so, struct sockaddr *nam) +sobindlock(struct socket *so, struct sockaddr *nam, int dolock) { struct proc *p = current_proc(); int error = 0; - struct socket_filter_entry *filter; - int filtered = 0; - - socket_lock(so, 1); + if (dolock) + socket_lock(so, 1); + VERIFY(so->so_usecount > 1); + + so_update_last_owner_locked(so, p); + so_update_policy(so); + +#if NECP + so_update_necp_policy(so, nam, NULL); +#endif /* NECP */ + /* - * If this is a bind request on a previously-accepted socket - * that has been marked as inactive, reject it now before - * we go any further. + * 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; } /* Socket filter */ - error = 0; - for (filter = so->so_filt; filter && (error == 0); - filter = filter->sfe_next_onsocket) { - if (filter->sfe_filter->sf_filter.sf_bind) { - if (filtered == 0) { - filtered = 1; - sflt_use(so); - socket_unlock(so, 0); - } - error = filter->sfe_filter->sf_filter. - sf_bind(filter->sfe_cookie, so, nam); - } - } - if (filtered != 0) { - socket_lock(so, 0); - sflt_unuse(so); - } - /* End socket filter */ + error = sflt_bind(so, nam); if (error == 0) error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, p); out: - socket_unlock(so, 1); + if (dolock) + socket_unlock(so, 1); if (error == EJUSTRETURN) error = 0; @@ -663,17 +803,31 @@ out: void sodealloc(struct socket *so) { - so->so_gencnt = ++so_gencnt; + kauth_cred_unref(&so->so_cred); + + /* Remove any filters */ + sflt_termsock(so); + +#if CONTENT_FILTER + cfil_sock_detach(so); +#endif /* CONTENT_FILTER */ + + /* 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 = OSIncrementAtomic64((SInt64 *)&so_gencnt); #if CONFIG_MACF_SOCKET mac_socket_label_destroy(so); #endif /* MAC_SOCKET */ - if (so->cached_in_sock_layer == 1) { + + if (so->cached_in_sock_layer) { cached_sock_free(so); } else { - if (so->cached_in_sock_layer == -1) - panic("sodealloc: double dealloc: so=%p\n", so); - so->cached_in_sock_layer = -1; FREE_ZONE(so, sizeof (*so), so->so_zone); } } @@ -702,10 +856,16 @@ solisten(struct socket *so, int backlog) { struct proc *p = current_proc(); int error = 0; - struct socket_filter_entry *filter; - int filtered = 0; socket_lock(so, 1); + + so_update_last_owner_locked(so, p); + so_update_policy(so); + +#if NECP + so_update_necp_policy(so, NULL, NULL); +#endif /* NECP */ + if (so->so_proto == NULL) { error = EINVAL; goto out; @@ -717,42 +877,30 @@ solisten(struct socket *so, int backlog) /* * If the listen request is made on a socket that is not fully - * disconnected, or on a previously-accepted socket that has - * been marked as inactive, reject the request now. + * 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_DENYIN) != 0) { + if ((so->so_restrictions & SO_RESTRICT_DENY_IN) != 0) { error = EPERM; goto out; } - error = 0; - for (filter = so->so_filt; filter && (error == 0); - filter = filter->sfe_next_onsocket) { - if (filter->sfe_filter->sf_filter.sf_listen) { - if (filtered == 0) { - filtered = 1; - sflt_use(so); - socket_unlock(so, 0); - } - error = filter->sfe_filter->sf_filter. - sf_listen(filter->sfe_cookie, so); - } - } - if (filtered != 0) { - socket_lock(so, 0); - sflt_unuse(so); - } - - if (error == 0) { + error = sflt_listen(so); + if (error == 0) error = (*so->so_proto->pr_usrreqs->pru_listen)(so, p); - } if (error) { if (error == EJUSTRETURN) @@ -792,17 +940,12 @@ sofreelastref(struct socket *so, int dealloc) /* Assume socket is locked */ - /* Remove any filters - may be called more than once */ - sflt_termsock(so); - - if ((!(so->so_flags & SOF_PCBCLEARING)) || - ((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); - so->so_rcv.sb_flags &= ~SB_UPCALL; - so->so_snd.sb_flags &= ~SB_UPCALL; -#endif + so->so_rcv.sb_flags &= ~(SB_SEL|SB_UPCALL); + so->so_snd.sb_flags &= ~(SB_SEL|SB_UPCALL); + so->so_event = sonullevent; return; } if (head != NULL) { @@ -817,12 +960,11 @@ sofreelastref(struct socket *so, int dealloc) * 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); - so->so_rcv.sb_flags &= ~SB_UPCALL; - so->so_snd.sb_flags &= ~SB_UPCALL; -#endif + so->so_rcv.sb_flags &= ~(SB_SEL|SB_UPCALL); + so->so_snd.sb_flags &= ~(SB_SEL|SB_UPCALL); + so->so_event = sonullevent; socket_unlock(head, 1); return; } else { @@ -833,15 +975,19 @@ sofreelastref(struct socket *so, int dealloc) 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); +#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 = sonullevent; if (dealloc) sodealloc(so); @@ -862,12 +1008,12 @@ soclose_wait_locked(struct socket *so) * Double check here and return if there's no outstanding upcall; * otherwise proceed further only if SOF_UPCALLCLOSEWAIT is set. */ - if (!(so->so_flags & SOF_UPCALLINUSE) || - !(so->so_flags & SOF_UPCALLCLOSEWAIT)) + 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_upcall, mutex_held, (PZERO - 1), + (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; @@ -887,10 +1033,25 @@ soclose_locked(struct socket *so) if (so->so_usecount == 0) { panic("soclose: so=%p refcount=0\n", so); + /* NOTREACHED */ } sflt_notify(so, sock_evt_closing, NULL); + if (so->so_upcallusecount) + soclose_wait_locked(so); + +#if CONTENT_FILTER + /* + * We have to wait until the content filters are done + */ + if ((so->so_flags & SOF_CONTENT_FILTER) != 0) { + cfil_sock_close_wait(so); + cfil_sock_is_closed(so); + cfil_sock_detach(so); + } +#endif /* CONTENT_FILTER */ + if ((so->so_options & SO_ACCEPTCONN)) { struct socket *sp, *sonext; int socklock = 0; @@ -901,26 +1062,30 @@ soclose_locked(struct socket *so) */ so->so_options &= ~SO_ACCEPTCONN; - for (sp = TAILQ_FIRST(&so->so_incomp); sp != NULL; sp = sonext) { + for (sp = TAILQ_FIRST(&so->so_incomp); + sp != NULL; sp = sonext) { sonext = TAILQ_NEXT(sp, so_list); - /* Radar 5350314 + /* + * 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) + 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. + /* + * 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; + socklock = 1; } TAILQ_REMOVE(&so->so_incomp, sp, so_list); @@ -933,7 +1098,7 @@ soclose_locked(struct socket *so) (void) soabort(sp); } - if (socklock) + if (socklock) socket_unlock(sp, 1); } @@ -960,7 +1125,7 @@ soclose_locked(struct socket *so) } } } - if (so->so_pcb == 0) { + if (so->so_pcb == NULL) { /* 3915887: mark the socket as ready for dealloc */ so->so_flags |= SOF_PCBCLEARING; goto discard; @@ -998,23 +1163,36 @@ soclose_locked(struct socket *so) } } drop: - if (so->so_usecount == 0) + if (so->so_usecount == 0) { panic("soclose: usecount is zero so=%p\n", so); - if (so->so_pcb && !(so->so_flags & SOF_PCBCLEARING)) { + /* NOTREACHED */ + } + if (so->so_pcb != NULL && !(so->so_flags & SOF_PCBCLEARING)) { int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so); if (error == 0) error = error2; } - if (so->so_usecount <= 0) + 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); return (error); @@ -1026,9 +1204,6 @@ soclose(struct socket *so) int error = 0; socket_lock(so, 1); - if (so->so_flags & SOF_UPCALLINUSE) - soclose_wait_locked(so); - if (so->so_retaincnt == 0) { error = soclose_locked(so); } else { @@ -1083,6 +1258,12 @@ soacceptlock(struct socket *so, struct sockaddr **nam, int dolock) if (dolock) socket_lock(so, 1); + so_update_last_owner_locked(so, PROC_NULL); + so_update_policy(so); +#if NECP + so_update_necp_policy(so, NULL, NULL); +#endif /* NECP */ + if ((so->so_state & SS_NOFDREF) == 0) panic("soaccept: !NOFDREF"); so->so_state &= ~SS_NOFDREF; @@ -1103,15 +1284,13 @@ int soacceptfilter(struct socket *so) { struct sockaddr *local = NULL, *remote = NULL; - struct socket_filter_entry *filter; - int error = 0, filtered = 0; + 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 protect against the - * head or peer going away + * 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 || @@ -1125,29 +1304,7 @@ soacceptfilter(struct socket *so) goto done; } - /* - * At this point, we have a reference on the listening socket - * so we know it won't be going away. Do the same for the newly - * accepted socket while we invoke the accept callback routine. - */ - for (filter = so->so_filt; filter != NULL && error == 0; - filter = filter->sfe_next_onsocket) { - if (filter->sfe_filter->sf_filter.sf_accept != NULL) { - if (!filtered) { - filtered = 1; - sflt_use(so); - socket_unlock(so, 0); - } - error = filter->sfe_filter->sf_filter. - sf_accept(filter->sfe_cookie, - head, so, local, remote); - } - } - - if (filtered) { - socket_lock(so, 0); - sflt_unuse(so); - } + error = sflt_accept(head, so, local, remote); /* * If we get EJUSTRETURN from one of the filters, mark this socket @@ -1156,10 +1313,8 @@ soacceptfilter(struct socket *so) */ if (error == EJUSTRETURN) { error = 0; - so->so_flags |= SOF_DEFUNCT; - /* Prevent data from being appended to the socket buffers */ - so->so_snd.sb_flags |= SB_DROP; - so->so_rcv.sb_flags |= SB_DROP; + (void) sosetdefunct(current_proc(), so, + SHUTDOWN_SOCKET_LEVEL_DISCONNECT_INTERNAL, FALSE); } if (error != 0) { @@ -1206,17 +1361,31 @@ soconnectlock(struct socket *so, struct sockaddr *nam, int dolock) if (dolock) socket_lock(so, 1); + so_update_last_owner_locked(so, p); + so_update_policy(so); + +#if NECP + so_update_necp_policy(so, NULL, nam); +#endif /* NECP */ + /* * 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 (EOPNOTSUPP); + return (error); } - if ((so->so_restrictions & SO_RESTRICT_DENYOUT) != 0) { + if ((so->so_restrictions & SO_RESTRICT_DENY_OUT) != 0) { if (dolock) socket_unlock(so, 1); return (EPERM); @@ -1237,36 +1406,14 @@ soconnectlock(struct socket *so, struct sockaddr *nam, int dolock) * Run connect filter before calling protocol: * - non-blocking connect returns before completion; */ - struct socket_filter_entry *filter; - int filtered = 0; - - error = 0; - for (filter = so->so_filt; filter && (error == 0); - filter = filter->sfe_next_onsocket) { - if (filter->sfe_filter->sf_filter.sf_connect_out) { - if (filtered == 0) { - filtered = 1; - sflt_use(so); - socket_unlock(so, 0); - } - error = filter->sfe_filter->sf_filter. - sf_connect_out(filter->sfe_cookie, so, nam); - } - } - if (filtered != 0) { - socket_lock(so, 0); - sflt_unuse(so); - } - - if (error) { + error = sflt_connectout(so, nam); + if (error != 0) { if (error == EJUSTRETURN) error = 0; - if (dolock) - socket_unlock(so, 1); - return (error); + } else { + error = (*so->so_proto->pr_usrreqs->pru_connect) + (so, nam, p); } - - error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, p); } if (dolock) socket_unlock(so, 1); @@ -1305,27 +1452,85 @@ soconnect2(struct socket *so1, struct socket *so2) } int -sodisconnectlocked(struct socket *so) +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 ((so->so_state & SS_ISCONNECTED) == 0) { - error = ENOTCONN; - goto bad; - } - if (so->so_state & SS_ISDISCONNECTING) { - error = EALREADY; - goto bad; + 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)); + } + return (error); } - error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); + if ((so->so_restrictions & SO_RESTRICT_DENY_OUT) != 0) + return (EPERM); - if (error == 0) { - sflt_notify(so, sock_evt_disconnected, NULL); - } -bad: - return (error); -} + /* + * 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); + } + } + + return (error); +} + +int +sodisconnectlocked(struct socket *so) +{ + int error; + + if ((so->so_state & SS_ISCONNECTED) == 0) { + error = ENOTCONN; + goto bad; + } + if (so->so_state & SS_ISDISCONNECTING) { + 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 @@ -1339,7 +1544,45 @@ sodisconnect(struct socket *so) return (error); } -#define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_DONTWAIT : M_WAIT) +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) { + /* + * 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 @@ -1353,11 +1596,12 @@ sodisconnect(struct socket *so) * sbwait:EINTR * [so_error]:??? */ -static int -sosendcheck(struct socket *so, struct sockaddr *addr, int32_t resid, int32_t clen, - int32_t atomic, int flags, int *sblocked) +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; + int error = 0; int32_t space; int assumelock = 0; @@ -1376,6 +1620,8 @@ restart: } else { error = sblock(&so->so_snd, SBLOCKWAIT(flags)); if (error) { + if (so->so_flags & SOF_DEFUNCT) + goto defunct; return (error); } *sblocked = 1; @@ -1383,16 +1629,33 @@ restart: } /* - * If a send attempt is made on a previously-accepted socket - * that has been marked as inactive (disconnected), reject - * the request. + * 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) - return (ENOTCONN); - - if (so->so_state & SS_CANTSENDMORE) - return (EPIPE); + 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) { +#if CONTENT_FILTER + /* + * Can re-inject data of half closed connections + */ + if ((so->so_state & SS_ISDISCONNECTED) == 0 && + so->so_snd.sb_cfil_thread == current_thread() && + cfil_sock_data_pending(&so->so_snd) != 0) + CFIL_LOG(LOG_INFO, + "so %llx ignore SS_CANTSENDMORE", + (uint64_t)VM_KERNEL_ADDRPERM(so)); + else +#endif /* CONTENT_FILTER */ + return (EPIPE); + } if (so->so_error) { error = so->so_error; so->so_error = 0; @@ -1401,34 +1664,52 @@ restart: 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)) + if (((so->so_state & SS_ISCONFIRMING) == 0) && + (resid != 0 || clen == 0)) { +#if MPTCP + /* + * MPTCP Fast Join sends data before the + * socket is truly connected. + */ + if ((so->so_flags & (SOF_MP_SUBFLOW | + SOF_MPTCP_FASTJOIN)) != + (SOF_MP_SUBFLOW | SOF_MPTCP_FASTJOIN)) +#endif /* MPTCP */ return (ENOTCONN); + } } else if (addr == 0 && !(flags&MSG_HOLD)) { return ((so->so_proto->pr_flags & PR_CONNREQUIRED) ? ENOTCONN : EDESTADDRREQ); } } - space = sbspace(&so->so_snd); + 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)) { + + 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, 1); + 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); } @@ -1497,23 +1778,36 @@ 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 int32_t space, len, resid; + 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 mbuf *control_copy = NULL; - if (uio) { - // LP64todo - fix this! + if (uio != NULL) resid = uio_resid(uio); - } else { + 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); socket_lock(so, 1); + + /* + * Re-injection should not affect process accounting + */ + if ((flags & MSG_SKIPCFIL) == 0) { + so_update_last_owner_locked(so, p); + so_update_policy(so); + +#if NECP + so_update_necp_policy(so, NULL, addr); +#endif /* NECP */ + } + if (so->so_type != SOCK_STREAM && (flags & MSG_OOB) != 0) { error = EOPNOTSUPP; socket_unlock(so, 1); @@ -1527,37 +1821,41 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, * 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 int value as we use + * imin() to set bytes_to_copy -- radr://14558484 */ - if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { + if (resid < 0 || resid > INT_MAX || (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); OSIncrementAtomicLong(&p->p_stats->p_ru.ru_msgsnd); - if (control) + + if (control != NULL) clen = control->m_len; do { error = sosendcheck(so, addr, resid, clen, atomic, flags, - &sblocked); - if (error) { + &sblocked, control); + if (error) goto release; - } + mp = ⊤ - space = sbspace(&so->so_snd) - clen + ((flags & MSG_OOB) ? - 1024 : 0); + 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 { - struct socket_filter_entry *filter; - int filtered; - boolean_t recursive; - if (uio == NULL) { /* * Data is prepackaged in "top". @@ -1569,15 +1867,23 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, int chainlength; int bytes_to_copy; boolean_t jumbocl; + boolean_t bigcl; bytes_to_copy = imin(resid, space); - if (sosendminchain > 0) { + if (sosendminchain > 0) chainlength = 0; - } else { + else chainlength = sosendmaxchain; - } + /* + * Use big 4 KB cluster only when outgoing + * interface does not want 2 LB clusters + */ + bigcl = + !(so->so_flags1 & SOF1_IF_2KCL) || + sosendbigcl_ignore_capab; + /* * Attempt to use larger than system page-size * clusters for large writes only if there is @@ -1586,13 +1892,14 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, */ jumbocl = sosendjcl && njcl > 0 && ((so->so_flags & SOF_MULTIPAGES) || - sosendjcl_ignore_capab); + sosendjcl_ignore_capab) && + bigcl; socket_unlock(so, 0); do { int num_needed; - int hdrs_needed = (top == 0) ? 1 : 0; + int hdrs_needed = (top == NULL) ? 1 : 0; /* * try to maintain a local cache of mbuf @@ -1610,7 +1917,8 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, * haven't yet consumed. */ if (freelist == NULL && - bytes_to_copy > NBPG && jumbocl) { + bytes_to_copy > MBIGCLBYTES && + jumbocl) { num_needed = bytes_to_copy / M16KCLBYTES; @@ -1631,12 +1939,13 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, } if (freelist == NULL && - bytes_to_copy > MCLBYTES) { + bytes_to_copy > MCLBYTES && + bigcl) { num_needed = - bytes_to_copy / NBPG; + bytes_to_copy / MBIGCLBYTES; if ((bytes_to_copy - - (num_needed * NBPG)) >= + (num_needed * MBIGCLBYTES)) >= MINCLSIZE) num_needed++; @@ -1644,7 +1953,7 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, m_getpackets_internal( (unsigned int *)&num_needed, hdrs_needed, M_WAIT, 0, - NBPG); + MBIGCLBYTES); /* * Fall back to cluster size * if allocation failed @@ -1673,7 +1982,7 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, } if (freelist == NULL) { - if (top == 0) + if (top == NULL) MGETHDR(freelist, M_WAIT, MT_DATA); else @@ -1690,7 +1999,7 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, * leave room for protocol * headers in first mbuf. */ - if (atomic && top == 0 && + if (atomic && top == NULL && bytes_to_copy < MHLEN) { MH_ALIGN(freelist, bytes_to_copy); @@ -1743,7 +2052,7 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, if (flags & (MSG_HOLD|MSG_SEND)) { /* Enqueue for later, go away if HOLD */ - register struct mbuf *mb1; + struct mbuf *mb1; if (so->so_temp && (flags & MSG_FLUSH)) { m_freem(so->so_temp); so->so_temp = NULL; @@ -1774,83 +2083,65 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, */ ((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 - */ - recursive = (so->so_send_filt_thread != NULL); - filtered = 0; - error = 0; - for (filter = so->so_filt; filter && (error == 0); - filter = filter->sfe_next_onsocket) { - if (filter->sfe_filter->sf_filter.sf_data_out) { - int so_flags = 0; - if (filtered == 0) { - filtered = 1; - so->so_send_filt_thread = - current_thread(); - sflt_use(so); - socket_unlock(so, 0); - so_flags = - (sendflags & MSG_OOB) ? - sock_data_filt_flag_oob : 0; + (resid <= 0)) ? PRUS_EOF : + /* If there is more to send set PRUS_MORETOCOME */ + (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; + + if ((flags & MSG_SKIPCFIL) == 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; } - error = filter->sfe_filter->sf_filter. - sf_data_out(filter->sfe_cookie, so, - addr, &top, &control, so_flags); + goto release; } - } - - if (filtered) { +#if CONTENT_FILTER /* - * At this point, we've run at least one - * filter. The socket is unlocked as is - * the socket buffer. Clear the recorded - * filter thread only when we are outside - * of a filter's context. This allows for - * a filter to issue multiple inject calls - * from its sf_data_out callback routine. + * Content filter processing */ - socket_lock(so, 0); - sflt_unuse(so); - if (!recursive) - so->so_send_filt_thread = 0; + error = cfil_sock_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 = 0; - top = 0; - } - + control = NULL; + top = NULL; + } goto release; } +#endif /* CONTENT_FILTER */ } - /* - * End Socket filter processing - */ - - if (error == EJUSTRETURN) { - /* A socket filter handled this data */ - error = 0; - } else { - error = (*so->so_proto->pr_usrreqs->pru_send) - (so, sendflags, top, addr, control, p); + 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); } -#ifdef __APPLE__ + error = (*so->so_proto->pr_usrreqs->pru_send) + (so, sendflags, top, addr, control, p); + if (flags & MSG_SEND) so->so_temp = NULL; -#endif + if (dontroute) so->so_options &= ~SO_DONTROUTE; clen = 0; - control = 0; - top = 0; + control = control_copy; + control_copy = NULL; + top = NULL; mp = ⊤ if (error) goto release; @@ -1859,16 +2150,18 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, release: if (sblocked) - sbunlock(&so->so_snd, 0); /* will unlock socket */ + 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) + 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); @@ -1876,161 +2169,459 @@ out: return (error); } -/* - * Implement receive operations on a socket. - * We depend on the way that records are added to the sockbuf - * by sbappend*. In particular, each record (mbufs linked through m_next) - * must begin with an address if the protocol so specifies, - * followed by an optional mbuf or mbufs containing ancillary data, - * and then zero or more mbufs of data. - * In order to avoid blocking network interrupts for the entire time here, - * we splx() while doing the actual copy to user space. - * Although the sockbuf is locked, new data may still be appended, - * and thus we must maintain consistency of the sockbuf during that time. - * - * 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(struct socket *so, struct sockaddr **psa, struct uio *uio, - struct mbuf **mp0, struct mbuf **controlp, int *flagsp) +sosend_list(struct socket *so, struct sockaddr *addr, struct uio **uioarray, + u_int uiocnt, struct mbuf *top, struct mbuf *control, int flags) { - register struct mbuf *m, **mp, *ml = NULL; - register int flags, len, error, offset; - struct protosw *pr = so->so_proto; - struct mbuf *nextrecord; - int moff, type = 0; - int orig_resid = uio_resid(uio); - struct mbuf *free_list; - int delayed_copy_len; - int can_delay; - int need_event; + struct mbuf *m, *freelist = NULL; + user_ssize_t len, resid; + int clen = 0, error, dontroute, mlen; + int atomic = sosendallatonce(so) || top; + int sblocked = 0; struct proc *p = current_proc(); + u_int uiofirst = 0; + u_int uiolast = 0; - // LP64todo - fix this! - 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); - - socket_lock(so, 1); + KERNEL_DEBUG((DBG_FNC_SOSEND_LIST | DBG_FUNC_START), so, uiocnt, + so->so_snd.sb_cc, so->so_snd.sb_lowat, so->so_snd.sb_hiwat); -#ifdef MORE_LOCKING_DEBUG - if (so->so_usecount == 1) - panic("soreceive: so=%x no other reference on socket\n", so); -#endif - mp = mp0; - if (psa) - *psa = 0; - if (controlp) - *controlp = 0; - if (flagsp) - flags = *flagsp &~ MSG_EOR; + if (so->so_type != SOCK_DGRAM) { + error = EINVAL; + goto out; + } + if (atomic == 0) { + error = EINVAL; + goto out; + } + if (so->so_proto->pr_usrreqs->pru_send_list == NULL) { + error = EPROTONOSUPPORT; + goto out; + } + if (flags & ~(MSG_DONTWAIT | MSG_NBIO)) { + error = EINVAL; + goto out; + } + if (uioarray != NULL) + resid = uio_array_resid(uioarray, uiocnt); else - flags = 0; + resid = mbuf_pkt_list_len(top); /* - * If a recv attempt is made on a previously-accepted socket - * that has been marked as inactive (disconnected), reject - * the request. + * 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. + * + * Note: We limit resid to be a positive int value as we use + * imin() to set bytes_to_copy -- radr://14558484 */ - if (so->so_flags & SOF_DEFUNCT) { - struct sockbuf *sb = &so->so_rcv; - - /* - * This socket should have been disconnected and flushed - * prior to being returned from accept; there should be - * no data on its receive list, so panic otherwise. - */ - sb_empty_assert(sb, __func__); - socket_unlock(so, 1); - return (ENOTCONN); + if (resid < 0 || resid > INT_MAX) { + error = EINVAL; + goto out; } - /* - * 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. + * Disallow functionality not currently supported + * Note: Will need to treat arrays of addresses and controls */ - 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) { - 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), - imin(uio_resid(uio), m->m_len), uio); - m = m_free(m); - } while (uio_resid(uio) && error == 0 && m); - socket_lock(so, 0); -bad: - if (m) - 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. - */ - error = 0; - goto nooob; - } else if (error == 0 && flagsp) { - *flagsp |= MSG_OOB; - } - } + if (addr != NULL) { + printf("%s addr not supported\n", __func__); + error = EOPNOTSUPP; + goto out; + } + if (control != NULL) { + printf("%s control not supported\n", __func__); + error = EOPNOTSUPP; + goto out; + } + + socket_lock(so, 1); + so_update_last_owner_locked(so, p); + so_update_policy(so); + +#if NECP + so_update_necp_policy(so, NULL, addr); +#endif /* NECP */ + + dontroute = (flags & MSG_DONTROUTE) && + (so->so_options & SO_DONTROUTE) == 0 && + (so->so_proto->pr_flags & PR_ATOMIC); + OSIncrementAtomicLong(&p->p_stats->p_ru.ru_msgsnd); + + if (control != NULL) + clen = control->m_len; + + error = sosendcheck(so, addr, resid, clen, atomic, flags, + &sblocked, control); + if (error) + goto release; + + do { + int i; + + if (uioarray == NULL) { + /* + * Data is prepackaged in "top". + */ + resid = 0; + } else { + int num_needed = 0; + int chainlength; + size_t maxpktlen = 0; + + if (sosendminchain > 0) + chainlength = 0; + else + chainlength = sosendmaxchain; + + socket_unlock(so, 0); + + /* + * Find a set of uio that fit in a reasonable number + * of mbuf packets + */ + for (i = uiofirst; i < uiocnt; i++) { + struct uio *auio = uioarray[i]; + + len = uio_resid(auio); + + /* Do nothing for empty messages */ + if (len == 0) + continue; + + num_needed += 1; + uiolast += 1; + + if (len > maxpktlen) + maxpktlen = len; + + chainlength += len; + if (chainlength > sosendmaxchain) + break; + } + /* + * Nothing left to send + */ + if (num_needed == 0) { + socket_lock(so, 0); + break; + } + /* + * Allocate the mbuf packets at once + */ + freelist = m_allocpacket_internal( + (unsigned int *)&num_needed, + maxpktlen, NULL, M_WAIT, 1, 0); + + if (freelist == NULL) { + socket_lock(so, 0); + error = ENOMEM; + goto release; + } + /* + * Copy each uio of the set into its own mbuf packet + */ + for (i = uiofirst, m = freelist; + i < uiolast && m != NULL; + i++) { + int bytes_to_copy; + struct mbuf *n; + struct uio *auio = uioarray[i]; + + bytes_to_copy = uio_resid(auio); + + /* Do nothing for empty messages */ + if (bytes_to_copy == 0) + continue; + + for (n = m; n != NULL; n = n->m_next) { + mlen = mbuf_maxlen(n); + + len = imin(mlen, bytes_to_copy); + + /* + * Note: uiomove() decrements the iovec + * length + */ + error = uiomove(mtod(n, caddr_t), + len, auio); + if (error != 0) + break; + n->m_len = len; + m->m_pkthdr.len += len; + + VERIFY(m->m_pkthdr.len <= maxpktlen); + + bytes_to_copy -= len; + resid -= len; + } + if (m->m_pkthdr.len == 0) { + printf("%s so %llx pkt %llx len null\n", + __func__, + (uint64_t)VM_KERNEL_ADDRPERM(so), + (uint64_t)VM_KERNEL_ADDRPERM(m)); + } + if (error != 0) + break; + m = m->m_nextpkt; + } + + socket_lock(so, 0); + + if (error) + goto release; + top = freelist; + freelist = NULL; + } + + if (dontroute) + so->so_options |= SO_DONTROUTE; + + if ((flags & MSG_SKIPCFIL) == 0) { + struct mbuf **prevnextp = NULL; + + for (i = uiofirst, m = top; + i < uiolast && m != NULL; + i++) { + struct mbuf *nextpkt = m->m_nextpkt; + + /* + * Socket filter processing + */ + error = sflt_data_out(so, addr, &m, + &control, 0); + if (error != 0 && error != EJUSTRETURN) + goto release; + +#if CONTENT_FILTER + if (error == 0) { + /* + * Content filter processing + */ + error = cfil_sock_data_out(so, addr, m, + control, 0); + if (error != 0 && error != EJUSTRETURN) + goto release; + } +#endif /* CONTENT_FILTER */ + /* + * Remove packet from the list when + * swallowed by a filter + */ + if (error == EJUSTRETURN) { + error = 0; + if (prevnextp != NULL) + *prevnextp = nextpkt; + else + top = nextpkt; + } + + m = nextpkt; + if (m != NULL) + prevnextp = &m->m_nextpkt; + } + } + if (top != NULL) + error = (*so->so_proto->pr_usrreqs->pru_send_list) + (so, 0, top, addr, control, p); + + if (dontroute) + so->so_options &= ~SO_DONTROUTE; + + clen = 0; + top = NULL; + uiofirst = uiolast; + } while (resid > 0 && error == 0); +release: + if (sblocked) + sbunlock(&so->so_snd, FALSE); /* will unlock socket */ + else + socket_unlock(so, 1); +out: + if (top != NULL) + m_freem(top); + if (control != NULL) + m_freem(control); + if (freelist != NULL) + m_freem_list(freelist); + + KERNEL_DEBUG(DBG_FNC_SOSEND_LIST | DBG_FUNC_END, so, resid, + so->so_snd.sb_cc, 0, error); + + return (error); +} + +/* + * Implement receive operations on a socket. + * We depend on the way that records are added to the sockbuf + * by sbappend*. In particular, each record (mbufs linked through m_next) + * must begin with an address if the protocol so specifies, + * followed by an optional mbuf or mbufs containing ancillary data, + * and then zero or more mbufs of data. + * In order to avoid blocking network interrupts for the entire time here, + * we splx() while doing the actual copy to user space. + * Although the sockbuf is locked, new data may still be appended, + * and thus we must maintain consistency of the sockbuf during that time. + * + * 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(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; + 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); + + /* + * Sanity check on the length passed by caller as we are making 'int' + * comparisons + */ + if (orig_resid < 0 || orig_resid > INT_MAX) + return (EINVAL); + + socket_lock(so, 1); + so_update_last_owner_locked(so, p); + so_update_policy(so); + +#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 != NULL) + *psa = NULL; + if (controlp != NULL) + *controlp = NULL; + if (flagsp != NULL) + flags = *flagsp &~ MSG_EOR; + else + flags = 0; + + /* + * 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) { + 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), + imin(uio_resid(uio), m->m_len), uio); + m = m_free(m); + } while (uio_resid(uio) && error == 0 && m != NULL); + socket_lock(so, 0); +bad: + if (m != NULL) + m_freem(m); + + 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. + */ + error = 0; + goto nooob; + } else if (error == 0 && flagsp != NULL) { + *flagsp |= MSG_OOB; + } + } socket_unlock(so, 1); KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error, 0, 0, 0, 0); -#endif + return (error); } nooob: - if (mp) - *mp = (struct mbuf *)0; - if (so->so_state & SS_ISCONFIRMING && uio_resid(uio)) - (*pr->pr_usrreqs->pru_rcvd)(so, 0); + 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); + printf("soreceive: sblock so=0x%llx ref=%d on socket\n", + (uint64_t)VM_KERNEL_ADDRPERM(so), so->so_usecount); #endif /* * See if the socket has been closed (SS_NOFDREF|SS_CANTRCVMORE) @@ -2043,9 +2634,13 @@ restart: * 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)) { + (SS_NOFDREF | SS_CANTRCVMORE) && !(so->so_flags & SOF_MP_SUBFLOW)) { socket_unlock(so, 1); return (0); } @@ -2070,11 +2665,11 @@ 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 && + 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 == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { + 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 @@ -2082,12 +2677,10 @@ restart: * end up with false positives during select() or poll() * which could put the application in a bad state. */ - if (m == NULL && so->so_rcv.sb_cc != 0) - panic("soreceive corrupted so_rcv: m %p cc %u", - m, so->so_rcv.sb_cc); + SB_MB_CHECK(&so->so_rcv); if (so->so_error) { - if (m) + if (m != NULL) goto dontblock; error = so->so_error; if ((flags & MSG_PEEK) == 0) @@ -2095,12 +2688,23 @@ restart: goto release; } if (so->so_state & SS_CANTRCVMORE) { - if (m) +#if CONTENT_FILTER + /* + * Deal with half closed connections + */ + if ((so->so_state & SS_ISDISCONNECTED) == 0 && + cfil_sock_data_pending(&so->so_rcv) != 0) + CFIL_LOG(LOG_INFO, + "so %llx ignore SS_CANTRCVMORE", + (uint64_t)VM_KERNEL_ADDRPERM(so)); + else +#endif /* CONTENT_FILTER */ + if (m != NULL) goto dontblock; else goto release; } - for (; m; m = m->m_next) + 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; @@ -2119,7 +2723,7 @@ restart: } SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); - sbunlock(&so->so_rcv, 1); + sbunlock(&so->so_rcv, TRUE); /* keep socket locked */ #if EVEN_MORE_LOCKING_DEBUG if (socket_debug) printf("Waiting for socket data\n"); @@ -2130,9 +2734,11 @@ restart: if (socket_debug) printf("SORECEIVE - sbwait returned %d\n", error); #endif - if (so->so_usecount < 1) - panic("soreceive: after 2nd sblock so=%p ref=%d on " - "socket\n", so, so->so_usecount); + 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) { socket_unlock(so, 1); KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error, @@ -2175,6 +2781,7 @@ dontblock: 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) { /* @@ -2192,10 +2799,18 @@ dontblock: * Process the next record or wait for one. */ socket_lock(so, 0); - sbunlock(&so->so_rcv, 1); + 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 @@ -2216,10 +2831,10 @@ dontblock: } #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; } @@ -2228,9 +2843,11 @@ dontblock: m = m->m_next; } else { sbfree(&so->so_rcv, m); - if (m->m_next == 0 && so->so_rcv.sb_cc != 0) - panic("soreceive: about to create invalid " - "socketbuf"); + 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) { @@ -2252,6 +2869,7 @@ dontblock: 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 @@ -2264,7 +2882,24 @@ dontblock: 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; @@ -2332,11 +2967,38 @@ dontblock: } cm = cmn; } - orig_resid = 0; - if (sb_rcv->sb_mb != NULL) + /* + * 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 = NULL; + 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 != NULL) { @@ -2349,10 +3011,13 @@ dontblock: * 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("soreceive: post-control !sync so=%p " - "m=%p nextrecord=%p\n", so, m, nextrecord); - + 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; } @@ -2361,7 +3026,6 @@ dontblock: flags |= MSG_OOB; } else { if (!(flags & MSG_PEEK)) { - so->so_rcv.sb_mb = nextrecord; SB_EMPTY_FIXUP(&so->so_rcv); } } @@ -2378,7 +3042,8 @@ dontblock: need_event = 0; - while (m && (uio_resid(uio) - 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; @@ -2408,7 +3073,7 @@ 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) { @@ -2469,19 +3134,41 @@ dontblock: sbfree(&so->so_rcv, m); m->m_nextpkt = NULL; - if (mp) { - *mp = m; - mp = &m->m_next; - so->so_rcv.sb_mb = m = m->m_next; - *mp = (struct mbuf *)0; - } else { + /* + * 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 != NULL) { + *mp = m; + mp = &m->m_next; + so->so_rcv.sb_mb = m = m->m_next; + *mp = NULL; + } else { if (free_list == NULL) 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 != NULL) { m->m_nextpkt = nextrecord; @@ -2498,8 +3185,26 @@ dontblock: if (flags & MSG_PEEK) { moff += len; } else { - if (mp) - *mp = m_copym(m, 0, len, M_WAIT); + 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; @@ -2534,10 +3239,14 @@ dontblock: * count but without error. Keep sockbuf locked * against other readers. */ - while (flags & (MSG_WAITALL|MSG_WAITSTREAM) && m == 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) + if (so->so_error || ((so->so_state & SS_CANTRCVMORE) +#if CONTENT_FILTER + && cfil_sock_data_pending(&so->so_rcv) == 0 +#endif /* CONTENT_FILTER */ + )) goto release; /* @@ -2588,29 +3297,28 @@ dontblock: 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("soreceive: after big while so=%p ref=%d on socket\n", - so, so->so_usecount); + 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 (m != NULL && pr->pr_flags & PR_ATOMIC) { if (so->so_options & SO_DONTTRUNC) { flags |= MSG_RCVMORE; } else { -#endif flags |= MSG_TRUNC; if ((flags & MSG_PEEK) == 0) (void) sbdroprecord(&so->so_rcv); -#ifdef __APPLE__ } -#endif } /* @@ -2625,7 +3333,7 @@ dontblock: flags |= MSG_HAVEMORE; if ((flags & MSG_PEEK) == 0) { - if (m == 0) { + if (m == NULL) { so->so_rcv.sb_mb = nextrecord; /* * First part is an inline SB_EMPTY_FIXUP(). Second @@ -2638,49 +3346,50 @@ dontblock: } 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); } -#ifdef __APPLE__ + if (delayed_copy_len) { error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len); - if (error) 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_resid(uio) && orig_resid && (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { - sbunlock(&so->so_rcv, 1); + sbunlock(&so->so_rcv, TRUE); /* keep socket locked */ goto restart; } - if (flagsp) + if (flagsp != NULL) *flagsp |= flags; release: #ifdef MORE_LOCKING_DEBUG - if (so->so_usecount <= 1) - panic("soreceive: release so=%p ref=%d on socket\n", + if (so->so_usecount <= 1) { + panic("%s: release so=%p ref=%d on socket\n", __func__, so, so->so_usecount); + /* NOTREACHED */ + } #endif - if (delayed_copy_len) { + if (delayed_copy_len) error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len); - } - if (free_list) { - m_freem_list((struct mbuf *)free_list); - } - sbunlock(&so->so_rcv, 0); /* will unlock socket */ - // LP64todo - fix this! + 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_resid(uio), so->so_rcv.sb_cc, 0, error); @@ -2693,7 +3402,7 @@ release: */ static int sodelayed_copy(struct socket *so, struct uio *uio, struct mbuf **free_list, - int *resid) + user_ssize_t *resid) { int error = 0; struct mbuf *m; @@ -2702,15 +3411,13 @@ sodelayed_copy(struct socket *so, struct uio *uio, struct mbuf **free_list, socket_unlock(so, 0); - while (m && error == 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; socket_lock(so, 0); @@ -2718,131 +3425,704 @@ sodelayed_copy(struct socket *so, struct uio *uio, struct mbuf **free_list, 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(struct socket *so, int how) +soreceive_list(struct socket *so, struct sockaddr **psa, struct uio **uioarray, + u_int uiocnt, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { - int error; + struct mbuf *m, **mp; + struct mbuf *nextrecord; + struct mbuf *ml = NULL, *free_list = NULL; + int flags, error, offset; + user_ssize_t len; + struct protosw *pr = so->so_proto; + user_ssize_t orig_resid, resid; + struct proc *p = current_proc(); + struct uio *auio = NULL; + int i = 0; + int sblocked = 0; - 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); - } - socket_unlock(so, 1); - break; - default: + KERNEL_DEBUG(DBG_FNC_SORECEIVE_LIST | DBG_FUNC_START, + so, uiocnt, + so->so_rcv.sb_cc, so->so_rcv.sb_lowat, so->so_rcv.sb_hiwat); + + mp = mp0; + if (psa != NULL) + *psa = NULL; + if (controlp != NULL) + *controlp = NULL; + if (flagsp != NULL) + flags = *flagsp &~ MSG_EOR; + else + flags = 0; + /* + * Disallow functionality not currently supported + */ + if (mp0 != NULL) { + printf("%s mp0 not supported\n", __func__); + error = EOPNOTSUPP; + goto out; + } + if (psa != NULL) { + printf("%s sockaddr not supported\n", __func__); + error = EOPNOTSUPP; + goto out; + } + if (controlp != NULL) { + printf("%s control not supported\n", __func__); + error = EOPNOTSUPP; + goto out; + } + + /* + * Sanity checks: + * - Only supports don't wait flags + * - Only support datagram sockets (could be extended to raw) + * - Must be atomic + * - Protocol must support packet chains + * - The uio array is NULL (should we panic?) + */ + if (flags & ~(MSG_DONTWAIT | MSG_NBIO)) { + printf("%s flags not supported\n", __func__); + error = EOPNOTSUPP; + goto out; + } + if (so->so_type != SOCK_DGRAM) { error = EINVAL; - break; + goto out; + } + if (sosendallatonce(so) == 0) { + error = EINVAL; + goto out; + } + if (so->so_proto->pr_usrreqs->pru_send_list == NULL) { + error = EPROTONOSUPPORT; + goto out; + } + if (uioarray == NULL) { + printf("%s uioarray is NULL\n", __func__); + error = EINVAL; + goto out; + } + if (uiocnt == 0) { + printf("%s uiocnt is 0\n", __func__); + error = EINVAL; + goto out; + } + /* + * Sanity check on the length passed by caller as we are making 'int' + * comparisons + */ + resid = orig_resid = uio_array_resid(uioarray, uiocnt); + if (orig_resid < 0 || orig_resid > INT_MAX) { + error = EINVAL; + goto out; } - return (error); -} + socket_lock(so, 1); + so_update_last_owner_locked(so, p); + so_update_policy(so); -int -soshutdownlock(struct socket *so, int how) -{ - struct protosw *pr = so->so_proto; - int error = 0; +#if NECP + so_update_necp_policy(so, NULL, NULL); +#endif /* NECP */ + + /* + * 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; - sflt_notify(so, sock_evt_shutdown, &how); + 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__); + goto release; + } + if (mp != NULL) + *mp = NULL; +restart: + /* + * 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. + */ + if ((so->so_state & (SS_NOFDREF | SS_CANTRCVMORE)) == + (SS_NOFDREF | SS_CANTRCVMORE)) { + error = 0; + goto release; + } - if (how != SHUT_WR) { - if ((so->so_state & SS_CANTRCVMORE) != 0) { - /* read already shut down */ - error = ENOTCONN; - goto done; + error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); + if (error) { + goto release; + } + sblocked = 1; + + /* + * Skip empty uio + */ + auio = uioarray[i]; + while (uio_resid(auio) == 0) { + i++; + if (i >= uiocnt) { + error = 0; + goto release; } - sorflush(so); - postevent(so, 0, EV_RCLOSED); } - if (how != SHUT_RD) { - if ((so->so_state & SS_CANTSENDMORE) != 0) { - /* write already shut down */ + + m = so->so_rcv.sb_mb; + /* + * Block awaiting more datagram if needed + */ + if (m == NULL) { + /* + * 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); + + if (so->so_error) { + error = so->so_error; + goto release; + } + if (so->so_state & SS_CANTRCVMORE) { + goto release; + } + if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && + (so->so_proto->pr_flags & PR_CONNREQUIRED)) { error = ENOTCONN; - goto done; + goto release; } - 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); -} + if ((so->so_state & SS_NBIO) || + (flags & (MSG_DONTWAIT|MSG_NBIO))) { + error = EWOULDBLOCK; + goto release; + } + /* + * Do not block if we got some data + * Note: We could use MSG_WAITALL to wait + */ + resid = uio_array_resid(uioarray, uiocnt); + if (resid != orig_resid) { + error = 0; + goto release; + } + + SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); -void -sorflush(struct socket *so) -{ - register struct sockbuf *sb = &so->so_rcv; - register struct protosw *pr = so->so_proto; - struct sockbuf asb; + sbunlock(&so->so_rcv, TRUE); /* keep socket locked */ + sblocked = 0; -#ifdef MORE_LOCKING_DEBUG - lck_mtx_t *mutex_held; + error = sbwait(&so->so_rcv); + if (error) { + goto release; + } + goto restart; + } - 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 (m->m_pkthdr.len == 0) { + printf("%s so %llx pkt %llx len is null\n", + __func__, + (uint64_t)VM_KERNEL_ADDRPERM(so), + (uint64_t)VM_KERNEL_ADDRPERM(m)); + goto restart; + } + OSIncrementAtomicLong(&p->p_stats->p_ru.ru_msgrcv); + SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); - sflt_notify(so, sock_evt_flush_read, NULL); + /* + * Consume the current uio index as we have a datagram + */ + i += 1; + nextrecord = m->m_nextpkt; - sb->sb_flags |= SB_NOINTR; - (void) sblock(sb, M_WAIT); - socantrcvmore(so); - sbunlock(sb, 1); -#ifdef __APPLE__ - selthreadclear(&sb->sb_sel); -#endif - asb = *sb; - bzero((caddr_t)sb, sizeof (*sb)); - sb->sb_so = so; /* reestablish link to socket */ - if (asb.sb_flags & SB_KNOTE) { - sb->sb_sel.si_note = asb.sb_sel.si_note; - sb->sb_flags = SB_KNOTE; - } - if (asb.sb_flags & SB_DROP) - sb->sb_flags |= SB_DROP; - if (asb.sb_flags & SB_UNIX) - sb->sb_flags |= SB_UNIX; - if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose) { - (*pr->pr_domain->dom_dispose)(asb.sb_mb); +#if SO_RECEIVE_LIST_SOCKADDR_NOT_YET + if ((pr->pr_flags & PR_ADDR) && m->m_type == MT_SONAME) { + /* + * to be adapted from soreceive() + */ } - sbrelease(&asb); -} +#endif /* SO_RECEIVE_LIST_SOCKADDR_NOT_YET */ -/* - * Perhaps this routine, and sooptcopyout(), below, ought to come in - * an additional variant to handle the case where the option value needs - * 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. +#if SO_RECEIVE_LIST_CONTROL_NOT_YET + /* + * 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) { + /* + * to be adapted from soreceive() + */ + } +#endif /* SO_RECEIVE_LIST_CONTROL_NOT_YET */ + + offset = 0; + + /* + * Loop to copy out the mbufs of the current record + */ + while (m != NULL && uio_resid(auio) > 0 && error == 0) { + len = uio_resid(auio); + + if (m->m_len == 0) + printf("%s: so %llx m %llx m_len is 0\n", + __func__, + (uint64_t)VM_KERNEL_ADDRPERM(so), + (uint64_t)VM_KERNEL_ADDRPERM(m)); + + /* + * Clip to the residual length + */ + if (len > m->m_len) + len = m->m_len; + /* + * If mp is set, just pass back the mbufs. + * Otherwise copy them out via the uio, then free. + * Sockbuf must be consistent here (points to current mbuf, + * it points to next record) when we drop priority; + * we must note any additions to the sockbuf when we + * block interrupts again. + */ + if (mp != NULL) { + uio_setresid(auio, (uio_resid(auio) - len)); + } else { + SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); + + socket_unlock(so, 0); + error = uiomove(mtod(m, caddr_t), (int)len, auio); + socket_lock(so, 0); + + if (error) + goto release; + } + if (len == m->m_len) { + /* + * m was entirely copied + */ + nextrecord = m->m_nextpkt; + sbfree(&so->so_rcv, m); + m->m_nextpkt = NULL; + + /* + * Move to m_next + */ + if (mp != NULL) { + *mp = m; + mp = &m->m_next; + so->so_rcv.sb_mb = m = m->m_next; + *mp = NULL; + } else { + if (free_list == NULL) + free_list = m; + else + ml->m_next = m; + ml = m; + so->so_rcv.sb_mb = m = m->m_next; + ml->m_next = NULL; + ml->m_nextpkt = NULL; + } + 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 { + /* + * Stop the loop on partial copy + */ + 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(auio, + (uio_resid(auio) + len)); + error = ENOMEM; + break; + } + } + break; + } + } +#ifdef MORE_LOCKING_DEBUG + if (so->so_usecount <= 1) { + panic("%s: after big while so=%llx ref=%d on socket\n", + __func__, + (uint64_t)VM_KERNEL_ADDRPERM(so), so->so_usecount); + /* NOTREACHED */ + } +#endif + /* + * Tell the caller we made a partial copy + */ + if (m != NULL) { + if (so->so_options & SO_DONTTRUNC) { + m->m_data += len; + m->m_len -= len; + so->so_rcv.sb_cc -= len; + flags |= MSG_RCVMORE; + } else { + (void) sbdroprecord(&so->so_rcv); + nextrecord = so->so_rcv.sb_mb; + m = NULL; + flags |= MSG_TRUNC; + } + } + + 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"); + + /* + * We can continue to the next packet as long as: + * - We haven't exhausted the uio array + * - There was no error + * - A packet was not truncated + * - We can still receive more data + */ + if (i < uiocnt && error == 0 && + (flags & (MSG_RCVMORE | MSG_TRUNC)) == 0 + && (so->so_state & SS_CANTRCVMORE) == 0) { + sbunlock(&so->so_rcv, TRUE); /* keep socket locked */ + sblocked = 0; + + goto restart; + } + +release: + /* + * pru_rcvd may cause more data to be received if the socket lock + * is dropped so we set MSG_HAVEMORE now based on what we know. + * That way the caller won't be surprised if it receives less data than requested. + */ + if ((so->so_options & SO_WANTMORE) && so->so_rcv.sb_cc > 0) + flags |= MSG_HAVEMORE; + + if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) + (*pr->pr_usrreqs->pru_rcvd)(so, flags); + + if (flagsp != NULL) + *flagsp |= flags; + if (sblocked) + sbunlock(&so->so_rcv, FALSE); /* will unlock socket */ + else + socket_unlock(so, 1); +out: + /* + * Amortize the cost + */ + if (free_list != NULL) + m_freem_list(free_list); + + KERNEL_DEBUG(DBG_FNC_SORECEIVE_LIST | DBG_FUNC_END, error, + 0, 0, 0, 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(struct socket *so, int how) +{ + int error; + + KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_START, how, 0, 0, 0, 0); + + 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); + } + socket_unlock(so, 1); + break; + default: + error = EINVAL; + break; + } + + KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_END, how, error, 0, 0, 0); + + return (error); +} + +int +soshutdownlock_final(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) { + 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, how, 1, 0, 0, 0); + return (error); +} + +int +soshutdownlock(struct socket *so, int how) +{ + int error = 0; + +#if CONTENT_FILTER + /* + * A content filter may delay the actual shutdown until it + * has processed the pending data + */ + if (so->so_flags & SOF_CONTENT_FILTER) { + error = cfil_sock_shutdown(so, &how); + if (error == EJUSTRETURN) { + error = 0; + goto done; + } else if (error != 0) { + goto done; + } + } +#endif /* CONTENT_FILTER */ + + error = soshutdownlock_final(so, how); + +done: + return (error); +} + +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(struct socket *so) +{ + struct sockbuf *sb = &so->so_rcv; + struct protosw *pr = so->so_proto; + struct sockbuf asb; +#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 */ + + sflt_notify(so, sock_evt_flush_read, NULL); + + socantrcvmore(so); + + /* + * 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); + + if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose) + (*pr->pr_domain->dom_dispose)(asb.sb_mb); + + sbrelease(&asb); +} + +/* + * Perhaps this routine, and sooptcopyout(), below, ought to come in + * an additional variant to handle the case where the option value needs + * 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 @@ -2879,52 +4159,55 @@ sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) * the top 32-bits assigning tv64.tv_sec to tv_p->tv_sec. */ static int -sooptcopyin_timeval(struct sockopt *sopt, struct timeval * tv_p) +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)) { + if (sopt->sopt_valsize < sizeof (tv64)) return (EINVAL); - } - sopt->sopt_valsize = sizeof(tv64); + + sopt->sopt_valsize = sizeof (tv64); if (sopt->sopt_p != kernproc) { - error = copyin(sopt->sopt_val, &tv64, sizeof(tv64)); + 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)); + sizeof (tv64)); } - if (tv64.tv_sec < 0 || tv64.tv_sec > LONG_MAX - || tv64.tv_usec < 0 || tv64.tv_usec >= 1000000) { + 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)) { + if (sopt->sopt_valsize < sizeof (tv32)) return (EINVAL); - } - sopt->sopt_valsize = sizeof(tv32); + + sopt->sopt_valsize = sizeof (tv32); if (sopt->sopt_p != kernproc) { - error = copyin(sopt->sopt_val, &tv32, sizeof(tv32)); + 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)); + 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) { +#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; @@ -2953,74 +4236,66 @@ sooptcopyin_timeval(struct sockopt *sopt, struct timeval * tv_p) * their filter to return. */ int -sosetopt(struct socket *so, struct sockopt *sopt) +sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) { int error, optval; struct linger l; struct timeval tv; - struct socket_filter_entry *filter; - int filtered = 0; #if CONFIG_MACF_SOCKET struct mac extmac; #endif /* MAC_SOCKET */ - socket_lock(so, 1); - if ((so->so_state & (SS_CANTRCVMORE | SS_CANTSENDMORE)) - == (SS_CANTRCVMORE | SS_CANTSENDMORE) && + 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 bad; - } - - if (sopt->sopt_dir != SOPT_SET) { - sopt->sopt_dir = SOPT_SET; - } - - error = 0; - for (filter = so->so_filt; filter && (error == 0); - filter = filter->sfe_next_onsocket) { - if (filter->sfe_filter->sf_filter.sf_setoption) { - if (filtered == 0) { - filtered = 1; - sflt_use(so); - socket_unlock(so, 0); - } - error = filter->sfe_filter->sf_filter. - sf_setoption(filter->sfe_cookie, so, sopt); - } + goto out; } - if (filtered != 0) { - socket_lock(so, 0); - sflt_unuse(so); - - if (error) { - if (error == EJUSTRETURN) - error = 0; - goto bad; - } + 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) { + if (so->so_proto != NULL && + so->so_proto->pr_ctloutput != NULL) { error = (*so->so_proto->pr_ctloutput)(so, sopt); - socket_unlock(so, 1); - return (error); + 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: case SO_LINGER_SEC: error = sooptcopyin(sopt, &l, sizeof (l), sizeof (l)); - if (error) - goto bad; + if (error != 0) + goto out; so->so_linger = (sopt->sopt_name == SO_LINGER) ? l.l_linger : l.l_linger * hz; - if (l.l_onoff) + if (l.l_onoff != 0) so->so_options |= SO_LINGER; else so->so_options &= ~SO_LINGER; @@ -3035,15 +4310,15 @@ sosetopt(struct socket *so, struct sockopt *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 + case SO_NOWAKEFROMSLEEP: error = sooptcopyin(sopt, &optval, sizeof (optval), sizeof (optval)); - if (error) - goto bad; + if (error != 0) + goto out; if (optval) so->so_options |= sopt->sopt_name; else @@ -3056,8 +4331,8 @@ sosetopt(struct socket *so, struct sockopt *sopt) case SO_RCVLOWAT: error = sooptcopyin(sopt, &optval, sizeof (optval), sizeof (optval)); - if (error) - goto bad; + if (error != 0) + goto out; /* * Values < 1 make no sense for any of these @@ -3065,46 +4340,68 @@ sosetopt(struct socket *so, struct sockopt *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_int32_t) 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; } - if (sopt->sopt_name == SO_SNDBUF) - so->so_snd.sb_flags |= SB_USRSIZE; - else - so->so_rcv.sb_flags |= SB_USRSIZE; + 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. */ - case SO_SNDLOWAT: + case SO_SNDLOWAT: { + int space = sbspace(&so->so_snd); + u_int32_t hiwat = so->so_snd.sb_hiwat; + + if (so->so_snd.sb_flags & SB_UNIX) { + struct unpcb *unp = + (struct unpcb *)(so->so_pcb); + if (unp != NULL && unp->unp_conn != NULL) { + hiwat += unp->unp_conn->unp_cc; + } + } + so->so_snd.sb_lowat = - (optval > so->so_snd.sb_hiwat) ? - so->so_snd.sb_hiwat : optval; + (optval > hiwat) ? + hiwat : optval; + + if (space >= so->so_snd.sb_lowat) { + sowwakeup(so); + } break; - case SO_RCVLOWAT: + } + case SO_RCVLOWAT: { + int64_t data_len; so->so_rcv.sb_lowat = (optval > so->so_rcv.sb_hiwat) ? so->so_rcv.sb_hiwat : optval; + data_len = so->so_rcv.sb_cc + - so->so_rcv.sb_ctl; + if (data_len >= so->so_rcv.sb_lowat) + sorwakeup(so); break; } + } break; case SO_SNDTIMEO: case SO_RCVTIMEO: error = sooptcopyin_timeval(sopt, &tv); - if (error) - goto bad; + if (error != 0) + goto out; switch (sopt->sopt_name) { case SO_SNDTIMEO: @@ -3116,152 +4413,368 @@ sosetopt(struct socket *so, struct sockopt *sopt) } break; - case SO_NKE: - { + case SO_NKE: { struct so_nke nke; error = sooptcopyin(sopt, &nke, sizeof (nke), sizeof (nke)); - if (error) - goto bad; + if (error != 0) + goto out; - error = sflt_attach_private(so, NULL, - nke.nke_handle, 1); + 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) + 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) + 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) - goto bad; - if (optval) + if (error != 0) + goto out; + if (optval != 0) so->so_flags |= SOF_REUSESHAREUID; else so->so_flags &= ~SOF_REUSESHAREUID; break; -#ifdef __APPLE_API_PRIVATE + case SO_NOTIFYCONFLICT: if (kauth_cred_issuser(kauth_cred_get()) == 0) { error = EPERM; - goto bad; + goto out; } error = sooptcopyin(sopt, &optval, sizeof (optval), sizeof (optval)); - if (error) - goto bad; - if (optval) + if (error != 0) + goto out; + if (optval != 0) so->so_flags |= SOF_NOTIFYCONFLICT; else so->so_flags &= ~SOF_NOTIFYCONFLICT; break; -#endif + case SO_RESTRICTIONS: - if (kauth_cred_issuser(kauth_cred_get()) == 0) { + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + + error = so_set_restrictions(so, optval); + break; + + case SO_AWDL_UNRESTRICTED: + if (SOCK_DOM(so) != PF_INET && + SOCK_DOM(so) != PF_INET6) { + error = EOPNOTSUPP; + goto out; + } + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) + goto out; + if (optval != 0) { + kauth_cred_t cred = NULL; + proc_t ep = PROC_NULL; + + if (so->so_flags & SOF_DELEGATED) { + ep = proc_find(so->e_pid); + if (ep) + cred = kauth_cred_proc_ref(ep); + } + error = priv_check_cred( + cred ? cred : so->so_cred, + PRIV_NET_RESTRICTED_AWDL, 0); + if (error == 0) + inp_set_awdl_unrestricted( + sotoinpcb(so)); + if (cred) + kauth_cred_unref(&cred); + if (ep != PROC_NULL) + proc_rele(ep); + } else + inp_clear_awdl_unrestricted(sotoinpcb(so)); + 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 bad; + 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 " : "")); } - error = sooptcopyin(sopt, &optval, sizeof (optval), - sizeof (optval)); - if (error) - goto bad; - so->so_restrictions = (optval & (SO_RESTRICT_DENYIN | - SO_RESTRICT_DENYOUT | SO_RESTRICT_DENYSET)); break; - case SO_LABEL: -#if CONFIG_MACF_SOCKET - if ((error = sooptcopyin(sopt, &extmac, sizeof (extmac), - sizeof (extmac))) != 0) - goto bad; - - error = mac_setsockopt_label(proc_ucred(sopt->sopt_p), - so, &extmac); -#else - error = EOPNOTSUPP; -#endif /* MAC_SOCKET */ + case SO_ISDEFUNCT: + /* This option is not settable */ + error = EINVAL; break; -#ifdef __APPLE_API_PRIVATE - case SO_UPCALLCLOSEWAIT: + case SO_OPPORTUNISTIC: error = sooptcopyin(sopt, &optval, sizeof (optval), sizeof (optval)); - if (error) - goto bad; - if (optval) - so->so_flags |= SOF_UPCALLCLOSEWAIT; - else - so->so_flags &= ~SOF_UPCALLCLOSEWAIT; + if (error == 0) + error = so_set_opportunistic(so, optval); break; -#endif - case SO_RANDOMPORT: + 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) - goto bad; - if (optval) - so->so_flags |= SOF_BINDRANDOMPORT; - else - so->so_flags &= ~SOF_BINDRANDOMPORT; + if (error == 0) + error = so_set_recv_anyif(so, optval); break; - case SO_NP_EXTENSIONS: { - struct so_np_extensions sonpx; + case SO_TRAFFIC_MGT_BACKGROUND: { + /* This option is handled by lower layer(s) */ + error = 0; + break; + } - error = sooptcopyin(sopt, &sonpx, sizeof(sonpx), sizeof(sonpx)); - if (error) - goto bad; - if (sonpx.npx_mask & ~SONPX_MASK_VALID) { - error = EINVAL; - goto bad; - } - /* - * 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; - } +#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; } + +#if NECP + case SO_NECP_ATTRIBUTES: + error = necp_set_socket_attributes(so, sopt); + break; +#endif /* NECP */ + +#if MPTCP + case SO_MPTCP_FASTJOIN: + if (!((so->so_flags & SOF_MP_SUBFLOW) || + ((SOCK_CHECK_DOM(so, PF_MULTIPATH)) && + (SOCK_CHECK_PROTO(so, IPPROTO_TCP))))) { + error = ENOPROTOOPT; + break; + } + + error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval)); + if (error != 0) + goto out; + if (optval == 0) + so->so_flags &= ~SOF_MPTCP_FASTJOIN; + else + so->so_flags |= SOF_MPTCP_FASTJOIN; + break; +#endif /* MPTCP */ 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: - socket_unlock(so, 1); +out: + if (dolock) + socket_unlock(so, 1); return (error); } @@ -3295,7 +4808,7 @@ sooptcopyout(struct sockopt *sopt, void *buf, size_t len) } static int -sooptcopyout_timeval(struct sockopt *sopt, const struct timeval * tv_p) +sooptcopyout_timeval(struct sockopt *sopt, const struct timeval *tv_p) { int error; size_t len; @@ -3306,12 +4819,12 @@ sooptcopyout_timeval(struct sockopt *sopt, const struct timeval * tv_p) error = 0; if (proc_is64bit(sopt->sopt_p)) { - len = sizeof(tv64); + len = sizeof (tv64); tv64.tv_sec = tv_p->tv_sec; tv64.tv_usec = tv_p->tv_usec; val = &tv64; } else { - len = sizeof(tv32); + len = sizeof (tv32); tv32.tv_sec = tv_p->tv_sec; tv32.tv_usec = tv_p->tv_usec; val = &tv32; @@ -3335,63 +4848,52 @@ sooptcopyout_timeval(struct sockopt *sopt, const struct timeval * tv_p) * :??? */ int -sogetopt(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 socket_filter_entry *filter; - int filtered = 0; #if CONFIG_MACF_SOCKET struct mac extmac; #endif /* MAC_SOCKET */ - if (sopt->sopt_dir != SOPT_GET) { + if (sopt->sopt_dir != SOPT_GET) sopt->sopt_dir = SOPT_GET; - } - - socket_lock(so, 1); - error = 0; - for (filter = so->so_filt; filter && (error == 0); - filter = filter->sfe_next_onsocket) { - if (filter->sfe_filter->sf_filter.sf_getoption) { - if (filtered == 0) { - filtered = 1; - sflt_use(so); - socket_unlock(so, 0); - } - error = filter->sfe_filter->sf_filter. - sf_getoption(filter->sfe_cookie, so, sopt); - } - } - if (filtered != 0) { - socket_lock(so, 0); - sflt_unuse(so); + if (dolock) + socket_lock(so, 1); - if (error) { - if (error == EJUSTRETURN) - error = 0; - socket_unlock(so, 1); - return (error); - } + 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) { + if (so->so_proto != NULL && + so->so_proto->pr_ctloutput != NULL) { error = (*so->so_proto->pr_ctloutput)(so, sopt); - socket_unlock(so, 1); - return (error); - } else { - socket_unlock(so, 1); - return (ENOPROTOOPT); + 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: case SO_LINGER_SEC: - l.l_onoff = so->so_options & SO_LINGER; + 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)); @@ -3406,11 +4908,11 @@ sogetopt(struct socket *so, struct sockopt *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 + case SO_NOWAKEFROMSLEEP: optval = so->so_options & sopt->sopt_name; integer: error = sooptcopyout(sopt, &optval, sizeof (optval)); @@ -3420,7 +4922,6 @@ integer: optval = so->so_type; goto integer; -#ifdef __APPLE__ case SO_NREAD: if (so->so_proto->pr_flags & PR_ATOMIC) { int pkt_total; @@ -3428,9 +4929,10 @@ integer: pkt_total = 0; m1 = so->so_rcv.sb_mb; - while (m1) { - if (m1->m_type == MT_DATA || m1->m_type == MT_HEADER || - m1->m_type == MT_OOBDATA) + while (m1 != NULL) { + if (m1->m_type == MT_DATA || + m1->m_type == MT_HEADER || + m1->m_type == MT_OOBDATA) pkt_total += m1->m_len; m1 = m1->m_next; } @@ -3439,20 +4941,50 @@ integer: optval = so->so_rcv.sb_cc - so->so_rcv.sb_ctl; } goto integer; - + + case SO_NUMRCVPKT: + if (so->so_proto->pr_flags & PR_ATOMIC) { + int cnt = 0; + struct mbuf *m1; + + 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) + cnt += 1; + m1 = m1->m_nextpkt; + } + optval = cnt; + goto integer; + } else { + error = EINVAL; + break; + } + case SO_NWRITE: optval = so->so_snd.sb_cc; goto integer; -#endif + case SO_ERROR: optval = so->so_error; so->so_error = 0; goto integer; - case SO_SNDBUF: - optval = so->so_snd.sb_hiwat; - goto integer; + case SO_SNDBUF: { + u_int32_t hiwat = so->so_snd.sb_hiwat; + + if (so->so_snd.sb_flags & SB_UNIX) { + struct unpcb *unp = + (struct unpcb *)(so->so_pcb); + if (unp != NULL && unp->unp_conn != NULL) { + hiwat += unp->unp_conn->unp_cc; + } + } + optval = hiwat; + goto integer; + } case SO_RCVBUF: optval = so->so_rcv.sb_hiwat; goto integer; @@ -3485,16 +5017,25 @@ integer: optval = (so->so_flags & SOF_REUSESHAREUID); goto integer; -#ifdef __APPLE_API_PRIVATE + case SO_NOTIFYCONFLICT: optval = (so->so_flags & SOF_NOTIFYCONFLICT); goto integer; -#endif + case SO_RESTRICTIONS: - optval = so->so_restrictions & (SO_RESTRICT_DENYIN | - SO_RESTRICT_DENYOUT | SO_RESTRICT_DENYSET); + optval = so_get_restrictions(so); goto integer; + case SO_AWDL_UNRESTRICTED: + if (SOCK_DOM(so) == PF_INET || + SOCK_DOM(so) == PF_INET6) { + optval = inp_get_awdl_unrestricted( + sotoinpcb(so)); + goto integer; + } else + error = EOPNOTSUPP; + break; + case SO_LABEL: #if CONFIG_MACF_SOCKET if ((error = sooptcopyin(sopt, &extmac, sizeof (extmac), @@ -3535,22 +5076,117 @@ integer: case SO_NP_EXTENSIONS: { struct so_np_extensions sonpx; - sonpx.npx_flags = (so->so_flags & SOF_NPX_SETOPTSHUT) ? SONPX_SETOPTSHUT : 0; + 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; + 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 */ + +#if NECP + case SO_NECP_ATTRIBUTES: + error = necp_get_socket_attributes(so, sopt); + break; +#endif /* NECP */ + +#if CONTENT_FILTER + case SO_CFIL_SOCK_ID: { + cfil_sock_id_t sock_id; + + sock_id = cfil_sock_id_from_socket(so); + + error = sooptcopyout(sopt, &sock_id, + sizeof(cfil_sock_id_t)); + break; } +#endif /* CONTENT_FILTER */ + +#if MPTCP + case SO_MPTCP_FASTJOIN: + if (!((so->so_flags & SOF_MP_SUBFLOW) || + ((SOCK_CHECK_DOM(so, PF_MULTIPATH)) && + (SOCK_CHECK_PROTO(so, IPPROTO_TCP))))) { + error = ENOPROTOOPT; + break; + } + optval = (so->so_flags & SOF_MPTCP_FASTJOIN); + break; +#endif /* MPTCP */ + default: error = ENOPROTOOPT; break; } - socket_unlock(so, 1); - return (error); } +out: + if (dolock) + socket_unlock(so, 1); + return (error); } -/* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ +/* + * 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. + */ int soopt_getm(struct sockopt *sopt, struct mbuf **mp) { @@ -3558,12 +5194,12 @@ soopt_getm(struct sockopt *sopt, struct mbuf **mp) int sopt_size = sopt->sopt_valsize; int how; - if (sopt_size > MAX_SOOPTGETM_SIZE) + if (sopt_size <= 0 || sopt_size > MCLBYTES) return (EMSGSIZE); how = sopt->sopt_p != kernproc ? M_WAIT : M_DONTWAIT; MGET(m, how, MT_DATA); - if (m == 0) + if (m == NULL) return (ENOBUFS); if (sopt_size > MLEN) { MCLGET(m, how); @@ -3579,9 +5215,9 @@ soopt_getm(struct sockopt *sopt, struct mbuf **mp) *mp = m; m_prev = m; - while (sopt_size) { + while (sopt_size > 0) { MGET(m, how, MT_DATA); - if (m == 0) { + if (m == NULL) { m_freem(*mp); return (ENOBUFS); } @@ -3589,6 +5225,7 @@ soopt_getm(struct sockopt *sopt, struct mbuf **mp) MCLGET(m, how); if ((m->m_flags & M_EXT) == 0) { m_freem(*mp); + m_freem(m); return (ENOBUFS); } m->m_len = min(MCLBYTES, sopt_size); @@ -3602,7 +5239,7 @@ soopt_getm(struct sockopt *sopt, struct mbuf **mp) 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) { @@ -3628,12 +5265,15 @@ soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 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"); + /* 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) { @@ -3673,7 +5313,6 @@ soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) void sohasoutofband(struct socket *so) { - if (so->so_pgid < 0) gsignal(-so->so_pgid, SIGURG); else if (so->so_pgid > 0) @@ -3682,12 +5321,15 @@ sohasoutofband(struct socket *so) } int -sopoll(struct socket *so, int events, __unused kauth_cred_t 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; socket_lock(so, 1); + so_update_last_owner_locked(so, PROC_NULL); + so_update_policy(so); if (events & (POLLIN | POLLRDNORM)) if (soreadable(so)) @@ -3726,16 +5368,22 @@ sopoll(struct socket *so, int events, __unused kauth_cred_t cred, void * wql) } int -soo_kqfilter(__unused struct fileproc *fp, struct knote *kn, - __unused struct proc *p) +soo_kqfilter(struct fileproc *fp, struct knote *kn, vfs_context_t ctx) { +#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 sockbuf *sb; + 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(p), kn, so) != 0) { + if (mac_socket_check_kqfilter(proc_ucred(vfs_context_proc(ctx)), + kn, so) != 0) { socket_unlock(so, 1); return (1); } @@ -3744,19 +5392,37 @@ soo_kqfilter(__unused struct fileproc *fp, struct knote *kn, switch (kn->kn_filter) { case EVFILT_READ: kn->kn_fop = &soread_filtops; - sb = &so->so_rcv; + 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 (KNOTE_ATTACH(&sb->sb_sel.si_note, kn)) - sb->sb_flags |= SB_KNOTE; + 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); } @@ -3785,9 +5451,10 @@ filt_soread(struct knote *kn, long hint) 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. + /* + * 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; @@ -3813,7 +5480,11 @@ filt_soread(struct knote *kn, long hint) kn->kn_data = so->so_oobmark; kn->kn_flags |= EV_OOBAND; } else { - if (so->so_state & SS_CANTRCVMORE) { + if ((so->so_state & SS_CANTRCVMORE) +#if CONTENT_FILTER + && cfil_sock_data_pending(&so->so_rcv) == 0 +#endif /* CONTENT_FILTER */ + ) { kn->kn_flags |= EV_EOF; kn->kn_fflags = so->so_error; if ((hint & SO_FILT_HINT_LOCKED) == 0) @@ -3842,12 +5513,18 @@ filt_soread(struct knote *kn, long hint) return (1); } + 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 >= ((kn->kn_sfflags & NOTE_LOWAT) ? - kn->kn_sdata : so->so_rcv.sb_lowat)); + return ((kn->kn_flags & EV_OOBAND) || kn->kn_data >= lowwat); } static void @@ -3862,11 +5539,24 @@ filt_sowdetach(struct knote *kn) 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_fglob->fg_data; + int ret = 0; if ((hint & SO_FILT_HINT_LOCKED) == 0) socket_lock(so, 1); @@ -3875,42 +5565,194 @@ filt_sowrite(struct knote *kn, long hint) if (so->so_state & SS_CANTSENDMORE) { kn->kn_flags |= EV_EOF; kn->kn_fflags = so->so_error; - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (1); + ret = 1; + goto out; } if (so->so_error) { /* temporary udp error */ - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (1); + ret = 1; + goto out; } if (((so->so_state & SS_ISCONNECTED) == 0) && (so->so_proto->pr_flags & PR_CONNREQUIRED)) { - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (0); + 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) { + if ((SOCK_DOM(so) == PF_INET + || SOCK_DOM(so) == PF_INET6) + && so->so_type == SOCK_STREAM) { + ret = tcp_notsent_lowat_check(so); + } +#if MPTCP + else if ((SOCK_DOM(so) == PF_MULTIPATH) && + (SOCK_PROTO(so) == IPPROTO_TCP)) { + ret = mptcp_notsent_lowat_check(so); + } +#endif + else { + return (1); + } + } 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); +} + +static int +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) +#if CONTENT_FILTER + && cfil_sock_data_pending(&so->so_rcv) == 0 +#endif /* CONTENT_FILTER */ + ) + 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 ((hint & SO_FILT_HINT_LOCKED) == 0) + + 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); - if (kn->kn_sfflags & NOTE_LOWAT) - return (kn->kn_data >= kn->kn_sdata); - return (kn->kn_data >= so->so_snd.sb_lowat); -} -#define SO_LOCK_HISTORY_STR_LEN (2 * SO_LCKDBG_MAX * (2 + sizeof(void *) + 1) + 1) + return (ret); +} -__private_extern__ const char * solockhistory_nr(struct socket *so) +void +get_sockev_state(struct socket *so, u_int32_t *statep) { - size_t n = 0; - int i; - static char lock_history_str[SO_LOCK_HISTORY_STR_LEN]; + 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; +} - for (i = SO_LCKDBG_MAX - 1; i >= 0; i--) { - n += snprintf(lock_history_str + n, SO_LOCK_HISTORY_STR_LEN - n, "%lx:%lx ", - (uintptr_t) so->lock_lr[(so->next_lock_lr + i) % SO_LCKDBG_MAX], - (uintptr_t) so->unlock_lr[(so->next_unlock_lr + i) % SO_LCKDBG_MAX]); +#define SO_LOCK_HISTORY_STR_LEN \ + (2 * SO_LCKDBG_MAX * (2 + (2 * sizeof (void *)) + 1) + 1) + +__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; + return (lock_history_str); } int @@ -3947,8 +5789,10 @@ socket_unlock(struct socket *so, int refcount) lr_saved = __builtin_return_address(0); - if (so->so_proto == NULL) - panic("socket_unlock null so_proto so=%p\n", so); + 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); @@ -3961,16 +5805,17 @@ socket_unlock(struct socket *so, int refcount) so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; if (refcount) { - if (so->so_usecount <= 0) - panic("socket_unlock: bad refcount=%d so=%p (%d, %d, %d) lrh=%s", - so->so_usecount, so, so->so_proto->pr_domain->dom_family, - so->so_type, so->so_proto->pr_protocol, - solockhistory_nr(so)); - + 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) { + if (so->so_usecount == 0) sofreelastref(so, 1); - } } lck_mtx_unlock(mutex_held); } @@ -3982,8 +5827,8 @@ socket_unlock(struct socket *so, int refcount) 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 @@ -4021,15 +5866,653 @@ somultipages(struct socket *so, boolean_t set) so->so_flags &= ~SOF_MULTIPAGES; } +void +soif2kcl(struct socket *so, boolean_t set) +{ + if (set) + so->so_flags1 |= SOF1_IF_2KCL; + else + so->so_flags1 &= ~SOF1_IF_2KCL; +} + int so_isdstlocal(struct socket *so) { struct inpcb *inp = (struct inpcb *)so->so_pcb; - if (so->so_proto->pr_domain->dom_family == AF_INET) { - return inaddr_local(inp->inp_faddr); - } else if (so->so_proto->pr_domain->dom_family == AF_INET6) { - return in6addr_local(&inp->in6p_faddr); - } - return 0; + 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); + + so->so_flags1 |= SOF1_DEFUNCTINPROG; + 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_final(so, SHUT_RD); + (void) soshutdownlock_final(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 noexpensive_old, noexpensive_new; + + /* + * 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); + noexpensive_old = (so->so_restrictions & SO_RESTRICT_DENY_EXPENSIVE); + so->so_restrictions |= (vals & (SO_RESTRICT_DENY_IN | + SO_RESTRICT_DENY_OUT | SO_RESTRICT_DENY_CELLULAR | + SO_RESTRICT_DENY_EXPENSIVE)); + nocell_new = (so->so_restrictions & SO_RESTRICT_DENY_CELLULAR); + noexpensive_new = (so->so_restrictions & SO_RESTRICT_DENY_EXPENSIVE); + + /* we can only set, not clear restrictions */ + if ((nocell_new - nocell_old) == 0 && + (noexpensive_new - noexpensive_old) == 0) + return (0); +#if INET6 + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) { +#else + if (SOCK_DOM(so) == PF_INET) { +#endif /* !INET6 */ + if (nocell_new - nocell_old != 0) { + /* if deny cellular is now set, do what's needed for INPCB */ + inp_set_nocellular(sotoinpcb(so)); + } + if (noexpensive_new - noexpensive_old != 0) { + inp_set_noexpensive(sotoinpcb(so)); + } + } + + return (0); +} + +uint32_t +so_get_restrictions(struct socket *so) +{ + return (so->so_restrictions & (SO_RESTRICT_DENY_IN | + SO_RESTRICT_DENY_OUT | + SO_RESTRICT_DENY_CELLULAR | SO_RESTRICT_DENY_EXPENSIVE)); +} + +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); + } + + /* Update this socket's policy upon success */ + if (error == 0) { + so->so_policy_gencnt *= -1; + so_update_policy(so); +#if NECP + so_update_necp_policy(so, NULL, NULL); +#endif /* NECP */ + } + + 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); + } + + /* Update this socket's policy upon success */ + if (error == 0) { + so->so_policy_gencnt *= -1; + so_update_policy(so); +#if NECP + so_update_necp_policy(so, NULL, NULL); +#endif /* NECP */ + } + + 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); +} + +void +socket_post_kev_msg(uint32_t ev_code, + struct kev_socket_event_data *ev_data, + uint32_t ev_datalen) +{ + struct kev_msg ev_msg; + + 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_SOCKET_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); +} + +void +socket_post_kev_msg_closed(struct socket *so) +{ + struct kev_socket_closed ev; + struct sockaddr *socksa = NULL, *peersa = NULL; + int err; + bzero(&ev, sizeof(ev)); + err = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, &socksa); + if (err == 0) { + err = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, + &peersa); + if (err == 0) { + memcpy(&ev.ev_data.kev_sockname, socksa, + min(socksa->sa_len, + sizeof (ev.ev_data.kev_sockname))); + memcpy(&ev.ev_data.kev_peername, peersa, + min(peersa->sa_len, + sizeof (ev.ev_data.kev_peername))); + socket_post_kev_msg(KEV_SOCKET_CLOSED, + &ev.ev_data, sizeof (ev)); + } + } + if (socksa != NULL) + FREE(socksa, M_SONAME); + if (peersa != NULL) + FREE(peersa, M_SONAME); }