X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/89b3af67bb32e691275bf6fa803d1834b2284115..f427ee49d309d8fc33ebf3042c3a775f2f530ded:/bsd/kern/uipc_socket.c diff --git a/bsd/kern/uipc_socket.c b/bsd/kern/uipc_socket.c index 9ba75e943..607af6d3c 100644 --- a/bsd/kern/uipc_socket.c +++ b/bsd/kern/uipc_socket.c @@ -1,8 +1,8 @@ /* - * Copyright (c) 2000-2005 Apple Computer, Inc. All rights reserved. + * Copyright (c) 1998-2020 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, @@ -22,10 +22,9 @@ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. * Please see the License for the specific language governing rights and * limitations under the License. - * + * * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ -/* Copyright (c) 1998, 1999 Apple Computer, Inc. All Rights Reserved */ /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */ /* * Copyright (c) 1982, 1986, 1988, 1990, 1993 @@ -60,12 +59,18 @@ * SUCH DAMAGE. * * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 - * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.16 2001/06/14 20:46:06 ume Exp $ + */ +/* + * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce + * support for mandatory and extensible security protections. This notice + * is included in support of clause 2.2 (b) of the Apple Public License, + * Version 2.0. */ #include #include #include +#include #include #include #include @@ -82,304 +87,524 @@ #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 +#include +#include + +#include +#include +#include +#include + +#if CONFIG_MACF +#include +#endif /* MAC */ + +#if MULTIPATH +#include +#include +#endif /* MULTIPATH */ + +#define ROUNDUP(a, b) (((a) + ((b) - 1)) & (~((b) - 1))) + +#if DEBUG || DEVELOPMENT +#define DEBUG_KERNEL_ADDRPERM(_v) (_v) +#else +#define DEBUG_KERNEL_ADDRPERM(_v) VM_KERNEL_ADDRPERM(_v) +#endif -int so_cache_hw = 0; -int so_cache_timeouts = 0; -int so_cache_max_freed = 0; -int cached_sock_count = 0; -struct socket *socket_cache_head = 0; -struct socket *socket_cache_tail = 0; -u_long so_cache_time = 0; -int so_cache_init_done = 0; -struct zone *so_cache_zone; -extern int get_inpcb_str_size(); -extern int get_tcp_str_size(); - -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; +/* TODO: this should be in a header file somewhere */ +extern char *proc_name_address(void *p); + +static u_int32_t so_cache_hw; /* High water mark for socache */ +static u_int32_t so_cache_timeouts; /* number of timeouts */ +static u_int32_t so_cache_max_freed; /* max freed per timeout */ +static u_int32_t cached_sock_count = 0; +STAILQ_HEAD(, socket) so_cache_head; +int max_cached_sock_count = MAX_CACHED_SOCKETS; +static u_int32_t so_cache_time; +static int socketinit_done; +static struct zone *so_cache_zone; + +static lck_grp_t *so_cache_mtx_grp; +static lck_attr_t *so_cache_mtx_attr; +static lck_grp_attr_t *so_cache_mtx_grp_attr; +static lck_mtx_t *so_cache_mtx; #include +static int filt_sorattach(struct knote *kn, struct kevent_qos_s *kev); static void filt_sordetach(struct knote *kn); static int filt_soread(struct knote *kn, long hint); +static int filt_sortouch(struct knote *kn, struct kevent_qos_s *kev); +static int filt_sorprocess(struct knote *kn, struct kevent_qos_s *kev); + +static int filt_sowattach(struct knote *kn, struct kevent_qos_s *kev); static void filt_sowdetach(struct knote *kn); static int filt_sowrite(struct knote *kn, long hint); -static int filt_solisten(struct knote *kn, long hint); +static int filt_sowtouch(struct knote *kn, struct kevent_qos_s *kev); +static int filt_sowprocess(struct knote *kn, struct kevent_qos_s *kev); + +static int filt_sockattach(struct knote *kn, struct kevent_qos_s *kev); +static void filt_sockdetach(struct knote *kn); +static int filt_sockev(struct knote *kn, long hint); +static int filt_socktouch(struct knote *kn, struct kevent_qos_s *kev); +static int filt_sockprocess(struct knote *kn, struct kevent_qos_s *kev); + +static int sooptcopyin_timeval(struct sockopt *, struct timeval *); +static int sooptcopyout_timeval(struct sockopt *, const struct timeval *); + +SECURITY_READ_ONLY_EARLY(struct filterops) soread_filtops = { + .f_isfd = 1, + .f_attach = filt_sorattach, + .f_detach = filt_sordetach, + .f_event = filt_soread, + .f_touch = filt_sortouch, + .f_process = filt_sorprocess, +}; + +SECURITY_READ_ONLY_EARLY(struct filterops) sowrite_filtops = { + .f_isfd = 1, + .f_attach = filt_sowattach, + .f_detach = filt_sowdetach, + .f_event = filt_sowrite, + .f_touch = filt_sowtouch, + .f_process = filt_sowprocess, +}; + +SECURITY_READ_ONLY_EARLY(struct filterops) sock_filtops = { + .f_isfd = 1, + .f_attach = filt_sockattach, + .f_detach = filt_sockdetach, + .f_event = filt_sockev, + .f_touch = filt_socktouch, + .f_process = filt_sockprocess, +}; + +SECURITY_READ_ONLY_EARLY(struct filterops) soexcept_filtops = { + .f_isfd = 1, + .f_attach = filt_sorattach, + .f_detach = filt_sordetach, + .f_event = filt_soread, + .f_touch = filt_sortouch, + .f_process = filt_sorprocess, +}; -static struct filterops solisten_filtops = - { 1, NULL, filt_sordetach, filt_solisten }; -static struct filterops soread_filtops = - { 1, NULL, filt_sordetach, filt_soread }; -static struct filterops sowrite_filtops = - { 1, NULL, filt_sowdetach, filt_sowrite }; +SYSCTL_DECL(_kern_ipc); #define EVEN_MORE_LOCKING_DEBUG 0 + int socket_debug = 0; -int socket_zone = M_SOCKET; -so_gen_t so_gencnt; /* generation count for sockets */ +SYSCTL_INT(_kern_ipc, OID_AUTO, socket_debug, + CTLFLAG_RW | CTLFLAG_LOCKED, &socket_debug, 0, ""); + +static unsigned long sodefunct_calls = 0; +SYSCTL_LONG(_kern_ipc, OID_AUTO, sodefunct_calls, CTLFLAG_LOCKED, + &sodefunct_calls, ""); + +ZONE_DECLARE(socket_zone, "socket", sizeof(struct socket), ZC_ZFREE_CLEARMEM); +so_gen_t so_gencnt; /* generation count for sockets */ MALLOC_DEFINE(M_SONAME, "soname", "socket name"); MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); -#define DBG_LAYER_IN_BEG NETDBG_CODE(DBG_NETSOCK, 0) -#define DBG_LAYER_IN_END NETDBG_CODE(DBG_NETSOCK, 2) -#define DBG_LAYER_OUT_BEG NETDBG_CODE(DBG_NETSOCK, 1) -#define DBG_LAYER_OUT_END NETDBG_CODE(DBG_NETSOCK, 3) -#define DBG_FNC_SOSEND NETDBG_CODE(DBG_NETSOCK, (4 << 8) | 1) -#define DBG_FNC_SORECEIVE NETDBG_CODE(DBG_NETSOCK, (8 << 8)) +#define DBG_LAYER_IN_BEG NETDBG_CODE(DBG_NETSOCK, 0) +#define DBG_LAYER_IN_END NETDBG_CODE(DBG_NETSOCK, 2) +#define DBG_LAYER_OUT_BEG NETDBG_CODE(DBG_NETSOCK, 1) +#define DBG_LAYER_OUT_END NETDBG_CODE(DBG_NETSOCK, 3) +#define DBG_FNC_SOSEND NETDBG_CODE(DBG_NETSOCK, (4 << 8) | 1) +#define DBG_FNC_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) - +#define MAX_SOOPTGETM_SIZE (128 * MCLBYTES) -SYSCTL_DECL(_kern_ipc); - -static int somaxconn = SOMAXCONN; -SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, &somaxconn, - 0, ""); +int somaxconn = SOMAXCONN; +SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, + CTLFLAG_RW | CTLFLAG_LOCKED, &somaxconn, 0, ""); /* Should we get a maximum also ??? */ static int sosendmaxchain = 65536; static int sosendminchain = 16384; static int sorecvmincopy = 16384; -SYSCTL_INT(_kern_ipc, OID_AUTO, sosendminchain, CTLFLAG_RW, &sosendminchain, - 0, ""); -SYSCTL_INT(_kern_ipc, OID_AUTO, sorecvmincopy, CTLFLAG_RW, &sorecvmincopy, - 0, ""); +SYSCTL_INT(_kern_ipc, OID_AUTO, sosendminchain, + CTLFLAG_RW | CTLFLAG_LOCKED, &sosendminchain, 0, ""); +SYSCTL_INT(_kern_ipc, OID_AUTO, sorecvmincopy, + CTLFLAG_RW | CTLFLAG_LOCKED, &sorecvmincopy, 0, ""); + +/* + * Set to enable jumbo clusters (if available) for large writes when + * the socket is marked with SOF_MULTIPAGES; see below. + */ +int sosendjcl = 1; +SYSCTL_INT(_kern_ipc, OID_AUTO, sosendjcl, + CTLFLAG_RW | CTLFLAG_LOCKED, &sosendjcl, 0, ""); + +/* + * Set this to ignore SOF_MULTIPAGES and use jumbo clusters for large + * writes on the socket for all protocols on any network interfaces, + * depending upon sosendjcl above. Be extra careful when setting this + * to 1, because sending down packets that cross physical pages down to + * broken drivers (those that falsely assume that the physical pages + * are contiguous) might lead to system panics or silent data corruption. + * When set to 0, the system will respect SOF_MULTIPAGES, which is set + * only for TCP sockets whose outgoing interface is IFNET_MULTIPAGES + * capable. Set this to 1 only for testing/debugging purposes. + */ +int sosendjcl_ignore_capab = 0; +SYSCTL_INT(_kern_ipc, OID_AUTO, sosendjcl_ignore_capab, + CTLFLAG_RW | CTLFLAG_LOCKED, &sosendjcl_ignore_capab, 0, ""); + +/* + * Set this to ignore SOF1_IF_2KCL and use big clusters for large + * writes on the socket for all protocols on any network interfaces. + * Be extra careful when setting this to 1, because sending down packets with + * clusters larger that 2 KB might lead to system panics or data corruption. + * When set to 0, the system will respect SOF1_IF_2KCL, which is set + * on the outgoing interface + * Set this to 1 for testing/debugging purposes only. + */ +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, ""); + +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"); + +int soreserveheadroom = 1; +SYSCTL_INT(_kern_ipc, OID_AUTO, soreserveheadroom, CTLFLAG_RW | CTLFLAG_LOCKED, + &soreserveheadroom, 0, "To allocate contiguous datagram buffers"); -void so_cache_timer(); +#if (DEBUG || DEVELOPMENT) +int so_notsent_lowat_check = 1; +SYSCTL_INT(_kern_ipc, OID_AUTO, notsent_lowat, CTLFLAG_RW | CTLFLAG_LOCKED, + &so_notsent_lowat_check, 0, "enable/disable notsnet lowat check"); +#endif /* DEBUG || DEVELOPMENT */ + +int so_accept_list_waits = 0; +#if (DEBUG || DEVELOPMENT) +SYSCTL_INT(_kern_ipc, OID_AUTO, accept_list_waits, CTLFLAG_RW | CTLFLAG_LOCKED, + &so_accept_list_waits, 0, "number of waits for listener incomp list"); +#endif /* DEBUG || DEVELOPMENT */ + +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); + +vm_size_t so_cache_zone_element_size; + +static int sodelayed_copy(struct socket *, struct uio *, struct mbuf **, + user_ssize_t *); +static void cached_sock_alloc(struct socket **, zalloc_flags_t); +static void cached_sock_free(struct socket *); /* - * 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. + * Maximum of extended background idle sockets per process + * Set to zero to disable further setting of the option */ -#ifdef __APPLE__ +#define SO_IDLE_BK_IDLE_MAX_PER_PROC 1 +#define SO_IDLE_BK_IDLE_TIME 600 +#define SO_IDLE_BK_IDLE_RCV_HIWAT 131072 + +struct soextbkidlestat soextbkidlestat; + +SYSCTL_UINT(_kern_ipc, OID_AUTO, maxextbkidleperproc, + CTLFLAG_RW | CTLFLAG_LOCKED, &soextbkidlestat.so_xbkidle_maxperproc, 0, + "Maximum of extended background idle sockets per process"); + +SYSCTL_UINT(_kern_ipc, OID_AUTO, extbkidletime, CTLFLAG_RW | CTLFLAG_LOCKED, + &soextbkidlestat.so_xbkidle_time, 0, + "Time in seconds to keep extended background idle sockets"); -vm_size_t so_cache_zone_element_size; +SYSCTL_UINT(_kern_ipc, OID_AUTO, extbkidlercvhiwat, CTLFLAG_RW | CTLFLAG_LOCKED, + &soextbkidlestat.so_xbkidle_rcvhiwat, 0, + "High water mark for extended background idle sockets"); -static int sodelayed_copy(struct socket *so, struct uio *uio, struct mbuf **free_list, int *resid); +SYSCTL_STRUCT(_kern_ipc, OID_AUTO, extbkidlestat, CTLFLAG_RD | CTLFLAG_LOCKED, + &soextbkidlestat, soextbkidlestat, ""); +int so_set_extended_bk_idle(struct socket *, int); -void socketinit() + +/* + * 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 = 0; +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))); + +#ifdef __LP64__ + _CASSERT(sizeof(struct sa_endpoints) == sizeof(struct user64_sa_endpoints)); + _CASSERT(offsetof(struct sa_endpoints, sae_srcif) == offsetof(struct user64_sa_endpoints, sae_srcif)); + _CASSERT(offsetof(struct sa_endpoints, sae_srcaddr) == offsetof(struct user64_sa_endpoints, sae_srcaddr)); + _CASSERT(offsetof(struct sa_endpoints, sae_srcaddrlen) == offsetof(struct user64_sa_endpoints, sae_srcaddrlen)); + _CASSERT(offsetof(struct sa_endpoints, sae_dstaddr) == offsetof(struct user64_sa_endpoints, sae_dstaddr)); + _CASSERT(offsetof(struct sa_endpoints, sae_dstaddrlen) == offsetof(struct user64_sa_endpoints, sae_dstaddrlen)); +#else + _CASSERT(sizeof(struct sa_endpoints) == sizeof(struct user32_sa_endpoints)); + _CASSERT(offsetof(struct sa_endpoints, sae_srcif) == offsetof(struct user32_sa_endpoints, sae_srcif)); + _CASSERT(offsetof(struct sa_endpoints, sae_srcaddr) == offsetof(struct user32_sa_endpoints, sae_srcaddr)); + _CASSERT(offsetof(struct sa_endpoints, sae_srcaddrlen) == offsetof(struct user32_sa_endpoints, sae_srcaddrlen)); + _CASSERT(offsetof(struct sa_endpoints, sae_dstaddr) == offsetof(struct user32_sa_endpoints, sae_dstaddr)); + _CASSERT(offsetof(struct sa_endpoints, sae_dstaddrlen) == offsetof(struct user32_sa_endpoints, sae_dstaddrlen)); +#endif - 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)); /* * 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); - so_cache_mtx_grp = lck_grp_alloc_init("so_cache", so_cache_mtx_grp_attr); - /* * allocate the lock attribute for socket cache mutex */ so_cache_mtx_attr = lck_attr_alloc_init(); - so_cache_init_done = 1; - - so_cache_mtx = lck_mtx_alloc_init(so_cache_mtx_grp, so_cache_mtx_attr); /* cached sockets mutex */ - - if (so_cache_mtx == NULL) - return; /* we're hosed... */ + /* 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); - str_size = (vm_size_t)( sizeof(struct socket) + 4 + - get_inpcb_str_size() + 4 + - get_tcp_str_size()); - so_cache_zone = zinit (str_size, 120000*str_size, 8192, "socache zone"); -#if TEMPDEBUG - printf("cached_sock_alloc -- so_cache_zone size is %x\n", str_size); -#endif - timeout(so_cache_timer, NULL, (SO_CACHE_FLUSH_INTERVAL * hz)); + so_cache_zone_element_size = (vm_size_t)(sizeof(struct socket) + 4 + + get_inpcb_str_size() + 4 + get_tcp_str_size()); - so_cache_zone_element_size = str_size; + so_cache_zone = zone_create("socache zone", so_cache_zone_element_size, + ZC_ZFREE_CLEARMEM | ZC_NOENCRYPT); - sflt_init(); + bzero(&soextbkidlestat, sizeof(struct soextbkidlestat)); + soextbkidlestat.so_xbkidle_maxperproc = SO_IDLE_BK_IDLE_MAX_PER_PROC; + soextbkidlestat.so_xbkidle_time = SO_IDLE_BK_IDLE_TIME; + soextbkidlestat.so_xbkidle_rcvhiwat = SO_IDLE_BK_IDLE_RCV_HIWAT; + in_pcbinit(); + sflt_init(); + socket_tclass_init(); +#if MULTIPATH + mp_pcbinit(); +#endif /* MULTIPATH */ } -void cached_sock_alloc(so, waitok) -struct socket **so; -int waitok; - +static void +cached_sock_alloc(struct socket **so, zalloc_flags_t how) { - caddr_t temp; - register u_long offset; - + caddr_t temp; + 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 %x - count == %d\n", *so, - cached_sock_count); -#endif - (*so)->so_saved_pcb = temp; - (*so)->cached_in_sock_layer = 1; + temp = (*so)->so_saved_pcb; + bzero((caddr_t)*so, sizeof(struct socket)); - } - else { -#if TEMPDEBUG - kprintf("Allocating cached sock %x from memory\n", *so); -#endif + (*so)->so_saved_pcb = temp; + } else { + lck_mtx_unlock(so_cache_mtx); - lck_mtx_unlock(so_cache_mtx); - - if (waitok) - *so = (struct socket *) zalloc(so_cache_zone); - else - *so = (struct socket *) zalloc_noblock(so_cache_zone); - - if (*so == 0) - return; - - bzero((caddr_t)*so, sizeof(struct socket)); - - /* - * Define offsets for extra structures into our single block of - * memory. Align extra structures on longword boundaries. - */ - - - offset = (u_long) *so; - offset += sizeof(struct socket); - if (offset & 0x3) { - offset += 4; - offset &= 0xfffffffc; - } - (*so)->so_saved_pcb = (caddr_t) offset; - offset += get_inpcb_str_size(); - if (offset & 0x3) { - offset += 4; - offset &= 0xfffffffc; - } - - ((struct inpcb *) (*so)->so_saved_pcb)->inp_saved_ppcb = (caddr_t) offset; -#if TEMPDEBUG - kprintf("Allocating cached socket - %x, pcb=%x tcpcb=%x\n", *so, - (*so)->so_saved_pcb, - ((struct inpcb *)(*so)->so_saved_pcb)->inp_saved_ppcb); -#endif - } + *so = zalloc_flags(so_cache_zone, how | Z_ZERO); - (*so)->cached_in_sock_layer = 1; -} + /* + * Define offsets for extra structures into our + * single block of memory. Align extra structures + * on longword boundaries. + */ + offset = (uintptr_t)*so; + offset += sizeof(struct socket); -void cached_sock_free(so) -struct socket *so; -{ + offset = ALIGN(offset); + + (*so)->so_saved_pcb = (caddr_t)offset; + offset += get_inpcb_str_size(); + offset = ALIGN(offset); + + ((struct inpcb *)(void *)(*so)->so_saved_pcb)->inp_saved_ppcb = + (caddr_t)offset; + } + + OSBitOrAtomic(SOF1_CACHED_IN_SOCK_LAYER, &(*so)->so_flags1); +} + +static void +cached_sock_free(struct socket *so) +{ lck_mtx_lock(so_cache_mtx); - if (++cached_sock_count > MAX_CACHED_SOCKETS) { + 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 %x\n", so); -#endif zfree(so_cache_zone, so); - } - else { -#if TEMPDEBUG - kprintf("Freeing socket %x into cache\n", so); -#endif - if (so_cache_hw < cached_sock_count) + } else { + 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 %x 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)); + if (so->so_proto != NULL && so->so_proto->pr_update_last_owner != NULL) { + (*so->so_proto->pr_update_last_owner)(so, self, NULL); + } + } + proc_pidoriginatoruuid(so->so_vuuid, sizeof(so->so_vuuid)); + } } +void +so_update_policy(struct socket *so) +{ + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) { + (void) inp_update_policy(sotoinpcb(so)); + } +} -void so_cache_timer() +#if NECP +static void +so_update_necp_policy(struct socket *so, struct sockaddr *override_local_addr, + struct sockaddr *override_remote_addr) { - register struct socket *p; - register int n_freed = 0; + 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 */ +boolean_t +so_cache_timer(void) +{ + struct socket *p; + int n_freed = 0; + boolean_t rc = FALSE; lck_mtx_lock(so_cache_mtx); + so_cache_timeouts++; + so_cache_time = net_uptime(); + + while (!STAILQ_EMPTY(&so_cache_head)) { + VERIFY(cached_sock_count > 0); + p = STAILQ_FIRST(&so_cache_head); + if ((so_cache_time - p->cache_timestamp) < + SO_CACHE_TIME_LIMIT) { + break; + } - ++so_cache_time; - - while ( (p = socket_cache_tail) ) - { - 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); - - if (++n_freed >= SO_CACHE_MAX_FREE_BATCH) - { - so_cache_max_freed++; + + if (++n_freed >= SO_CACHE_MAX_FREE_BATCH) { + so_cache_max_freed++; break; } } - 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. @@ -389,319 +614,656 @@ void so_cache_timer() * the protocols can be easily modified to do this. */ struct socket * -soalloc(waitok, dom, type) - int waitok; - int dom; - int type; +soalloc(int waitok, int dom, int type) { + zalloc_flags_t how = waitok ? Z_WAITOK : Z_NOWAIT; struct socket *so; - if ((dom == PF_INET) && (type == SOCK_STREAM)) - cached_sock_alloc(&so, waitok); - else - { - MALLOC_ZONE(so, struct socket *, sizeof(*so), socket_zone, M_WAITOK); - if (so) - bzero(so, sizeof *so); + if ((dom == PF_INET) && (type == SOCK_STREAM)) { + cached_sock_alloc(&so, how); + } else { + so = zalloc_flags(socket_zone, how | Z_ZERO); } - /* XXX race condition for reentrant kernel */ -//###LD Atomic add for so_gencnt - if (so) { - so->so_gencnt = ++so_gencnt; - so->so_zone = socket_zone; + if (so != NULL) { + so->so_gencnt = OSIncrementAtomic64((SInt64 *)&so_gencnt); + + /* + * Increment the socket allocation statistics + */ + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_alloc_total); } return so; } int -socreate(dom, aso, type, proto) - int dom; - struct socket **aso; - register int type; - int proto; +socreate_internal(int dom, struct socket **aso, int type, int proto, + struct proc *p, uint32_t flags, struct proc *ep) { - struct proc *p = current_proc(); - register struct protosw *prp; - register struct socket *so; - register int error = 0; + struct protosw *prp; + struct socket *so; + int error = 0; +#if defined(XNU_TARGET_OS_OSX) + pid_t rpid = -1; +#endif + #if TCPDEBUG extern int tcpconsdebug; #endif - if (proto) + + VERIFY(aso != NULL); + *aso = NULL; + + if (proto != 0) { prp = pffindproto(dom, proto, type); - else + } else { prp = pffindtype(dom, type); + } + + if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL) { + if (pffinddomain(dom) == NULL) { + return EAFNOSUPPORT; + } + if (proto != 0) { + if (pffindprotonotype(dom, proto) != NULL) { + return EPROTOTYPE; + } + } + return EPROTONOSUPPORT; + } + if (prp->pr_type != type) { + return EPROTOTYPE; + } + so = soalloc(1, dom, type); + if (so == NULL) { + return ENOBUFS; + } - if (prp == 0 || prp->pr_usrreqs->pru_attach == 0) - return (EPROTONOSUPPORT); -#ifndef __APPLE__ + switch (dom) { + case PF_LOCAL: + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_domain_local_total); + break; + case PF_INET: + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_domain_inet_total); + if (type == SOCK_STREAM) { + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_inet_stream_total); + } else { + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_inet_dgram_total); + } + break; + case PF_ROUTE: + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_domain_route_total); + break; + case PF_NDRV: + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_domain_ndrv_total); + break; + case PF_KEY: + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_domain_key_total); + break; + case PF_INET6: + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_domain_inet6_total); + if (type == SOCK_STREAM) { + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_inet6_stream_total); + } else { + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_inet6_dgram_total); + } + break; + case PF_SYSTEM: + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_domain_system_total); + break; + case PF_MULTIPATH: + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_domain_multipath_total); + break; + default: + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_domain_other_total); + break; + } - if (p->p_prison && jail_socket_unixiproute_only && - prp->pr_domain->dom_family != PF_LOCAL && - prp->pr_domain->dom_family != PF_INET && - prp->pr_domain->dom_family != PF_ROUTE) { - return (EPROTONOSUPPORT); + if (flags & SOCF_MPTCP) { + so->so_state |= SS_NBIO; } - -#endif - if (prp->pr_type != type) - return (EPROTOTYPE); - so = soalloc(p != 0, dom, type); - if (so == 0) - return (ENOBUFS); 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; +#if defined(XNU_TARGET_OS_OSX) + if (ep->p_responsible_pid != so->e_pid) { + rpid = ep->p_responsible_pid; + } +#endif + } -#ifdef __APPLE__ - if (p != 0) { - so->so_uid = kauth_cred_getuid(kauth_cred_get()); - if (!suser(kauth_cred_get(),NULL)) - so->so_state = SS_PRIV; +#if defined(XNU_TARGET_OS_OSX) + if (rpid < 0 && p->p_responsible_pid != so->last_pid) { + rpid = p->p_responsible_pid; + } + + so->so_rpid = -1; + uuid_clear(so->so_ruuid); + if (rpid >= 0) { + proc_t rp = proc_find(rpid); + if (rp != PROC_NULL) { + proc_getexecutableuuid(rp, so->so_ruuid, sizeof(so->so_ruuid)); + so->so_rpid = rpid; + proc_rele(rp); + } } -#else - so->so_cred = kauth_cred_get_with_ref(); #endif + + so->so_cred = kauth_cred_proc_ref(p); + if (!suser(kauth_cred_get(), NULL)) { + 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; - - -//### Attachement will create the per pcb lock if necessary and increase refcount - so->so_usecount++; /* 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) { - /* - * Warning: - * If so_pcb is not zero, the socket will be leaked, - * so protocol attachment handler must be coded carefuly + if (error != 0) { + /* + * Warning: + * If so_pcb is not zero, the socket will be leaked, + * so protocol attachment handler must be coded carefuly */ so->so_state |= SS_NOFDREF; + VERIFY(so->so_usecount > 0); so->so_usecount--; - sofreelastref(so, 1); /* will deallocate the socket */ - return (error); + sofreelastref(so, 1); /* will deallocate the socket */ + return error; } -#ifdef __APPLE__ - prp->pr_domain->dom_refs++; - TAILQ_INIT(&so->so_evlist); + + /* + * Note: needs so_pcb to be set after pru_attach + */ + if (prp->pr_update_last_owner != NULL) { + (*prp->pr_update_last_owner)(so, p, ep); + } + + atomic_add_32(&prp->pr_domain->dom_refs, 1); /* Attach socket filters for this protocol */ sflt_initsock(so); #if TCPDEBUG - if (tcpconsdebug == 2) + if (tcpconsdebug == 2) { so->so_options |= SO_DEBUG; + } #endif -#endif + so_set_default_traffic_class(so); + + /* + * If this thread or task is marked to create backgrounded sockets, + * mark the socket as background. + */ + if (!(flags & SOCF_MPTCP) && + proc_get_effective_thread_policy(current_thread(), TASK_POLICY_NEW_SOCKETS_BG)) { + socket_set_traffic_mgt_flags(so, TRAFFIC_MGT_SO_BACKGROUND); + so->so_background_thread = current_thread(); + } + + switch (dom) { + /* + * Don't mark Unix domain or system + * eligible for defunct by default. + */ + case PF_LOCAL: + case PF_SYSTEM: + 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); + + return 0; } +/* + * Returns: 0 Success + * EAFNOSUPPORT + * EPROTOTYPE + * EPROTONOSUPPORT + * ENOBUFS + * :ENOBUFS[AF_UNIX] + * :ENOBUFS[TCP] + * :ENOMEM[TCP] + * :??? [other protocol families, IPSEC] + */ int -sobind(so, nam) - struct socket *so; - struct sockaddr *nam; +socreate(int dom, struct socket **aso, int type, int proto) +{ + return socreate_internal(dom, aso, type, proto, current_proc(), 0, + PROC_NULL); +} +int +socreate_delegate(int dom, struct socket **aso, int type, int proto, pid_t epid) { - struct proc *p = current_proc(); int error = 0; - struct socket_filter_entry *filter; - int filtered = 0; - - socket_lock(so, 1); + struct proc *ep = PROC_NULL; - /* 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); + if ((proc_selfpid() != epid) && ((ep = proc_find(epid)) == PROC_NULL)) { + error = ESRCH; + goto done; } - /* End socket filter */ - - if (error == 0) - error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, p); - - socket_unlock(so, 1); - - if (error == EJUSTRETURN) - error = 0; - - return (error); -} -void -sodealloc(so) - struct socket *so; -{ - so->so_gencnt = ++so_gencnt; - -#ifndef __APPLE__ - if (so->so_rcv.sb_hiwat) - (void)chgsbsize(so->so_cred->cr_uidinfo, - &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); - if (so->so_snd.sb_hiwat) - (void)chgsbsize(so->so_cred->cr_uidinfo, - &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); -#ifdef INET - if (so->so_accf != NULL) { - if (so->so_accf->so_accept_filter != NULL && - so->so_accf->so_accept_filter->accf_destroy != NULL) { - so->so_accf->so_accept_filter->accf_destroy(so); - } - if (so->so_accf->so_accept_filter_str != NULL) - FREE(so->so_accf->so_accept_filter_str, M_ACCF); - FREE(so->so_accf, M_ACCF); - } -#endif /* INET */ - kauth_cred_unref(&so->so_cred); - zfreei(so->so_zone, so); -#else - if (so->cached_in_sock_layer == 1) - cached_sock_free(so); - else { - if (so->cached_in_sock_layer == -1) - panic("sodealloc: double dealloc: so=%x\n", so); - so->cached_in_sock_layer = -1; - FREE_ZONE(so, sizeof(*so), so->so_zone); + 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); } -#endif /* __APPLE__ */ + + return error; } +/* + * Returns: 0 Success + * :EINVAL Invalid argument [COMMON_START] + * :EAFNOSUPPORT Address family not supported + * :EADDRNOTAVAIL Address not available. + * :EINVAL Invalid argument + * :EAFNOSUPPORT Address family not supported [notdef] + * :EACCES Permission denied + * :EADDRINUSE Address in use + * :EAGAIN Resource unavailable, try again + * :EPERM Operation not permitted + * :??? + * :??? + * + * Notes: It's not possible to fully enumerate the return codes above, + * since socket filter authors and protocol family authors may + * not choose to limit their error returns to those listed, even + * though this may result in some software operating incorrectly. + * + * The error codes which are enumerated above are those known to + * be returned by the tcp_usr_bind function supplied. + */ int -solisten(so, backlog) - register struct socket *so; - int backlog; - +sobindlock(struct socket *so, struct sockaddr *nam, int dolock) { struct proc *p = current_proc(); - int error; + int error = 0; - socket_lock(so, 1); - - { - 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_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 (dolock) { + socket_lock(so, 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 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, %s]: defunct so 0x%llx [%d,%d] (%d)\n", + __func__, proc_pid(p), proc_best_name(p), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error); + goto out; } + /* Socket filter */ + error = sflt_bind(so, nam); + if (error == 0) { - error = (*so->so_proto->pr_usrreqs->pru_listen)(so, p); + error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, p); } - - if (error) { +out: + if (dolock) { socket_unlock(so, 1); - if (error == EJUSTRETURN) - error = 0; - return (error); } - - if (TAILQ_EMPTY(&so->so_comp)) - so->so_options |= SO_ACCEPTCONN; - if (backlog < 0 || backlog > somaxconn) - backlog = somaxconn; - so->so_qlimit = backlog; - socket_unlock(so, 1); - return (0); + if (error == EJUSTRETURN) { + error = 0; + } + + return error; } void -sofreelastref(so, dealloc) - register struct socket *so; - int dealloc; +sodealloc(struct socket *so) { - int error; - struct socket *head = so->so_head; - - /*### Assume socket is locked */ + kauth_cred_unref(&so->so_cred); - /* Remove any filters - may be called more than once */ + /* Remove any filters */ sflt_termsock(so); - - if ((!(so->so_flags & SOF_PCBCLEARING)) || ((so->so_state & SS_NOFDREF) == 0)) { -#ifdef __APPLE__ - selthreadclear(&so->so_snd.sb_sel); + +#if CONTENT_FILTER + cfil_sock_detach(so); +#endif /* CONTENT_FILTER */ + + so->so_gencnt = OSIncrementAtomic64((SInt64 *)&so_gencnt); + + if (so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) { + cached_sock_free(so); + } else { + zfree(socket_zone, so); + } +} + +/* + * Returns: 0 Success + * EINVAL + * EOPNOTSUPP + * :EINVAL[AF_UNIX] + * :EINVAL[TCP] + * :EADDRNOTAVAIL[TCP] Address not available. + * :EINVAL[TCP] Invalid argument + * :EAFNOSUPPORT[TCP] Address family not supported [notdef] + * :EACCES[TCP] Permission denied + * :EADDRINUSE[TCP] Address in use + * :EAGAIN[TCP] Resource unavailable, try again + * :EPERM[TCP] Operation not permitted + * :??? + * + * Notes: Other returns depend on the protocol family; all + * returns depend on what the filter author causes + * their filter to return. + */ +int +solisten(struct socket *so, int backlog) +{ + struct proc *p = current_proc(); + int error = 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; + } + if ((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0) { + error = EOPNOTSUPP; + goto out; + } + + /* + * If the listen request is made on a socket that is not fully + * disconnected, or on a socket that has been marked as inactive, + * reject the request now. + */ + if ((so->so_state & + (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) || + (so->so_flags & SOF_DEFUNCT)) { + error = EINVAL; + if (so->so_flags & SOF_DEFUNCT) { + SODEFUNCTLOG("%s[%d, %s]: defunct so 0x%llx [%d,%d] " + "(%d)\n", __func__, proc_pid(p), + proc_best_name(p), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error); + } + goto out; + } + + if ((so->so_restrictions & SO_RESTRICT_DENY_IN) != 0) { + error = EPERM; + goto out; + } + + error = sflt_listen(so); + if (error == 0) { + error = (*so->so_proto->pr_usrreqs->pru_listen)(so, p); + } + + if (error) { + if (error == EJUSTRETURN) { + error = 0; + } + goto out; + } + + if (TAILQ_EMPTY(&so->so_comp)) { + so->so_options |= SO_ACCEPTCONN; + } + /* + * POSIX: The implementation may have an upper limit on the length of + * the listen queue-either global or per accepting socket. If backlog + * exceeds this limit, the length of the listen queue is set to the + * limit. + * + * If listen() is called with a backlog argument value that is less + * than 0, the function behaves as if it had been called with a backlog + * argument value of 0. + * + * A backlog argument of 0 may allow the socket to accept connections, + * in which case the length of the listen queue may be set to an + * implementation-defined minimum value. + */ + if (backlog <= 0 || backlog > somaxconn) { + backlog = somaxconn; + } + + so->so_qlimit = backlog; +out: + socket_unlock(so, 1); + return error; +} + +/* + * The "accept list lock" protects the fields related to the listener queues + * because we can unlock a socket to respect the lock ordering between + * the listener socket and its clients sockets. The lock ordering is first to + * acquire the client socket before the listener socket. + * + * The accept list lock serializes access to the following fields: + * - of the listener socket: + * - so_comp + * - so_incomp + * - so_qlen + * - so_inqlen + * - of client sockets that are in so_comp or so_incomp: + * - so_head + * - so_list + * + * As one can see the accept list lock protects the consistent of the + * linkage of the client sockets. + * + * Note that those fields may be read without holding the accept list lock + * for a preflight provided the accept list lock is taken when committing + * to take an action based on the result of the preflight. The preflight + * saves the cost of doing the unlock/lock dance. + */ +void +so_acquire_accept_list(struct socket *head, struct socket *so) +{ + lck_mtx_t *mutex_held; + + if (head->so_proto->pr_getlock == NULL) { + return; + } + mutex_held = (*head->so_proto->pr_getlock)(head, PR_F_WILLUNLOCK); + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); + + if (!(head->so_flags1 & SOF1_ACCEPT_LIST_HELD)) { + head->so_flags1 |= SOF1_ACCEPT_LIST_HELD; + return; + } + if (so != NULL) { + socket_unlock(so, 0); + } + while (head->so_flags1 & SOF1_ACCEPT_LIST_HELD) { + so_accept_list_waits += 1; + msleep((caddr_t)&head->so_incomp, mutex_held, + PSOCK | PCATCH, __func__, NULL); + } + head->so_flags1 |= SOF1_ACCEPT_LIST_HELD; + if (so != NULL) { + socket_unlock(head, 0); + socket_lock(so, 0); + socket_lock(head, 0); + } +} + +void +so_release_accept_list(struct socket *head) +{ + if (head->so_proto->pr_getlock != NULL) { + lck_mtx_t *mutex_held; + + mutex_held = (*head->so_proto->pr_getlock)(head, 0); + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); + + head->so_flags1 &= ~SOF1_ACCEPT_LIST_HELD; + wakeup((caddr_t)&head->so_incomp); + } +} + +void +sofreelastref(struct socket *so, int dealloc) +{ + struct socket *head = so->so_head; + + /* Assume socket is locked */ + + 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) { - socket_lock(head, 1); + /* + * Need to lock the listener when the protocol has + * per socket locks + */ + if (head->so_proto->pr_getlock != NULL) { + socket_lock(head, 1); + so_acquire_accept_list(head, so); + } if (so->so_state & SS_INCOMP) { + so->so_state &= ~SS_INCOMP; TAILQ_REMOVE(&head->so_incomp, so, so_list); head->so_incqlen--; + head->so_qlen--; + so->so_head = NULL; + + if (head->so_proto->pr_getlock != NULL) { + so_release_accept_list(head); + socket_unlock(head, 1); + } } else if (so->so_state & SS_COMP) { + if (head->so_proto->pr_getlock != NULL) { + so_release_accept_list(head); + socket_unlock(head, 1); + } /* * We must not decommission a socket that's * on the accept(2) queue. If we do, then * 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 - socket_unlock(head, 1); + so->so_rcv.sb_flags &= ~(SB_SEL | SB_UPCALL); + so->so_snd.sb_flags &= ~(SB_SEL | SB_UPCALL); + so->so_event = sonullevent; return; } else { - panic("sofree: not queued"); + if (head->so_proto->pr_getlock != NULL) { + so_release_accept_list(head); + socket_unlock(head, 1); + } + printf("sofree: not queued\n"); } - head->so_qlen--; - so->so_state &= ~SS_INCOMP; - so->so_head = NULL; - socket_unlock(head, 1); } -#ifdef __APPLE__ - selthreadclear(&so->so_snd.sb_sel); - sbrelease(&so->so_snd); -#endif + sowflush(so); sorflush(so); - + +#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; - - if (dealloc) + so->so_snd.sb_flags &= ~(SB_UPCALL | SB_SNDBYTE_CNT); + so->so_event = sonullevent; + + if (dealloc) { sodealloc(so); + } +} + +void +soclose_wait_locked(struct socket *so) +{ + lck_mtx_t *mutex_held; + + if (so->so_proto->pr_getlock != NULL) { + mutex_held = (*so->so_proto->pr_getlock)(so, PR_F_WILLUNLOCK); + } else { + mutex_held = so->so_proto->pr_domain->dom_mtx; + } + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); + + /* + * Double check here and return if there's no outstanding upcall; + * otherwise proceed further only if SOF_UPCALLCLOSEWAIT is set. + */ + if (!so->so_upcallusecount || !(so->so_flags & SOF_UPCALLCLOSEWAIT)) { + return; + } + so->so_rcv.sb_flags &= ~SB_UPCALL; + so->so_snd.sb_flags &= ~SB_UPCALL; + so->so_flags |= SOF_CLOSEWAIT; + + (void) msleep((caddr_t)&so->so_upcallusecount, mutex_held, (PZERO - 1), + "soclose_wait_locked", NULL); + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); + so->so_flags &= ~SOF_CLOSEWAIT; } /* @@ -710,63 +1272,146 @@ sofreelastref(so, dealloc) * Free socket when disconnect complete. */ int -soclose_locked(so) - register struct socket *so; +soclose_locked(struct socket *so) { int error = 0; - lck_mtx_t * mutex_held; struct timespec ts; if (so->so_usecount == 0) { - panic("soclose: so=%x refcount=0\n", so); + 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_flags1 & SOF1_EXTEND_BK_IDLE_INPROG) { + soresume(current_proc(), so, 1); + so->so_flags1 &= ~SOF1_EXTEND_BK_IDLE_WANTED; + } + if ((so->so_options & SO_ACCEPTCONN)) { - struct socket *sp; - - /* We do not want new connection to be added to the connection queues */ + struct socket *sp, *sonext; + int persocklock = 0; + int incomp_overflow_only; + + /* + * We do not want new connection to be added + * to the connection queues + */ so->so_options &= ~SO_ACCEPTCONN; - - while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { - /* A bit tricky here. We need to keep - * a lock if it's a protocol global lock - * but we want the head, not the socket locked - * in the case of per-socket lock... + + /* + * We can drop the lock on the listener once + * we've acquired the incoming list + */ + if (so->so_proto->pr_getlock != NULL) { + persocklock = 1; + so_acquire_accept_list(so, NULL); + socket_unlock(so, 0); + } +again: + incomp_overflow_only = 1; + + TAILQ_FOREACH_SAFE(sp, &so->so_incomp, so_list, sonext) { + /* + * 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 (so->so_proto->pr_getlock != NULL) { - socket_unlock(so, 0); + if (sp->so_flags & SOF_OVERFLOW) { + continue; + } + + if (persocklock != 0) { socket_lock(sp, 1); } - (void) soabort(sp); - if (so->so_proto->pr_getlock != NULL) { + + /* + * Radar 27945981 + * The extra reference for the list insure the + * validity of the socket pointer when we perform the + * unlock of the head above + */ + if (sp->so_state & SS_INCOMP) { + sp->so_state &= ~SS_INCOMP; + sp->so_head = NULL; + TAILQ_REMOVE(&so->so_incomp, sp, so_list); + so->so_incqlen--; + so->so_qlen--; + + (void) soabort(sp); + } else { + panic("%s sp %p in so_incomp but !SS_INCOMP", + __func__, sp); + } + + if (persocklock != 0) { socket_unlock(sp, 1); - socket_lock(so, 0); } } - while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { + TAILQ_FOREACH_SAFE(sp, &so->so_comp, so_list, sonext) { /* Dequeue from so_comp since sofree() won't do it */ - TAILQ_REMOVE(&so->so_comp, sp, so_list); - so->so_qlen--; - - if (so->so_proto->pr_getlock != NULL) { - socket_unlock(so, 0); + if (persocklock != 0) { socket_lock(sp, 1); } - sp->so_state &= ~SS_COMP; - sp->so_head = NULL; + if (sp->so_state & SS_COMP) { + sp->so_state &= ~SS_COMP; + sp->so_head = NULL; + TAILQ_REMOVE(&so->so_comp, sp, so_list); + so->so_qlen--; - (void) soabort(sp); - if (so->so_proto->pr_getlock != NULL) { + (void) soabort(sp); + } else { + panic("%s sp %p in so_comp but !SS_COMP", + __func__, sp); + } + + if (persocklock) { socket_unlock(sp, 1); - socket_lock(so, 0); } } + + if (incomp_overflow_only == 0 && !TAILQ_EMPTY(&so->so_incomp)) { +#if (DEBUG | DEVELOPMENT) + panic("%s head %p so_comp not empty\n", __func__, so); +#endif /* (DEVELOPMENT || DEBUG) */ + + goto again; + } + + if (!TAILQ_EMPTY(&so->so_comp)) { +#if (DEBUG | DEVELOPMENT) + panic("%s head %p so_comp not empty\n", __func__, so); +#endif /* (DEVELOPMENT || DEBUG) */ + + goto again; + } + + if (persocklock) { + socket_lock(so, 0); + so_release_accept_list(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; @@ -774,223 +1419,447 @@ soclose_locked(so) if (so->so_state & SS_ISCONNECTED) { if ((so->so_state & SS_ISDISCONNECTING) == 0) { error = sodisconnectlocked(so); - if (error) + if (error) { goto drop; + } } if (so->so_options & SO_LINGER) { + lck_mtx_t *mutex_held; + if ((so->so_state & SS_ISDISCONNECTING) && - (so->so_state & SS_NBIO)) + (so->so_state & SS_NBIO)) { goto drop; - if (so->so_proto->pr_getlock != NULL) - mutex_held = (*so->so_proto->pr_getlock)(so, 0); - else + } + if (so->so_proto->pr_getlock != NULL) { + mutex_held = (*so->so_proto->pr_getlock)(so, PR_F_WILLUNLOCK); + } else { mutex_held = so->so_proto->pr_domain->dom_mtx; + } while (so->so_state & SS_ISCONNECTED) { - ts.tv_sec = (so->so_linger/100); - ts.tv_nsec = (so->so_linger % 100) * NSEC_PER_USEC * 1000 * 10; - error = msleep((caddr_t)&so->so_timeo, mutex_held, - PSOCK | PCATCH, "soclos", &ts); + ts.tv_sec = (so->so_linger / 100); + ts.tv_nsec = (so->so_linger % 100) * + NSEC_PER_USEC * 1000 * 10; + error = msleep((caddr_t)&so->so_timeo, + mutex_held, PSOCK | PCATCH, "soclose", &ts); if (error) { - /* It's OK when the time fires, don't report an error */ - if (error == EWOULDBLOCK) + /* + * It's OK when the time fires, + * don't report an error + */ + if (error == EWOULDBLOCK) { error = 0; + } break; } } } } drop: - if (so->so_usecount == 0) - panic("soclose: usecount is zero so=%x\n", so); - if (so->so_pcb && !(so->so_flags & SOF_PCBCLEARING)) { + if (so->so_usecount == 0) { + panic("soclose: usecount is zero so=%p\n", so); + /* NOTREACHED */ + } + if (so->so_pcb != NULL && !(so->so_flags & SOF_PCBCLEARING)) { int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so); - if (error == 0) + if (error == 0) { error = error2; + } + } + if (so->so_usecount <= 0) { + panic("soclose: usecount is zero so=%p\n", so); + /* NOTREACHED */ } - if (so->so_usecount <= 0) - panic("soclose: usecount is zero so=%x\n", so); 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--; - evsofree(so); -#endif + + 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); + + VERIFY(so->so_usecount > 0); so->so_usecount--; sofree(so); - return (error); + return error; } int -soclose(so) - register struct socket *so; +soclose(struct socket *so) { int error = 0; socket_lock(so, 1); - if (so->so_retaincnt == 0) + + if (so->so_retaincnt == 0) { error = soclose_locked(so); - else { /* if the FD is going away, but socket is retained in kernel remove its reference */ + } else { + /* + * if the FD is going away, but socket is + * retained in kernel remove its reference + */ so->so_usecount--; - if (so->so_usecount < 2) - panic("soclose: retaincnt non null and so=%x usecount=%x\n", so->so_usecount); + if (so->so_usecount < 2) { + panic("soclose: retaincnt non null and so=%p " + "usecount=%d\n", so, so->so_usecount); + } } socket_unlock(so, 1); - return (error); + return error; } - /* * Must be called at splnet... */ -//#### Should already be locked +/* Should already be locked */ int -soabort(so) - struct socket *so; +soabort(struct socket *so) { int error; #ifdef MORE_LOCKING_DEBUG - lck_mtx_t * mutex_held; + lck_mtx_t *mutex_held; - if (so->so_proto->pr_getlock != NULL) + if (so->so_proto->pr_getlock != NULL) { mutex_held = (*so->so_proto->pr_getlock)(so, 0); - else + } else { mutex_held = so->so_proto->pr_domain->dom_mtx; - lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); + } + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); #endif - error = (*so->so_proto->pr_usrreqs->pru_abort)(so); - if (error) { - sofree(so); - return error; + if ((so->so_flags & SOF_ABORTED) == 0) { + so->so_flags |= SOF_ABORTED; + error = (*so->so_proto->pr_usrreqs->pru_abort)(so); + if (error) { + sofree(so); + return error; + } } - return (0); + return 0; } int -soacceptlock(so, nam, dolock) - register struct socket *so; - struct sockaddr **nam; - int dolock; +soacceptlock(struct socket *so, struct sockaddr **nam, int dolock) { int error; - if (dolock) socket_lock(so, 1); + 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) + if ((so->so_state & SS_NOFDREF) == 0) { panic("soaccept: !NOFDREF"); + } so->so_state &= ~SS_NOFDREF; error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); - - if (dolock) socket_unlock(so, 1); - return (error); + + if (dolock) { + socket_unlock(so, 1); + } + return error; } + int -soaccept(so, nam) - register struct socket *so; - struct sockaddr **nam; +soaccept(struct socket *so, struct sockaddr **nam) { - return (soacceptlock(so, nam, 1)); + return soacceptlock(so, nam, 1); } int -soconnectlock(so, nam, dolock) - register struct socket *so; - struct sockaddr *nam; - int dolock; - +soacceptfilter(struct socket *so, struct socket *head) { - int s; - int error; - struct proc *p = current_proc(); - - if (dolock) socket_lock(so, 1); + struct sockaddr *local = NULL, *remote = NULL; + int error = 0; - if (so->so_options & SO_ACCEPTCONN) { - if (dolock) socket_unlock(so, 1); - return (EOPNOTSUPP); + /* + * Hold the lock even if this socket has not been made visible + * to the filter(s). For sockets with global locks, this protects + * against the head or peer going away + */ + socket_lock(so, 1); + if (sogetaddr_locked(so, &remote, 1) != 0 || + sogetaddr_locked(so, &local, 0) != 0) { + so->so_state &= ~SS_NOFDREF; + socket_unlock(so, 1); + soclose(so); + /* Out of resources; try it again next time */ + error = ECONNABORTED; + goto done; } + + error = sflt_accept(head, so, local, remote); + /* - * If protocol is connection-based, can only connect once. - * Otherwise, if connected, try to disconnect first. - * This allows user to disconnect by connecting to, e.g., - * a null address. + * If we get EJUSTRETURN from one of the filters, mark this socket + * as inactive and return it anyway. This newly accepted socket + * will be disconnected later before we hand it off to the caller. */ - if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && - ((so->so_proto->pr_flags & PR_CONNREQUIRED) || - (error = sodisconnectlocked(so)))) - error = EISCONN; - else { + if (error == EJUSTRETURN) { + error = 0; + (void) sosetdefunct(current_proc(), so, + SHUTDOWN_SOCKET_LEVEL_DISCONNECT_INTERNAL, FALSE); + } + + if (error != 0) { /* - * Run connect filter before calling protocol: - * - non-blocking connect returns before completion; + * This may seem like a duplication to the above error + * handling part when we return ECONNABORTED, except + * the following is done while holding the lock since + * the socket has been exposed to the filter(s) earlier. */ - { - 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) { - if (error == EJUSTRETURN) - error = 0; - if (dolock) socket_unlock(so, 1); - return error; - } - - error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, p); + so->so_state &= ~SS_NOFDREF; + socket_unlock(so, 1); + soclose(so); + /* Propagate socket filter's error code to the caller */ + } else { + socket_unlock(so, 1); } - if (dolock) socket_unlock(so, 1); - return (error); +done: + /* Callee checks for NULL pointer */ + sock_freeaddr(remote); + sock_freeaddr(local); + return error; +} + +/* + * Returns: 0 Success + * EOPNOTSUPP Operation not supported on socket + * EISCONN Socket is connected + * :EADDRNOTAVAIL Address not available. + * :EINVAL Invalid argument + * :EAFNOSUPPORT Address family not supported [notdef] + * :EACCES Permission denied + * :EADDRINUSE Address in use + * :EAGAIN Resource unavailable, try again + * :EPERM Operation not permitted + * :??? [anything a filter writer might set] + */ +int +soconnectlock(struct socket *so, struct sockaddr *nam, int dolock) +{ + int error; + struct proc *p = current_proc(); + + 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, %s]: defunct so 0x%llx [%d,%d] " + "(%d)\n", __func__, proc_pid(p), + proc_best_name(p), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error); + } + if (dolock) { + socket_unlock(so, 1); + } + return error; + } + + if ((so->so_restrictions & SO_RESTRICT_DENY_OUT) != 0) { + if (dolock) { + socket_unlock(so, 1); + } + return EPERM; + } + + /* + * If protocol is connection-based, can only connect once. + * Otherwise, if connected, try to disconnect first. + * 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_CONNREQUIRED) || + (error = sodisconnectlocked(so)))) { + error = EISCONN; + } else { + /* + * Run connect filter before calling protocol: + * - non-blocking connect returns before completion; + */ + error = sflt_connectout(so, nam); + if (error != 0) { + if (error == EJUSTRETURN) { + error = 0; + } + } else { + error = (*so->so_proto->pr_usrreqs->pru_connect) + (so, nam, p); + if (error != 0) { + so->so_state &= ~SS_ISCONNECTING; + } + } + } + if (dolock) { + socket_unlock(so, 1); + } + return error; } int -soconnect(so, nam) - register struct socket *so; - struct sockaddr *nam; +soconnect(struct socket *so, struct sockaddr *nam) { - return (soconnectlock(so, nam, 1)); + return soconnectlock(so, nam, 1); } +/* + * Returns: 0 Success + * :EINVAL[AF_UNIX] + * :EPROTOTYPE[AF_UNIX] + * :??? [other protocol families] + * + * Notes: is not supported by [TCP]. + */ int -soconnect2(so1, so2) - register struct socket *so1; - struct socket *so2; +soconnect2(struct socket *so1, struct socket *so2) { int error; socket_lock(so1, 1); - if (so2->so_proto->pr_lock) + if (so2->so_proto->pr_lock) { socket_lock(so2, 1); + } error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); - + socket_unlock(so1, 1); - if (so2->so_proto->pr_lock) + if (so2->so_proto->pr_lock) { socket_unlock(so2, 1); - return (error); + } + return error; } +int +soconnectxlocked(struct socket *so, struct sockaddr *src, + struct sockaddr *dst, struct proc *p, uint32_t ifscope, + sae_associd_t aid, sae_connid_t *pcid, uint32_t flags, void *arg, + uint32_t arglen, uio_t auio, user_ssize_t *bytes_written) +{ + int error; + + 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, %s]: defunct so 0x%llx [%d,%d] " + "(%d)\n", __func__, proc_pid(p), + proc_best_name(p), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error); + } + return error; + } + + if ((so->so_restrictions & SO_RESTRICT_DENY_OUT) != 0) { + return EPERM; + } + + /* + * If protocol is connection-based, can only connect once + * unless PR_MULTICONN is set. Otherwise, if connected, + * try to disconnect first. This allows user to disconnect + * by connecting to, e.g., a null address. + */ + if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) && + !(so->so_proto->pr_flags & PR_MULTICONN) && + ((so->so_proto->pr_flags & PR_CONNREQUIRED) || + (error = sodisconnectlocked(so)) != 0)) { + error = EISCONN; + } else { + if ((so->so_proto->pr_flags & PR_DATA_IDEMPOTENT) && + (flags & CONNECT_DATA_IDEMPOTENT)) { + so->so_flags1 |= SOF1_DATA_IDEMPOTENT; + + if (flags & CONNECT_DATA_AUTHENTICATED) { + so->so_flags1 |= SOF1_DATA_AUTHENTICATED; + } + } + + /* + * Case 1: CONNECT_RESUME_ON_READ_WRITE set, no data. + * Case 2: CONNECT_RESUME_ON_READ_WRITE set, with data (user error) + * Case 3: CONNECT_RESUME_ON_READ_WRITE not set, with data + * Case 3 allows user to combine write with connect even if they have + * no use for TFO (such as regular TCP, and UDP). + * Case 4: CONNECT_RESUME_ON_READ_WRITE not set, no data (regular case) + */ + if ((so->so_proto->pr_flags & PR_PRECONN_WRITE) && + ((flags & CONNECT_RESUME_ON_READ_WRITE) || auio)) { + so->so_flags1 |= SOF1_PRECONNECT_DATA; + } + + /* + * If a user sets data idempotent and does not pass an uio, or + * sets CONNECT_RESUME_ON_READ_WRITE, this is an error, reset + * SOF1_DATA_IDEMPOTENT. + */ + if (!(so->so_flags1 & SOF1_PRECONNECT_DATA) && + (so->so_flags1 & SOF1_DATA_IDEMPOTENT)) { + /* We should return EINVAL instead perhaps. */ + so->so_flags1 &= ~SOF1_DATA_IDEMPOTENT; + } + + /* + * Run connect filter before calling protocol: + * - non-blocking connect returns before completion; + */ + error = sflt_connectout(so, dst); + if (error != 0) { + /* Disable PRECONNECT_DATA, as we don't need to send a SYN anymore. */ + so->so_flags1 &= ~SOF1_PRECONNECT_DATA; + if (error == EJUSTRETURN) { + error = 0; + } + } else { + error = (*so->so_proto->pr_usrreqs->pru_connectx) + (so, src, dst, p, ifscope, aid, pcid, + flags, arg, arglen, auio, bytes_written); + if (error != 0) { + so->so_state &= ~SS_ISCONNECTING; + if (error != EINPROGRESS) { + so->so_flags1 &= ~SOF1_PRECONNECT_DATA; + } + } + } + } + + return error; +} int -sodisconnectlocked(so) - register struct socket *so; +sodisconnectlocked(struct socket *so) { int error; @@ -1002,55 +1871,88 @@ sodisconnectlocked(so) error = EALREADY; goto bad; } - + error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); - if (error == 0) { sflt_notify(so, sock_evt_disconnected, NULL); } bad: - return (error); + return error; } -//### Locking version + +/* Locking version */ int -sodisconnect(so) - register struct socket *so; +sodisconnect(struct socket *so) { - int error; + int error; socket_lock(so, 1); error = sodisconnectlocked(so); socket_unlock(so, 1); - return(error); + return error; +} + +int +sodisconnectxlocked(struct socket *so, sae_associd_t aid, sae_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, sae_associd_t aid, sae_connid_t cid) +{ + int error; + + socket_lock(so, 1); + error = sodisconnectxlocked(so, aid, cid); + socket_unlock(so, 1); + return error; } -#define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_DONTWAIT : M_WAIT) +#define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) /* * sosendcheck will lock the socket buffer if it isn't locked and * verify that there is space for the data being inserted. + * + * Returns: 0 Success + * EPIPE + * sblock:EWOULDBLOCK + * sblock:EINTR + * sbwait:EBADF + * sbwait:EINTR + * [so_error]:??? */ - -static int -sosendcheck( - struct socket *so, - struct sockaddr *addr, - long resid, - long clen, - long 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) { - int error = 0; - long space; - int assumelock = 0; + int error = 0; + int32_t space; + int assumelock = 0; restart: if (*sblocked == 0) { if ((so->so_snd.sb_flags & SB_LOCK) != 0 && - so->so_send_filt_thread != 0 && - so->so_send_filt_thread == current_thread()) { + so->so_send_filt_thread != 0 && + so->so_send_filt_thread == current_thread()) { /* * We're being called recursively from a filter, * allow this to continue. Radar 4150520. @@ -1058,59 +1960,107 @@ restart: * to perform an unlock later. */ assumelock = 1; - } - else { + } else { error = sblock(&so->so_snd, SBLOCKWAIT(flags)); if (error) { + if (so->so_flags & SOF_DEFUNCT) { + goto defunct; + } return error; } *sblocked = 1; } } - - if (so->so_state & SS_CANTSENDMORE) + + /* + * If a send attempt is made on a socket that has been marked + * as inactive (disconnected), reject the request. + */ + if (so->so_flags & SOF_DEFUNCT) { +defunct: + error = EPIPE; + SODEFUNCTLOG("%s[%d, %s]: defunct so 0x%llx [%d,%d] (%d)\n", + __func__, proc_selfpid(), proc_best_name(current_proc()), + (uint64_t)DEBUG_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)DEBUG_KERNEL_ADDRPERM(so)); + } else +#endif /* CONTENT_FILTER */ return EPIPE; - + } if (so->so_error) { error = so->so_error; so->so_error = 0; return error; } - + if ((so->so_state & SS_ISCONNECTED) == 0) { - /* - * `sendto' and `sendmsg' is allowed on a connection- - * based socket if it supports implied connect. - * Return ENOTCONN if not connected and no address is - * supplied. - */ - if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && - (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { - if ((so->so_state & SS_ISCONFIRMING) == 0 && - !(resid == 0 && clen != 0)) + if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { + if (((so->so_state & SS_ISCONFIRMING) == 0) && + (resid != 0 || clen == 0) && + !(so->so_flags1 & SOF1_PRECONNECT_DATA)) { return ENOTCONN; - } else if (addr == 0 && !(flags&MSG_HOLD)) - return (so->so_proto->pr_flags & PR_CONNREQUIRED) ? ENOTCONN : EDESTADDRREQ; + } + } else if (addr == 0) { + return (so->so_proto->pr_flags & PR_CONNREQUIRED) ? + ENOTCONN : EDESTADDRREQ; + } } + space = sbspace(&so->so_snd); - if (flags & MSG_OOB) + + if (flags & MSG_OOB) { space += 1024; + } if ((atomic && resid > so->so_snd.sb_hiwat) || - clen > so->so_snd.sb_hiwat) + clen > so->so_snd.sb_hiwat) { return EMSGSIZE; - if (space < resid + clen && - (atomic || space < so->so_snd.sb_lowat || space < clen)) { - if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO) || assumelock) { + } + + 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))) { + /* + * don't block the connectx call when there's more data + * than can be copied. + */ + if (so->so_flags1 & SOF1_PRECONNECT_DATA) { + if (space == 0) { + return EWOULDBLOCK; + } + if (space < (int32_t)so->so_snd.sb_lowat) { + return 0; + } + } + 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; } @@ -1130,44 +2080,106 @@ restart: * Returns nonzero on error, timeout or signal; callers * must check for short counts if EINTR/ERESTART are returned. * Data and control buffers are freed on return. - * Experiment: - * MSG_HOLD: go thru most of sosend(), but just enqueue the mbuf - * MSG_SEND: go thru as for MSG_HOLD on current fragment, then - * point at the mbuf chain being constructed and go from there. + * + * Returns: 0 Success + * EOPNOTSUPP + * EINVAL + * ENOBUFS + * uiomove:EFAULT + * sosendcheck:EPIPE + * sosendcheck:EWOULDBLOCK + * sosendcheck:EINTR + * sosendcheck:EBADF + * sosendcheck:EINTR + * sosendcheck:??? [value from so_error] + * :ECONNRESET[TCP] + * :EINVAL[TCP] + * :ENOBUFS[TCP] + * :EADDRINUSE[TCP] + * :EADDRNOTAVAIL[TCP] + * :EAFNOSUPPORT[TCP] + * :EACCES[TCP] + * :EAGAIN[TCP] + * :EPERM[TCP] + * :EMSGSIZE[TCP] + * :EHOSTUNREACH[TCP] + * :ENETUNREACH[TCP] + * :ENETDOWN[TCP] + * :ENOMEM[TCP] + * :ENOBUFS[TCP] + * :???[TCP] [ignorable: mostly IPSEC/firewall/DLIL] + * :EINVAL[AF_UNIX] + * :EOPNOTSUPP[AF_UNIX] + * :EPIPE[AF_UNIX] + * :ENOTCONN[AF_UNIX] + * :EISCONN[AF_UNIX] + * :???[AF_UNIX] [whatever a filter author chooses] + * :??? [whatever a filter author chooses] + * + * Notes: Other returns depend on the protocol family; all + * returns depend on what the filter author causes + * their filter to return. */ int -sosend(so, addr, uio, top, control, flags) - register struct socket *so; - struct sockaddr *addr; - struct uio *uio; - struct mbuf *top; - struct mbuf *control; - int flags; - +sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, + struct mbuf *top, struct mbuf *control, int flags) { struct mbuf **mp; - register struct mbuf *m, *freelist = NULL; - register long space, len, resid; + struct mbuf *m, *freelist = NULL; + user_ssize_t space, len, resid, orig_resid; int clen = 0, error, dontroute, mlen, sendflags; int atomic = sosendallatonce(so) || top; int sblocked = 0; struct proc *p = current_proc(); + uint16_t headroom = 0; + boolean_t en_tracing = FALSE; - 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); + 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); + /* + * trace if tracing & network (vs. unix) sockets & and + * non-loopback + */ + if (ENTR_SHOULDTRACE && + (SOCK_CHECK_DOM(so, AF_INET) || SOCK_CHECK_DOM(so, AF_INET6))) { + struct inpcb *inp = sotoinpcb(so); + if (inp->inp_last_outifp != NULL && + !(inp->inp_last_outifp->if_flags & IFF_LOOPBACK)) { + en_tracing = TRUE; + KERNEL_ENERGYTRACE(kEnTrActKernSockWrite, DBG_FUNC_START, + VM_KERNEL_ADDRPERM(so), + ((so->so_state & SS_NBIO) ? kEnTrFlagNonBlocking : 0), + (int64_t)resid); + orig_resid = resid; + } + } + + /* + * 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; + goto out_locked; + } + /* * In theory resid should be unsigned. * However, space must be signed, as it might be less than 0 @@ -1175,1773 +2187,5833 @@ sosend(so, addr, uio, top, control, flags) * 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. + * + * 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 && (flags & MSG_EOR))) { error = EINVAL; - socket_unlock(so, 1); - goto out; + goto out_locked; } - dontroute = - (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && + dontroute = (flags & MSG_DONTROUTE) && + (so->so_options & SO_DONTROUTE) == 0 && (so->so_proto->pr_flags & PR_ATOMIC); - if (p) - p->p_stats->p_ru.ru_msgsnd++; - if (control) + OSIncrementAtomicLong(&p->p_stats->p_ru.ru_msgsnd); + + if (control != NULL) { clen = control->m_len; + } + + if (soreserveheadroom != 0) { + headroom = so->so_pktheadroom; + } do { - error = sosendcheck(so, addr, resid, clen, atomic, flags, &sblocked); + error = sosendcheck(so, addr, resid, clen, atomic, flags, + &sblocked); if (error) { - goto release; + goto out_locked; } + mp = ⊤ - space = sbspace(&so->so_snd) - clen + ((flags & MSG_OOB) ? 1024 : 0); + space = sbspace(&so->so_snd) - clen; + space += ((flags & MSG_OOB) ? 1024 : 0); do { - - if (uio == NULL) { + if (uio == NULL) { /* * Data is prepackaged in "top". */ resid = 0; - if (flags & MSG_EOR) + if (flags & MSG_EOR) { top->m_flags |= M_EOR; + } } else { - int chainlength; - int bytes_to_copy; - - bytes_to_copy = min(resid, space); - + int chainlength; + int bytes_to_copy; + boolean_t jumbocl; + boolean_t bigcl; + int bytes_to_alloc; + + bytes_to_copy = imin(resid, space); + + bytes_to_alloc = bytes_to_copy; + if (top == NULL) { + bytes_to_alloc += headroom; + } + if (sosendminchain > 0) { chainlength = 0; - } else + } else { chainlength = sosendmaxchain; - + } + + /* + * Use big 4 KB cluster when the outgoing interface + * does not prefer 2 KB 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 + * a jumbo cluster pool and if the socket is + * marked accordingly. + */ + jumbocl = sosendjcl && njcl > 0 && + ((so->so_flags & SOF_MULTIPAGES) || + 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 clusters needed to complete this write - * the list is further limited to the number that are currently needed to fill the socket - * this mechanism allows a large number of mbufs/clusters to be grabbed under a single - * mbuf lock... if we can't get any clusters, than fall back to trying for mbufs - * if we fail early (or miscalcluate the number needed) make sure to release any clusters - * we haven't yet consumed. + * try to maintain a local cache of mbuf + * clusters needed to complete this + * write the list is further limited to + * the number that are currently needed + * to fill the socket this mechanism + * allows a large number of mbufs/ + * clusters to be grabbed under a single + * mbuf lock... if we can't get any + * clusters, than fall back to trying + * for mbufs if we fail early (or + * miscalcluate the number needed) make + * sure to release any clusters we + * haven't yet consumed. */ - if (freelist == NULL && bytes_to_copy > MCLBYTES) { - num_needed = bytes_to_copy / NBPG; + if (freelist == NULL && + bytes_to_alloc > MBIGCLBYTES && + jumbocl) { + num_needed = + bytes_to_alloc / M16KCLBYTES; + + if ((bytes_to_alloc - + (num_needed * M16KCLBYTES)) + >= MINCLSIZE) { + num_needed++; + } + + freelist = + m_getpackets_internal( + (unsigned int *)&num_needed, + hdrs_needed, M_WAIT, 0, + M16KCLBYTES); + /* + * Fall back to 4K cluster size + * if allocation failed + */ + } + + if (freelist == NULL && + bytes_to_alloc > MCLBYTES && + bigcl) { + num_needed = + bytes_to_alloc / MBIGCLBYTES; - if ((bytes_to_copy - (num_needed * NBPG)) >= MINCLSIZE) + if ((bytes_to_alloc - + (num_needed * MBIGCLBYTES)) >= + MINCLSIZE) { num_needed++; - - freelist = m_getpackets_internal(&num_needed, hdrs_needed, M_WAIT, 0, NBPG); - /* Fall back to cluster size if allocation failed */ + } + + freelist = + m_getpackets_internal( + (unsigned int *)&num_needed, + hdrs_needed, M_WAIT, 0, + MBIGCLBYTES); + /* + * Fall back to cluster size + * if allocation failed + */ } - - if (freelist == NULL && bytes_to_copy > MINCLSIZE) { - num_needed = bytes_to_copy / MCLBYTES; - - if ((bytes_to_copy - (num_needed * MCLBYTES)) >= MINCLSIZE) + + /* + * Allocate a cluster as we want to + * avoid to split the data in more + * that one segment and using MINCLSIZE + * would lead us to allocate two mbufs + */ + if (soreserveheadroom != 0 && + freelist == NULL && + ((top == NULL && + bytes_to_alloc > _MHLEN) || + bytes_to_alloc > _MLEN)) { + num_needed = ROUNDUP(bytes_to_alloc, MCLBYTES) / + MCLBYTES; + freelist = + m_getpackets_internal( + (unsigned int *)&num_needed, + hdrs_needed, M_WAIT, 0, + MCLBYTES); + /* + * Fall back to a single mbuf + * if allocation failed + */ + } else if (freelist == NULL && + bytes_to_alloc > MINCLSIZE) { + num_needed = + bytes_to_alloc / MCLBYTES; + + if ((bytes_to_alloc - + (num_needed * MCLBYTES)) >= + MINCLSIZE) { num_needed++; - - freelist = m_getpackets_internal(&num_needed, hdrs_needed, M_WAIT, 0, MCLBYTES); - /* Fall back to a single mbuf if allocation failed */ + } + + freelist = + m_getpackets_internal( + (unsigned int *)&num_needed, + hdrs_needed, M_WAIT, 0, + MCLBYTES); + /* + * Fall back to a single mbuf + * if allocation failed + */ + } + /* + * For datagram protocols, leave + * headroom for protocol headers + * in the first cluster of the chain + */ + if (freelist != NULL && atomic && + top == NULL && headroom > 0) { + freelist->m_data += headroom; } - + + /* + * Fall back to regular mbufs without + * reserving the socket headroom + */ if (freelist == NULL) { - if (top == 0) - MGETHDR(freelist, M_WAIT, MT_DATA); - else - MGET(freelist, M_WAIT, MT_DATA); + if (SOCK_TYPE(so) != SOCK_STREAM || bytes_to_alloc <= MINCLSIZE) { + if (top == NULL) { + MGETHDR(freelist, + M_WAIT, MT_DATA); + } else { + MGET(freelist, + M_WAIT, MT_DATA); + } + } if (freelist == NULL) { error = ENOBUFS; socket_lock(so, 0); - goto release; + goto out_locked; } /* - * For datagram protocols, leave room - * for protocol headers in first mbuf. + * For datagram protocols, + * leave room for protocol + * headers in first mbuf. */ - if (atomic && top == 0 && bytes_to_copy < MHLEN) - MH_ALIGN(freelist, bytes_to_copy); + if (atomic && top == NULL && + bytes_to_copy < MHLEN) { + MH_ALIGN(freelist, + bytes_to_copy); + } } m = freelist; freelist = m->m_next; m->m_next = NULL; - - if ((m->m_flags & M_EXT)) - mlen = m->m_ext.ext_size; - else if ((m->m_flags & M_PKTHDR)) - mlen = MHLEN - m_leadingspace(m); - else - mlen = MLEN; - len = min(mlen, bytes_to_copy); + + if ((m->m_flags & M_EXT)) { + mlen = m->m_ext.ext_size - + M_LEADINGSPACE(m); + } else if ((m->m_flags & M_PKTHDR)) { + mlen = + MHLEN - M_LEADINGSPACE(m); + } else { + mlen = MLEN - M_LEADINGSPACE(m); + } + len = imin(mlen, bytes_to_copy); chainlength += len; - + space -= len; - error = uiomove(mtod(m, caddr_t), (int)len, uio); - - // LP64todo - fix this! + error = uiomove(mtod(m, caddr_t), + len, uio); + resid = uio_resid(uio); - + m->m_len = len; *mp = m; top->m_pkthdr.len += len; - if (error) + if (error) { break; + } mp = &m->m_next; if (resid <= 0) { - if (flags & MSG_EOR) + if (flags & MSG_EOR) { top->m_flags |= M_EOR; + } break; } bytes_to_copy = min(resid, space); - - } while (space > 0 && (chainlength < sosendmaxchain || atomic || resid < MINCLSIZE)); - + } while (space > 0 && + (chainlength < sosendmaxchain || atomic || + resid < MINCLSIZE)); + socket_lock(so, 0); - - if (error) - goto release; - } - - if (flags & (MSG_HOLD|MSG_SEND)) - { - /* Enqueue for later, go away if HOLD */ - register struct mbuf *mb1; - if (so->so_temp && (flags & MSG_FLUSH)) - { - m_freem(so->so_temp); - so->so_temp = NULL; - } - if (so->so_temp) - so->so_tail->m_next = top; - else - so->so_temp = top; - mb1 = top; - while (mb1->m_next) - mb1 = mb1->m_next; - so->so_tail = mb1; - if (flags & MSG_HOLD) - { - top = NULL; - goto release; + + if (error) { + goto out_locked; } - top = so->so_temp; - } - if (dontroute) - so->so_options |= SO_DONTROUTE; - /* Compute flags here, for pru_send and NKEs */ - sendflags = (flags & MSG_OOB) ? PRUS_OOB : + } + + if (dontroute) { + so->so_options |= SO_DONTROUTE; + } + /* + * Compute flags here, for pru_send and NKEs + * * If the user set MSG_EOF, the protocol * understands this flag and nothing left to * send then use PRU_SEND_EOF instead of PRU_SEND. */ - ((flags & MSG_EOF) && - (so->so_proto->pr_flags & PR_IMPLOPCL) && - (resid <= 0)) ? - PRUS_EOF : - /* If there is more to send set PRUS_MORETOCOME */ - (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; - - /* - * Socket filter processing - */ - { - struct socket_filter_entry *filter; - int filtered; - - 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; - } - error = filter->sfe_filter->sf_filter.sf_data_out( - filter->sfe_cookie, so, addr, &top, &control, so_flags); + sendflags = (flags & MSG_OOB) ? PRUS_OOB : + ((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; + + 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; } + goto out_locked; } - - if (filtered) { - /* - * At this point, we've run at least one filter. - * The socket is unlocked as is the socket buffer. - */ - socket_lock(so, 0); - sflt_unuse(so); - so->so_send_filt_thread = 0; - if (error) { - if (error == EJUSTRETURN) { - error = 0; - clen = 0; - control = 0; - top = 0; - } - - goto release; +#if CONTENT_FILTER + /* + * Content filter processing + */ + error = cfil_sock_data_out(so, addr, top, + control, sendflags); + if (error) { + if (error == EJUSTRETURN) { + error = 0; + clen = 0; + control = NULL; + top = NULL; } + goto out_locked; } +#endif /* CONTENT_FILTER */ } - /* - * End Socket filter processing - */ - - if (error == EJUSTRETURN) { - /* A socket filter handled this data */ - error = 0; + error = (*so->so_proto->pr_usrreqs->pru_send) + (so, sendflags, top, addr, control, p); + + if (dontroute) { + so->so_options &= ~SO_DONTROUTE; } - else { - error = (*so->so_proto->pr_usrreqs->pru_send)(so, - sendflags, top, addr, control, p); + + clen = 0; + control = NULL; + top = NULL; + mp = ⊤ + if (error) { + goto out_locked; } -#ifdef __APPLE__ - if (flags & MSG_SEND) - so->so_temp = NULL; -#endif - if (dontroute) - so->so_options &= ~SO_DONTROUTE; - clen = 0; - control = 0; - top = 0; - mp = ⊤ - if (error) - goto release; } while (resid && space > 0); } while (resid); -release: - if (sblocked) - sbunlock(&so->so_snd, 0); /* will unlock socket */ - else +out_locked: + if (sblocked) { + sbunlock(&so->so_snd, FALSE); /* will unlock socket */ + } else { socket_unlock(so, 1); -out: - if (top) + } + if (top != NULL) { m_freem(top); - if (control) + } + if (control != NULL) { m_freem(control); - if (freelist) - m_freem_list(freelist); + } + if (freelist != NULL) { + m_freem_list(freelist); + } + + soclearfastopen(so); + + if (en_tracing) { + /* resid passed here is the bytes left in uio */ + KERNEL_ENERGYTRACE(kEnTrActKernSockWrite, DBG_FUNC_END, + VM_KERNEL_ADDRPERM(so), + ((error == EWOULDBLOCK) ? kEnTrFlagNoWork : 0), + (int64_t)(orig_resid - resid)); + } + KERNEL_DEBUG(DBG_FNC_SOSEND | DBG_FUNC_END, so, resid, + so->so_snd.sb_cc, space, error); + + return error; +} + +int +sosend_reinject(struct socket *so, struct sockaddr *addr, struct mbuf *top, struct mbuf *control, uint32_t sendflags) +{ + struct mbuf *m0 = NULL, *control_end = NULL; + + socket_lock_assert_owned(so); + + /* + * top must points to mbuf chain to be sent. + * If control is not NULL, top must be packet header + */ + VERIFY(top != NULL && + (control == NULL || top->m_flags & M_PKTHDR)); - KERNEL_DEBUG(DBG_FNC_SOSEND | DBG_FUNC_END, - so, - resid, - so->so_snd.sb_cc, - space, - error); + /* + * If control is not passed in, see if we can get it + * from top. + */ + if (control == NULL && (top->m_flags & M_PKTHDR) == 0) { + // Locate start of control if present and start of data + for (m0 = top; m0 != NULL; m0 = m0->m_next) { + if (m0->m_flags & M_PKTHDR) { + top = m0; + break; + } else if (m0->m_type == MT_CONTROL) { + if (control == NULL) { + // Found start of control + control = m0; + } + if (control != NULL && m0->m_next != NULL && m0->m_next->m_type != MT_CONTROL) { + // Found end of control + control_end = m0; + } + } + } + if (control_end != NULL) { + control_end->m_next = NULL; + } + } + + int error = (*so->so_proto->pr_usrreqs->pru_send) + (so, sendflags, top, addr, control, current_proc()); - return (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. + * Supported only connected sockets (no address) without ancillary data + * (control mbuf) for atomic protocols */ int -soreceive(so, psa, uio, mp0, controlp, flagsp) - register struct socket *so; - struct sockaddr **psa; - struct uio *uio; - struct mbuf **mp0; - struct mbuf **controlp; - int *flagsp; -{ - register struct mbuf *m, **mp, *ml = NULL; - register int flags, len, error, offset; - struct protosw *pr = so->so_proto; - struct mbuf *nextrecord; - int moff, type = 0; - // LP64todo - fix this! - int orig_resid = uio_resid(uio); - volatile struct mbuf *free_list; - volatile int delayed_copy_len; - int can_delay; - int need_event; +sosend_list(struct socket *so, struct uio **uioarray, u_int uiocnt, int flags) +{ + struct mbuf *m, *freelist = NULL; + user_ssize_t len, resid; + int error, dontroute, mlen; + int atomic = sosendallatonce(so); + int sblocked = 0; struct proc *p = current_proc(); + u_int uiofirst = 0; + u_int uiolast = 0; + struct mbuf *top = NULL; + uint16_t headroom = 0; + boolean_t bigcl; + + 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); + 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; + } + resid = uio_array_resid(uioarray, uiocnt); - // 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); + /* + * 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 (resid < 0 || resid > INT_MAX) { + error = EINVAL; + goto out; + } 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("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; - else - flags = 0; - /* - * When SO_WANTOOBFLAG is set we try to get out-of-band data - * regardless of the flags argument. Here is the case were - * out-of-band data is not inline. - */ - if ((flags & MSG_OOB) || - ((so->so_options & SO_WANTOOBFLAG) != 0 && - (so->so_options & SO_OOBINLINE) == 0 && - (so->so_oobmark || (so->so_state & SS_RCVATMARK)))) { - 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 { - // LP64todo - fix this! - error = uiomove(mtod(m, caddr_t), - (int) min(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; - } - 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 NECP + so_update_necp_policy(so, NULL, NULL); +#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); - free_list = (struct mbuf *)0; - delayed_copy_len = 0; -restart: -#ifdef MORE_LOCKING_DEBUG - if (so->so_usecount <= 1) - printf("soreceive: sblock so=%x ref=%d on socket\n", so, so->so_usecount); -#endif - error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); + error = sosendcheck(so, NULL, resid, 0, atomic, flags, &sblocked); if (error) { - socket_unlock(so, 1); - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error,0,0,0,0); - return (error); + goto release; } - m = so->so_rcv.sb_mb; /* - * If we have less data than requested, block awaiting more - * (subject to any timeout) if: - * 1. the current count is less than the low water mark, or - * 2. MSG_WAITALL is set, and it is possible to do the entire - * receive operation at once if we block (resid <= hiwat). - * 3. MSG_DONTWAIT is not set - * If MSG_WAITALL is set but resid is larger than the receive buffer, - * we have to do the receive in sections, and thus risk returning - * a short count if a timeout or signal occurs after we start. - */ - if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && - so->so_rcv.sb_cc < uio_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)) { - - KASSERT(m != 0 || !so->so_rcv.sb_cc, ("receive 1")); - if (so->so_error) { - if (m) - goto dontblock; - error = so->so_error; - if ((flags & MSG_PEEK) == 0) - so->so_error = 0; - goto release; - } - if (so->so_state & SS_CANTRCVMORE) { - if (m) - goto dontblock; - else - goto release; - } - for (; m; m = m->m_next) - if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { - m = so->so_rcv.sb_mb; - goto dontblock; - } - if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && - (so->so_proto->pr_flags & PR_CONNREQUIRED)) { - error = ENOTCONN; - goto release; - } - if (uio_resid(uio) == 0) - goto release; - if ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO))) { - error = EWOULDBLOCK; - goto release; - } - sbunlock(&so->so_rcv, 1); -#ifdef EVEN_MORE_LOCKING_DEBUG - if (socket_debug) - printf("Waiting for socket data\n"); -#endif + * Use big 4 KB clusters when the outgoing interface does not prefer + * 2 KB clusters + */ + bigcl = !(so->so_flags1 & SOF1_IF_2KCL) || sosendbigcl_ignore_capab; - error = sbwait(&so->so_rcv); -#ifdef EVEN_MORE_LOCKING_DEBUG - if (socket_debug) - printf("SORECEIVE - sbwait returned %d\n", error); -#endif - if (so->so_usecount < 1) - panic("soreceive: after 2nd sblock so=%x ref=%d on socket\n", so, so->so_usecount); - if (error) { - socket_unlock(so, 1); - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error,0,0,0,0); - return (error); - } - goto restart; + if (soreserveheadroom != 0) { + headroom = so->so_pktheadroom; } -dontblock: -#ifndef __APPLE__ - if (uio->uio_procp) - uio->uio_procp->p_stats->p_ru.ru_msgrcv++; -#else /* __APPLE__ */ - /* - * 2207985 - * This should be uio->uio-procp; however, some callers of this - * function use auto variables with stack garbage, and fail to - * fill out the uio structure properly. - */ - if (p) - p->p_stats->p_ru.ru_msgrcv++; -#endif /* __APPLE__ */ - nextrecord = m->m_nextpkt; - if ((pr->pr_flags & PR_ADDR) && m->m_type == MT_SONAME) { - KASSERT(m->m_type == MT_SONAME, ("receive 1a")); - orig_resid = 0; - if (psa) { - *psa = dup_sockaddr(mtod(m, struct sockaddr *), - mp0 == 0); - if ((*psa == 0) && (flags & MSG_NEEDSA)) { - error = EWOULDBLOCK; - goto release; - } - } - if (flags & MSG_PEEK) { - m = m->m_next; + + do { + int i; + int num_needed = 0; + int chainlength; + size_t maxpktlen = 0; + int bytes_to_alloc; + + if (sosendminchain > 0) { + chainlength = 0; } else { - sbfree(&so->so_rcv, m); - if (m->m_next == 0 && so->so_rcv.sb_cc != 0) - panic("soreceive: about to create invalid socketbuf"); - MFREE(m, so->so_rcv.sb_mb); - m = so->so_rcv.sb_mb; + chainlength = sosendmaxchain; } - } - while (m && m->m_type == MT_CONTROL && error == 0) { - if (flags & MSG_PEEK) { - if (controlp) - *controlp = m_copy(m, 0, m->m_len); - m = m->m_next; - } else { - sbfree(&so->so_rcv, m); - if (controlp) { - if (pr->pr_domain->dom_externalize && - mtod(m, struct cmsghdr *)->cmsg_type == - SCM_RIGHTS) { - socket_unlock(so, 0); /* release socket lock: see 3903171 */ - error = (*pr->pr_domain->dom_externalize)(m); - socket_lock(so, 0); - } - *controlp = m; - if (m->m_next == 0 && so->so_rcv.sb_cc != 0) - panic("soreceive: so->so_rcv.sb_mb->m_next == 0 && so->so_rcv.sb_cc != 0"); - so->so_rcv.sb_mb = m->m_next; - m->m_next = 0; - m = so->so_rcv.sb_mb; - } else { - MFREE(m, so->so_rcv.sb_mb); - m = so->so_rcv.sb_mb; + + 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; } - } - if (controlp) { - orig_resid = 0; - controlp = &(*controlp)->m_next; - } - } - if (m) { - if ((flags & MSG_PEEK) == 0) - m->m_nextpkt = nextrecord; - type = m->m_type; - if (type == MT_OOBDATA) - flags |= MSG_OOB; - } - moff = 0; - offset = 0; - if (!(flags & MSG_PEEK) && uio_resid(uio) > sorecvmincopy) - can_delay = 1; - else - can_delay = 0; + num_needed += 1; + uiolast += 1; - need_event = 0; + if (len > maxpktlen) { + maxpktlen = len; + } - while (m && (uio_resid(uio) - delayed_copy_len) > 0 && error == 0) { - if (m->m_type == MT_OOBDATA) { - if (type != MT_OOBDATA) + chainlength += len; + if (chainlength > sosendmaxchain) { break; - } else if (type == MT_OOBDATA) + } + } + /* + * Nothing left to send + */ + if (num_needed == 0) { + socket_lock(so, 0); break; -#ifndef __APPLE__ -/* - * This assertion needs rework. The trouble is Appletalk is uses many - * mbuf types (NOT listed in mbuf.h!) which will trigger this panic. - * For now just remove the assertion... CSM 9/98 - */ - else - KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, - ("receive 3")); -#else + } /* - * Make sure to allways set MSG_OOB event when getting - * out of band data inline. + * Allocate buffer large enough to include headroom space for + * network and link header + * */ - if ((so->so_options & SO_WANTOOBFLAG) != 0 && - (so->so_options & SO_OOBINLINE) != 0 && - (so->so_state & SS_RCVATMARK) != 0) { - flags |= MSG_OOB; + bytes_to_alloc = maxpktlen + headroom; + + /* + * Allocate a single contiguous buffer of the smallest available + * size when possible + */ + if (bytes_to_alloc > MCLBYTES && + bytes_to_alloc <= MBIGCLBYTES && bigcl) { + freelist = m_getpackets_internal( + (unsigned int *)&num_needed, + num_needed, M_WAIT, 1, + MBIGCLBYTES); + } else if (bytes_to_alloc > _MHLEN && + bytes_to_alloc <= MCLBYTES) { + freelist = m_getpackets_internal( + (unsigned int *)&num_needed, + num_needed, M_WAIT, 1, + MCLBYTES); + } else { + freelist = m_allocpacket_internal( + (unsigned int *)&num_needed, + bytes_to_alloc, NULL, M_WAIT, 1, 0); + } + + if (freelist == NULL) { + socket_lock(so, 0); + error = ENOMEM; + goto release; } -#endif - so->so_state &= ~SS_RCVATMARK; - // LP64todo - fix this! - len = uio_resid(uio) - delayed_copy_len; - if (so->so_oobmark && len > so->so_oobmark - offset) - len = so->so_oobmark - offset; - if (len > m->m_len - moff) - len = m->m_len - moff; /* - * 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. + * Copy each uio of the set into its own mbuf packet */ - if (mp == 0) { - if (can_delay && len == m->m_len) { - /* - * only delay the copy if we're consuming the - * mbuf and we're NOT in MSG_PEEK mode - * and we have enough data to make it worthwile - * to drop and retake the funnel... can_delay - * reflects the state of the 2 latter constraints - * moff should always be zero in these cases + 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; + } + /* + * Leave headroom for protocol headers + * in the first mbuf of the chain + */ + m->m_data += headroom; + + for (n = m; n != NULL; n = n->m_next) { + if ((m->m_flags & M_EXT)) { + mlen = m->m_ext.ext_size - + M_LEADINGSPACE(m); + } else if ((m->m_flags & M_PKTHDR)) { + mlen = + MHLEN - M_LEADINGSPACE(m); + } else { + mlen = MLEN - M_LEADINGSPACE(m); + } + len = imin(mlen, bytes_to_copy); + + /* + * Note: uiomove() decrements the iovec + * length */ - delayed_copy_len += len; - } else { + error = uiomove(mtod(n, caddr_t), + len, auio); + if (error != 0) { + break; + } + n->m_len = len; + m->m_pkthdr.len += len; - if (delayed_copy_len) { - error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len); + VERIFY(m->m_pkthdr.len <= maxpktlen); - if (error) { + bytes_to_copy -= len; + resid -= len; + } + if (m->m_pkthdr.len == 0) { + printf( + "%s:%d so %llx pkt %llx type %u len null\n", + __func__, __LINE__, + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + (uint64_t)DEBUG_KERNEL_ADDRPERM(m), + m->m_type); + } + 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, NULL, &m, + NULL, 0); + if (error != 0 && error != EJUSTRETURN) { + goto release; + } + +#if CONTENT_FILTER + if (error == 0) { + /* + * Content filter processing + */ + error = cfil_sock_data_out(so, NULL, m, + NULL, 0); + if (error != 0 && error != EJUSTRETURN) { goto release; } - if (m != so->so_rcv.sb_mb) { - /* - * can only get here if MSG_PEEK is not set - * therefore, m should point at the head of the rcv queue... - * if it doesn't, it means something drastically changed - * while we were out from behind the funnel in sodelayed_copy... - * perhaps a RST on the stream... in any event, the stream has - * been interrupted... it's probably best just to return - * whatever data we've moved and let the caller sort it out... - */ - break; + } +#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; } } - socket_unlock(so, 0); - error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); - socket_lock(so, 0); - if (error) - goto release; + m = nextpkt; + if (m != NULL) { + prevnextp = &m->m_nextpkt; + } } - } else - uio_setresid(uio, (uio_resid(uio) - len)); + } + if (top != NULL) { + error = (*so->so_proto->pr_usrreqs->pru_send_list) + (so, 0, top, NULL, NULL, p); + } - if (len == m->m_len - moff) { - if (m->m_flags & M_EOR) - flags |= MSG_EOR; - if (flags & MSG_PEEK) { - m = m->m_next; - moff = 0; - } else { - nextrecord = m->m_nextpkt; - sbfree(&so->so_rcv, m); - m->m_nextpkt = NULL; + if (dontroute) { + so->so_options &= ~SO_DONTROUTE; + } - if (mp) { - *mp = m; - mp = &m->m_next; - so->so_rcv.sb_mb = m = m->m_next; - *mp = (struct mbuf *)0; - } 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; - } - if (m) - m->m_nextpkt = nextrecord; - } + 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 (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; +} + +/* + * May return ERESTART when packet is dropped by MAC policy check + */ +static int +soreceive_addr(struct proc *p, struct socket *so, struct sockaddr **psa, + int flags, struct mbuf **mp, struct mbuf **nextrecordp, int canwait) +{ + int error = 0; + struct mbuf *m = *mp; + struct mbuf *nextrecord = *nextrecordp; + + KASSERT(m->m_type == MT_SONAME, ("receive 1a")); +#if CONFIG_MACF_SOCKET_SUBSET + /* + * Call the MAC framework for policy checking if we're in + * the user process context and the socket isn't connected. + */ + if (p != kernproc && !(so->so_state & SS_ISCONNECTED)) { + struct mbuf *m0 = m; + /* + * Dequeue this record (temporarily) from the receive + * list since we're about to drop the socket's lock + * where a new record may arrive and be appended to + * the list. Upon MAC policy failure, the record + * will be freed. Otherwise, we'll add it back to + * the head of the list. We cannot rely on SB_LOCK + * because append operation uses the socket's lock. + */ + do { + m->m_nextpkt = NULL; + sbfree(&so->so_rcv, m); + m = m->m_next; + } while (m != NULL); + m = m0; + so->so_rcv.sb_mb = nextrecord; + SB_EMPTY_FIXUP(&so->so_rcv); + SBLASTRECORDCHK(&so->so_rcv, "soreceive 1a"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 1a"); + socket_unlock(so, 0); + + if (mac_socket_check_received(proc_ucred(p), so, + mtod(m, struct sockaddr *)) != 0) { + /* + * MAC policy failure; free this record and + * process the next record (or block until + * one is available). We have adjusted sb_cc + * and sb_mbcnt above so there is no need to + * call sbfree() again. + */ + m_freem(m); + /* + * Clear SB_LOCK but don't unlock the socket. + * Process the next record or wait for one. + */ + socket_lock(so, 0); + sbunlock(&so->so_rcv, TRUE); /* stay locked */ + error = ERESTART; + goto done; + } + 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 done; + } + /* + * Re-adjust the socket receive list and re-enqueue + * the record in front of any packets which may have + * been appended while we dropped the lock. + */ + for (m = m0; m->m_next != NULL; m = m->m_next) { + sballoc(&so->so_rcv, m); + } + sballoc(&so->so_rcv, m); + if (so->so_rcv.sb_mb == NULL) { + so->so_rcv.sb_lastrecord = m0; + so->so_rcv.sb_mbtail = m; + } + m = m0; + nextrecord = m->m_nextpkt = so->so_rcv.sb_mb; + so->so_rcv.sb_mb = m; + SBLASTRECORDCHK(&so->so_rcv, "soreceive 1b"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 1b"); + } +#endif /* CONFIG_MACF_SOCKET_SUBSET */ + if (psa != NULL) { + *psa = dup_sockaddr(mtod(m, struct sockaddr *), canwait); + if ((*psa == NULL) && (flags & MSG_NEEDSA)) { + error = EWOULDBLOCK; + goto done; + } + } + if (flags & MSG_PEEK) { + m = m->m_next; + } else { + sbfree(&so->so_rcv, m); + if (m->m_next == NULL && so->so_rcv.sb_cc != 0) { + panic("%s: about to create invalid socketbuf", + __func__); + /* NOTREACHED */ + } + MFREE(m, so->so_rcv.sb_mb); + m = so->so_rcv.sb_mb; + if (m != NULL) { + m->m_nextpkt = nextrecord; } else { - if (flags & MSG_PEEK) - moff += len; - else { - if (mp) - *mp = m_copym(m, 0, len, M_WAIT); - m->m_data += len; - m->m_len -= len; - so->so_rcv.sb_cc -= len; - } + so->so_rcv.sb_mb = nextrecord; + SB_EMPTY_FIXUP(&so->so_rcv); } - if (so->so_oobmark) { - if ((flags & MSG_PEEK) == 0) { - so->so_oobmark -= len; - if (so->so_oobmark == 0) { - so->so_state |= SS_RCVATMARK; - /* - * delay posting the actual event until after - * any delayed copy processing has finished - */ - need_event = 1; - break; + } +done: + *mp = m; + *nextrecordp = nextrecord; + + return error; +} + +/* + * 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. + */ +static int +soreceive_ctl(struct socket *so, struct mbuf **controlp, int flags, + struct mbuf **mp, struct mbuf **nextrecordp) +{ + int error = 0; + struct mbuf *cm = NULL, *cmn; + struct mbuf **cme = &cm; + struct sockbuf *sb_rcv = &so->so_rcv; + struct mbuf **msgpcm = NULL; + struct mbuf *m = *mp; + struct mbuf *nextrecord = *nextrecordp; + struct protosw *pr = so->so_proto; + + /* + * Externalizing the control messages would require us to + * drop the socket's lock below. Once we re-acquire the + * lock, the mbuf chain might change. In order to preserve + * consistency, we unlink all control messages from the + * first mbuf chain in one shot and link them separately + * onto a different chain. + */ + do { + if (flags & MSG_PEEK) { + if (controlp != NULL) { + if (*controlp == NULL) { + msgpcm = controlp; } - } else { - offset += len; - if (offset == so->so_oobmark) - break; + *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 done; + } + controlp = &(*controlp)->m_next; } + m = m->m_next; + } else { + m->m_nextpkt = NULL; + sbfree(sb_rcv, m); + sb_rcv->sb_mb = m->m_next; + m->m_next = NULL; + *cme = m; + cme = &(*cme)->m_next; + m = sb_rcv->sb_mb; } - if (flags & MSG_EOR) - break; + } while (m != NULL && m->m_type == MT_CONTROL); + + if (!(flags & MSG_PEEK)) { + if (sb_rcv->sb_mb != NULL) { + sb_rcv->sb_mb->m_nextpkt = nextrecord; + } else { + sb_rcv->sb_mb = nextrecord; + SB_EMPTY_FIXUP(sb_rcv); + } + if (nextrecord == NULL) { + sb_rcv->sb_lastrecord = m; + } + } + + SBLASTRECORDCHK(&so->so_rcv, "soreceive ctl"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive ctl"); + + while (cm != NULL) { + int cmsg_type; + + cmn = cm->m_next; + cm->m_next = NULL; + cmsg_type = mtod(cm, struct cmsghdr *)->cmsg_type; + /* - * If the MSG_WAITALL or MSG_WAITSTREAM flag is set (for non-atomic socket), - * we must not quit until "uio->uio_resid == 0" or an error - * termination. If a signal/timeout occurs, return - * with a short count but without error. - * Keep sockbuf locked against other readers. + * Call the protocol to externalize SCM_RIGHTS message + * and return the modified message to the caller upon + * success. Otherwise, all other control messages are + * returned unmodified to the caller. Note that we + * only get into this loop if MSG_PEEK is not set. */ - while (flags & (MSG_WAITALL|MSG_WAITSTREAM) && m == 0 && (uio_resid(uio) - delayed_copy_len) > 0 && - !sosendallatonce(so) && !nextrecord) { - if (so->so_error || so->so_state & SS_CANTRCVMORE) - goto release; - - if (pr->pr_flags & PR_WANTRCVD && so->so_pcb && (((struct inpcb *)so->so_pcb)->inp_state != INPCB_STATE_DEAD)) - (*pr->pr_usrreqs->pru_rcvd)(so, flags); - if (sbwait(&so->so_rcv)) { - error = 0; - goto release; - } + if (pr->pr_domain->dom_externalize != NULL && + cmsg_type == SCM_RIGHTS) { /* - * have to wait until after we get back from the sbwait to do the copy because - * we will drop the funnel if we have enough data that has been delayed... by dropping - * the funnel we open up a window allowing the netisr thread to process the incoming packets - * and to change the state of this socket... we're issuing the sbwait because - * the socket is empty and we're expecting the netisr thread to wake us up when more - * packets arrive... if we allow that processing to happen and then sbwait, we - * could stall forever with packets sitting in the socket if no further packets - * arrive from the remote side. - * - * we want to copy before we've collected all the data to satisfy this request to - * allow the copy to overlap the incoming packet processing on an MP system + * Release socket lock: see 3903171. This + * would also allow more records to be appended + * to the socket buffer. We still have SB_LOCK + * set on it, so we can be sure that the head + * of the mbuf chain won't change. */ - if (delayed_copy_len > sorecvmincopy && (delayed_copy_len > (so->so_rcv.sb_hiwat / 2))) { - - error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len); + socket_unlock(so, 0); + error = (*pr->pr_domain->dom_externalize)(cm); + socket_lock(so, 0); + } else { + error = 0; + } - if (error) - goto release; - } - m = so->so_rcv.sb_mb; - if (m) { - nextrecord = m->m_nextpkt; - } + if (controlp != NULL && error == 0) { + *controlp = cm; + controlp = &(*controlp)->m_next; + } else { + (void) m_free(cm); } + cm = cmn; + } + /* + * Update the value of nextrecord in case we received new + * records when the socket was unlocked above for + * externalizing SCM_RIGHTS. + */ + if (m != NULL) { + nextrecord = sb_rcv->sb_mb->m_nextpkt; + } else { + nextrecord = sb_rcv->sb_mb; } -#ifdef MORE_LOCKING_DEBUG - if (so->so_usecount <= 1) - panic("soreceive: after big while so=%x ref=%d on socket\n", so, so->so_usecount); -#endif - if (m && pr->pr_flags & PR_ATOMIC) { -#ifdef __APPLE__ - 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__ +done: + *mp = m; + *nextrecordp = nextrecord; + + return error; +} + +/* + * If we have less data than requested, block awaiting more + * (subject to any timeout) if: + * 1. the current count is less than the low water mark, or + * 2. MSG_WAITALL is set, and it is possible to do the entire + * receive operation at once if we block (resid <= hiwat). + * 3. MSG_DONTWAIT is not set + * If MSG_WAITALL is set but resid is larger than the receive buffer, + * we have to do the receive in sections, and thus risk returning + * a short count if a timeout or signal occurs after we start. + */ +static boolean_t +so_should_wait(struct socket *so, struct uio *uio, struct mbuf *m, int flags) +{ + struct protosw *pr = so->so_proto; + + /* No mbufs in the receive-queue? Wait! */ + if (m == NULL) { + return true; + } + + /* Not enough data in the receive socket-buffer - we may have to wait */ + if ((flags & MSG_DONTWAIT) == 0 && so->so_rcv.sb_cc < uio_resid(uio) && + m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0) { + /* + * Application did set the lowater-mark, so we should wait for + * this data to be present. + */ + if (so->so_rcv.sb_cc < so->so_rcv.sb_lowat) { + return true; + } + + /* + * Application wants all the data - so let's try to do the + * receive-operation at once by waiting for everything to + * be there. + */ + if ((flags & MSG_WAITALL) && uio_resid(uio) <= so->so_rcv.sb_hiwat) { + return true; } + } + + return false; +} + +/* + * 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; + struct proc *p = current_proc(); + boolean_t en_tracing = FALSE; + + /* + * Sanity check on the length passed by caller as we are making 'int' + * comparisons + */ + if (orig_resid < 0 || orig_resid > INT_MAX) { + return EINVAL; + } + + 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); + 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 ((flags & MSG_PEEK) == 0) { - if (m == 0) - so->so_rcv.sb_mb = nextrecord; - if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) - (*pr->pr_usrreqs->pru_rcvd)(so, flags); + if (controlp != NULL) { + *controlp = NULL; + } + if (flagsp != NULL) { + flags = *flagsp & ~MSG_EOR; + } else { + flags = 0; } -#ifdef __APPLE__ - if ((so->so_options & SO_WANTMORE) && so->so_rcv.sb_cc > 0) - flags |= MSG_HAVEMORE; - if (delayed_copy_len) { - error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len); + /* + * 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, %s]: defunct so 0x%llx [%d,%d] (%d)\n", + __func__, proc_pid(p), proc_best_name(p), + (uint64_t)DEBUG_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; + } + + if ((so->so_flags1 & SOF1_PRECONNECT_DATA) && + pr->pr_usrreqs->pru_preconnect) { + /* + * A user may set the CONNECT_RESUME_ON_READ_WRITE-flag but not + * calling write() right after this. *If* the app calls a read + * we do not want to block this read indefinetely. Thus, + * we trigger a connect so that the session gets initiated. + */ + error = (*pr->pr_usrreqs->pru_preconnect)(so); + + if (error) { + socket_unlock(so, 1); + return error; + } + } + + if (ENTR_SHOULDTRACE && + (SOCK_CHECK_DOM(so, AF_INET) || SOCK_CHECK_DOM(so, AF_INET6))) { + /* + * enable energy tracing for inet sockets that go over + * non-loopback interfaces only. + */ + struct inpcb *inp = sotoinpcb(so); + if (inp->inp_last_outifp != NULL && + !(inp->inp_last_outifp->if_flags & IFF_LOOPBACK)) { + en_tracing = TRUE; + KERNEL_ENERGYTRACE(kEnTrActKernSockRead, DBG_FUNC_START, + VM_KERNEL_ADDRPERM(so), + ((so->so_state & SS_NBIO) ? + kEnTrFlagNonBlocking : 0), + (int64_t)orig_resid); + } + } + + /* + * 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); + if (en_tracing) { + KERNEL_ENERGYTRACE(kEnTrActKernSockRead, DBG_FUNC_END, + VM_KERNEL_ADDRPERM(so), 0, + (int64_t)(orig_resid - uio_resid(uio))); + } + KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error, + 0, 0, 0, 0); + + return error; + } +nooob: + if (mp != NULL) { + *mp = NULL; + } + + if (so->so_state & SS_ISCONFIRMING && uio_resid(uio)) { + (*pr->pr_usrreqs->pru_rcvd)(so, 0); + } + + free_list = NULL; + delayed_copy_len = 0; +restart: +#ifdef MORE_LOCKING_DEBUG + if (so->so_usecount <= 1) { + printf("soreceive: sblock so=0x%llx ref=%d on socket\n", + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), so->so_usecount); + } +#endif + /* + * See if the socket has been closed (SS_NOFDREF|SS_CANTRCVMORE) + * and if so just return to the caller. This could happen when + * soreceive() is called by a socket upcall function during the + * time the socket is freed. The socket buffer would have been + * locked across the upcall, therefore we cannot put this thread + * to sleep (else we will deadlock) or return EWOULDBLOCK (else + * we may livelock), because the lock on the socket buffer will + * only be released when the upcall routine returns to its caller. + * Because the socket has been officially closed, there can be + * no further read on it. + * + * A multipath subflow socket would have its SS_NOFDREF set by + * default, so check for SOF_MP_SUBFLOW socket flag; when the + * socket is closed for real, SOF_MP_SUBFLOW would be cleared. + */ + if ((so->so_state & (SS_NOFDREF | SS_CANTRCVMORE)) == + (SS_NOFDREF | SS_CANTRCVMORE) && !(so->so_flags & SOF_MP_SUBFLOW)) { + socket_unlock(so, 1); + return 0; + } + + error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); + if (error) { + socket_unlock(so, 1); + KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error, + 0, 0, 0, 0); + if (en_tracing) { + KERNEL_ENERGYTRACE(kEnTrActKernSockRead, DBG_FUNC_END, + VM_KERNEL_ADDRPERM(so), 0, + (int64_t)(orig_resid - uio_resid(uio))); + } + return error; + } + + m = so->so_rcv.sb_mb; + if (so_should_wait(so, uio, m, flags)) { + /* + * 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) { + if (m != NULL) { + goto dontblock; + } + error = so->so_error; + if ((flags & MSG_PEEK) == 0) { + so->so_error = 0; + } + goto release; + } + if (so->so_state & SS_CANTRCVMORE) { +#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)DEBUG_KERNEL_ADDRPERM(so)); + } else +#endif /* CONTENT_FILTER */ + if (m != NULL) { + goto dontblock; + } else { + goto release; + } + } + 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; + } + } + if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) == 0 && + (so->so_proto->pr_flags & PR_CONNREQUIRED)) { + error = ENOTCONN; + goto release; + } + if (uio_resid(uio) == 0) { + goto release; + } + + if ((so->so_state & SS_NBIO) || + (flags & (MSG_DONTWAIT | MSG_NBIO))) { + error = EWOULDBLOCK; + goto release; + } + SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); + sbunlock(&so->so_rcv, TRUE); /* keep socket locked */ +#if EVEN_MORE_LOCKING_DEBUG + if (socket_debug) { + printf("Waiting for socket data\n"); + } +#endif + + /* + * Depending on the protocol (e.g. TCP), the following + * might cause the socket lock to be dropped and later + * be reacquired, and more data could have arrived and + * have been appended to the receive socket buffer by + * the time it returns. Therefore, we only sleep in + * sbwait() below if and only if the wait-condition is still + * true. + */ + if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb != NULL) { + (*pr->pr_usrreqs->pru_rcvd)(so, flags); + } + + error = 0; + if (so_should_wait(so, uio, so->so_rcv.sb_mb, flags)) { + error = sbwait(&so->so_rcv); + } + +#if EVEN_MORE_LOCKING_DEBUG + if (socket_debug) { + printf("SORECEIVE - sbwait returned %d\n", error); + } +#endif + if (so->so_usecount < 1) { + panic("%s: after 2nd sblock so=%p ref=%d on socket\n", + __func__, so, so->so_usecount); + /* NOTREACHED */ + } + if (error) { + socket_unlock(so, 1); + KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error, + 0, 0, 0, 0); + if (en_tracing) { + KERNEL_ENERGYTRACE(kEnTrActKernSockRead, DBG_FUNC_END, + VM_KERNEL_ADDRPERM(so), 0, + (int64_t)(orig_resid - uio_resid(uio))); + } + return error; + } + goto restart; + } +dontblock: + OSIncrementAtomicLong(&p->p_stats->p_ru.ru_msgrcv); + SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); + nextrecord = m->m_nextpkt; + + if ((pr->pr_flags & PR_ADDR) && m->m_type == MT_SONAME) { + error = soreceive_addr(p, so, psa, flags, &m, &nextrecord, + mp0 == NULL); + if (error == ERESTART) { + goto restart; + } else if (error != 0) { + goto release; + } + orig_resid = 0; + } + + /* + * 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) { + error = soreceive_ctl(so, controlp, flags, &m, &nextrecord); + if (error != 0) { + goto release; + } + orig_resid = 0; + } + + if (m != NULL) { + if (!(flags & MSG_PEEK)) { + /* + * We get here because m points to an mbuf following + * any MT_SONAME or MT_CONTROL mbufs which have been + * processed above. In any case, m should be pointing + * to the head of the mbuf chain, and the nextrecord + * should be either NULL or equal to m->m_nextpkt. + * See comments above about SB_LOCK. + */ + if (m != so->so_rcv.sb_mb || + m->m_nextpkt != nextrecord) { + panic("%s: post-control !sync so=%p m=%p " + "nextrecord=%p\n", __func__, so, m, + nextrecord); + /* NOTREACHED */ + } + if (nextrecord == NULL) { + so->so_rcv.sb_lastrecord = m; + } + } + type = m->m_type; + if (type == MT_OOBDATA) { + flags |= MSG_OOB; + } + } else { + if (!(flags & MSG_PEEK)) { + SB_EMPTY_FIXUP(&so->so_rcv); + } + } + SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); + + moff = 0; + offset = 0; + + if (!(flags & MSG_PEEK) && uio_resid(uio) > sorecvmincopy) { + can_delay = 1; + } else { + can_delay = 0; + } + + while (m != NULL && + (uio_resid(uio) - delayed_copy_len) > 0 && error == 0) { + if (m->m_type == MT_OOBDATA) { + if (type != MT_OOBDATA) { + break; + } + } else if (type == MT_OOBDATA) { + break; + } + /* + * Make sure to allways set MSG_OOB event when getting + * out of band data inline. + */ + if ((so->so_options & SO_WANTOOBFLAG) != 0 && + (so->so_options & SO_OOBINLINE) != 0 && + (so->so_state & SS_RCVATMARK) != 0) { + flags |= MSG_OOB; + } + so->so_state &= ~SS_RCVATMARK; + len = uio_resid(uio) - delayed_copy_len; + if (so->so_oobmark && len > so->so_oobmark - offset) { + len = so->so_oobmark - offset; + } + if (len > m->m_len - moff) { + len = m->m_len - moff; + } + /* + * 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) { + SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); + if (can_delay && len == m->m_len) { + /* + * only delay the copy if we're consuming the + * mbuf and we're NOT in MSG_PEEK mode + * and we have enough data to make it worthwile + * to drop and retake the lock... can_delay + * reflects the state of the 2 latter + * constraints moff should always be zero + * in these cases + */ + delayed_copy_len += len; + } else { + if (delayed_copy_len) { + error = sodelayed_copy(so, uio, + &free_list, &delayed_copy_len); + + if (error) { + goto release; + } + /* + * can only get here if MSG_PEEK is not + * set therefore, m should point at the + * head of the rcv queue; if it doesn't, + * it means something drastically + * changed while we were out from behind + * the lock in sodelayed_copy. perhaps + * a RST on the stream. in any event, + * the stream has been interrupted. it's + * probably best just to return whatever + * data we've moved and let the caller + * sort it out... + */ + if (m != so->so_rcv.sb_mb) { + break; + } + } + socket_unlock(so, 0); + error = uiomove(mtod(m, caddr_t) + moff, + (int)len, uio); + socket_lock(so, 0); + + if (error) { + goto release; + } + } + } else { + uio_setresid(uio, (uio_resid(uio) - len)); + } + if (len == m->m_len - moff) { + if (m->m_flags & M_EOR) { + flags |= MSG_EOR; + } + if (flags & MSG_PEEK) { + m = m->m_next; + moff = 0; + } else { + nextrecord = m->m_nextpkt; + sbfree(&so->so_rcv, m); + m->m_nextpkt = NULL; + + 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; + } + if (m != NULL) { + m->m_nextpkt = nextrecord; + if (nextrecord == NULL) { + so->so_rcv.sb_lastrecord = m; + } + } else { + so->so_rcv.sb_mb = nextrecord; + SB_EMPTY_FIXUP(&so->so_rcv); + } + SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); + } + } else { + if (flags & MSG_PEEK) { + moff += len; + } else { + if (mp != NULL) { + int copy_flag; + + if (flags & MSG_DONTWAIT) { + copy_flag = M_DONTWAIT; + } else { + copy_flag = M_WAIT; + } + *mp = m_copym(m, 0, len, copy_flag); + /* + * Failed to allocate an mbuf? + * Adjust uio_resid back, it was + * adjusted down by len bytes which + * we didn't copy over. + */ + if (*mp == NULL) { + uio_setresid(uio, + (uio_resid(uio) + len)); + break; + } + } + m->m_data += len; + m->m_len -= len; + so->so_rcv.sb_cc -= len; + } + } + if (so->so_oobmark) { + if ((flags & MSG_PEEK) == 0) { + so->so_oobmark -= len; + if (so->so_oobmark == 0) { + so->so_state |= SS_RCVATMARK; + break; + } + } else { + offset += len; + if (offset == so->so_oobmark) { + break; + } + } + } + if (flags & MSG_EOR) { + break; + } + /* + * If the MSG_WAITALL or MSG_WAITSTREAM flag is set + * (for non-atomic socket), we must not quit until + * "uio->uio_resid == 0" or an error termination. + * If a signal/timeout occurs, return with a short + * count but without error. Keep sockbuf locked + * against other readers. + */ + 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 CONTENT_FILTER + && cfil_sock_data_pending(&so->so_rcv) == 0 +#endif /* CONTENT_FILTER */ + )) { + goto release; + } + + /* + * Depending on the protocol (e.g. TCP), the following + * might cause the socket lock to be dropped and later + * be reacquired, and more data could have arrived and + * have been appended to the receive socket buffer by + * the time it returns. Therefore, we only sleep in + * sbwait() below if and only if the socket buffer is + * empty, in order to avoid a false sleep. + */ + if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb != NULL) { + (*pr->pr_usrreqs->pru_rcvd)(so, flags); + } + + SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); + + if (so->so_rcv.sb_mb == NULL && sbwait(&so->so_rcv)) { + error = 0; + goto release; + } + /* + * have to wait until after we get back from the sbwait + * to do the copy because we will drop the lock if we + * have enough data that has been delayed... by dropping + * the lock we open up a window allowing the netisr + * thread to process the incoming packets and to change + * the state of this socket... we're issuing the sbwait + * because the socket is empty and we're expecting the + * netisr thread to wake us up when more packets arrive; + * if we allow that processing to happen and then sbwait + * we could stall forever with packets sitting in the + * socket if no further packets arrive from the remote + * side. + * + * we want to copy before we've collected all the data + * to satisfy this request to allow the copy to overlap + * the incoming packet processing on an MP system + */ + if (delayed_copy_len > sorecvmincopy && + (delayed_copy_len > (so->so_rcv.sb_hiwat / 2))) { + error = sodelayed_copy(so, uio, + &free_list, &delayed_copy_len); + + if (error) { + goto release; + } + } + m = so->so_rcv.sb_mb; + if (m != NULL) { + nextrecord = m->m_nextpkt; + } + SB_MB_CHECK(&so->so_rcv); + } + } +#ifdef MORE_LOCKING_DEBUG + if (so->so_usecount <= 1) { + panic("%s: after big while so=%p ref=%d on socket\n", + __func__, so, so->so_usecount); + /* NOTREACHED */ + } +#endif + + if (m != NULL && pr->pr_flags & PR_ATOMIC) { + if (so->so_options & SO_DONTTRUNC) { + flags |= MSG_RCVMORE; + } else { + flags |= MSG_TRUNC; + if ((flags & MSG_PEEK) == 0) { + (void) sbdroprecord(&so->so_rcv); + } + } + } + + /* + * pru_rcvd below (for TCP) may cause more data to be received + * if the socket lock is dropped prior to sending the ACK; some + * legacy OpenTransport applications don't handle this well + * (if it receives less data than requested while MSG_HAVEMORE + * is set), and so we set the flag now based on what we know + * prior to calling pru_rcvd. + */ + if ((so->so_options & SO_WANTMORE) && so->so_rcv.sb_cc > 0) { + flags |= MSG_HAVEMORE; + } + + if ((flags & MSG_PEEK) == 0) { + if (m == NULL) { + so->so_rcv.sb_mb = nextrecord; + /* + * First part is an inline SB_EMPTY_FIXUP(). Second + * part makes sure sb_lastrecord is up-to-date if + * there is still data in the socket buffer. + */ + if (so->so_rcv.sb_mb == NULL) { + so->so_rcv.sb_mbtail = NULL; + so->so_rcv.sb_lastrecord = NULL; + } else if (nextrecord->m_nextpkt == NULL) { + so->so_rcv.sb_lastrecord = nextrecord; + } + SB_MB_CHECK(&so->so_rcv); + } + SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); + if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) { + (*pr->pr_usrreqs->pru_rcvd)(so, flags); + } + } + + if (delayed_copy_len) { + error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len); + if (error) { + goto release; + } + } + if (free_list != NULL) { + m_freem_list(free_list); + free_list = NULL; + } + + if (orig_resid == uio_resid(uio) && orig_resid && + (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { + sbunlock(&so->so_rcv, TRUE); /* keep socket locked */ + goto restart; + } + + if (flagsp != NULL) { + *flagsp |= flags; + } +release: +#ifdef MORE_LOCKING_DEBUG + if (so->so_usecount <= 1) { + panic("%s: release so=%p ref=%d on socket\n", __func__, + so, so->so_usecount); + /* NOTREACHED */ + } +#endif + if (delayed_copy_len) { + error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len); + } + + if (free_list != NULL) { + m_freem_list(free_list); + } + + sbunlock(&so->so_rcv, FALSE); /* will unlock socket */ + + if (en_tracing) { + KERNEL_ENERGYTRACE(kEnTrActKernSockRead, DBG_FUNC_END, + VM_KERNEL_ADDRPERM(so), + ((error == EWOULDBLOCK) ? kEnTrFlagNoWork : 0), + (int64_t)(orig_resid - uio_resid(uio))); + } + KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, so, uio_resid(uio), + so->so_rcv.sb_cc, 0, error); + + return error; +} + +/* + * Returns: 0 Success + * uiomove:EFAULT + */ +static int +sodelayed_copy(struct socket *so, struct uio *uio, struct mbuf **free_list, + user_ssize_t *resid) +{ + int error = 0; + struct mbuf *m; + + m = *free_list; + + socket_unlock(so, 0); + + while (m != NULL && error == 0) { + error = uiomove(mtod(m, caddr_t), (int)m->m_len, uio); + m = m->m_next; + } + m_freem_list(*free_list); + + *free_list = NULL; + *resid = 0; + + socket_lock(so, 0); + + return error; +} + +static int +sodelayed_copy_list(struct socket *so, struct recv_msg_elem *msgarray, + u_int uiocnt, struct mbuf **free_list, user_ssize_t *resid) +{ +#pragma unused(so) + int error = 0; + struct mbuf *ml, *m; + int i = 0; + struct uio *auio; + + for (ml = *free_list, i = 0; ml != NULL && i < uiocnt; + ml = ml->m_nextpkt, i++) { + auio = msgarray[i].uio; + for (m = ml; m != NULL; m = m->m_next) { + error = uiomove(mtod(m, caddr_t), m->m_len, auio); + if (error != 0) { + goto out; + } + } + } +out: + m_freem_list(*free_list); + + *free_list = NULL; + *resid = 0; + + return error; +} + +int +soreceive_list(struct socket *so, struct recv_msg_elem *msgarray, u_int uiocnt, + int *flagsp) +{ + struct mbuf *m; + struct mbuf *nextrecord; + struct mbuf *ml = NULL, *free_list = NULL, *free_tail = NULL; + int error; + user_ssize_t len, pktlen, delayed_copy_len = 0; + struct protosw *pr = so->so_proto; + user_ssize_t resid; + struct proc *p = current_proc(); + struct uio *auio = NULL; + int npkts = 0; + int sblocked = 0; + struct sockaddr **psa = NULL; + struct mbuf **controlp = NULL; + int can_delay; + int flags; + struct mbuf *free_others = NULL; + + 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); + + /* + * 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 (flagsp != NULL) { + flags = *flagsp; + } else { + flags = 0; + } + if (flags & ~(MSG_PEEK | MSG_WAITALL | MSG_DONTWAIT | MSG_NEEDSA | + MSG_NBIO)) { + printf("%s invalid flags 0x%x\n", __func__, flags); + error = EINVAL; + goto out; + } + if (so->so_type != SOCK_DGRAM) { + error = EINVAL; + 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 (msgarray == 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 = recv_msg_array_resid(msgarray, uiocnt); + if (resid < 0 || resid > INT_MAX) { + error = EINVAL; + goto out; + } + + if (!(flags & MSG_PEEK) && sorecvmincopy > 0) { + can_delay = 1; + } else { + can_delay = 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 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, %s]: defunct so 0x%llx [%d,%d] (%d)\n", + __func__, proc_pid(p), proc_best_name(p), + (uint64_t)DEBUG_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; + } + +next: + /* + * The uio may be empty + */ + if (npkts >= uiocnt) { + error = 0; + goto release; + } +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; + } + + error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); + if (error) { + goto release; + } + sblocked = 1; + + m = so->so_rcv.sb_mb; + /* + * Block awaiting more datagram if needed + */ + if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && + (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || + ((flags & MSG_WAITALL) && npkts < uiocnt))))) { + /* + * 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; + if ((flags & MSG_PEEK) == 0) { + so->so_error = 0; + } + 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 release; + } + if ((so->so_state & SS_NBIO) || + (flags & (MSG_DONTWAIT | MSG_NBIO))) { + error = EWOULDBLOCK; + goto release; + } + /* + * Do not block if we got some data + */ + if (free_list != NULL) { + error = 0; + goto release; + } + + SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); + + sbunlock(&so->so_rcv, TRUE); /* keep socket locked */ + sblocked = 0; + + error = sbwait(&so->so_rcv); + if (error) { + goto release; + } + goto restart; + } + + OSIncrementAtomicLong(&p->p_stats->p_ru.ru_msgrcv); + SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); + + /* + * Consume the current uio index as we have a datagram + */ + auio = msgarray[npkts].uio; + resid = uio_resid(auio); + msgarray[npkts].which |= SOCK_MSG_DATA; + psa = (msgarray[npkts].which & SOCK_MSG_SA) ? + &msgarray[npkts].psa : NULL; + controlp = (msgarray[npkts].which & SOCK_MSG_CONTROL) ? + &msgarray[npkts].controlp : NULL; + npkts += 1; + nextrecord = m->m_nextpkt; + + if ((pr->pr_flags & PR_ADDR) && m->m_type == MT_SONAME) { + error = soreceive_addr(p, so, psa, flags, &m, &nextrecord, 1); + if (error == ERESTART) { + goto restart; + } else if (error != 0) { + goto release; + } + } + + if (m != NULL && m->m_type == MT_CONTROL) { + error = soreceive_ctl(so, controlp, flags, &m, &nextrecord); + if (error != 0) { + goto release; + } + } + + if (m->m_pkthdr.len == 0) { + printf("%s:%d so %llx pkt %llx type %u pktlen null\n", + __func__, __LINE__, + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + (uint64_t)DEBUG_KERNEL_ADDRPERM(m), + m->m_type); + } + + /* + * Loop to copy the mbufs of the current record + * Support zero length packets + */ + ml = NULL; + pktlen = 0; + while (m != NULL && (len = resid - pktlen) >= 0 && error == 0) { + if (m->m_len == 0) { + panic("%p m_len zero", m); + } + if (m->m_type == 0) { + panic("%p m_type zero", m); + } + /* + * Clip to the residual length + */ + if (len > m->m_len) { + len = m->m_len; + } + pktlen += len; + /* + * Copy the mbufs via the uio or delay the copy + * 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 (len > 0 && can_delay == 0) { + socket_unlock(so, 0); + error = uiomove(mtod(m, caddr_t), (int)len, auio); + socket_lock(so, 0); + if (error) { + goto release; + } + } else { + delayed_copy_len += len; + } + + if (len == m->m_len) { + /* + * m was entirely copied + */ + sbfree(&so->so_rcv, m); + nextrecord = m->m_nextpkt; + m->m_nextpkt = NULL; + + /* + * Set the first packet to the head of the free list + */ + if (free_list == NULL) { + free_list = m; + } + /* + * Link current packet to tail of free list + */ + if (ml == NULL) { + if (free_tail != NULL) { + free_tail->m_nextpkt = m; + } + free_tail = m; + } + /* + * Link current mbuf to last mbuf of current packet + */ + if (ml != NULL) { + ml->m_next = m; + } + ml = m; + + /* + * Move next buf to head of socket buffer + */ + so->so_rcv.sb_mb = m = ml->m_next; + ml->m_next = 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 + */ + 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)DEBUG_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) { + /* + * Copyout first the freelist then the partial mbuf + */ + socket_unlock(so, 0); + if (delayed_copy_len) { + error = sodelayed_copy_list(so, msgarray, + uiocnt, &free_list, &delayed_copy_len); + } + + if (error == 0) { + error = uiomove(mtod(m, caddr_t), (int)len, + auio); + } + socket_lock(so, 0); + if (error) { + goto release; + } + + 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 (npkts < 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 next; + } + if (flagsp != NULL) { + *flagsp |= flags; + } + +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 (sblocked) { + sbunlock(&so->so_rcv, FALSE); /* will unlock socket */ + } else { + socket_unlock(so, 1); + } + + if (delayed_copy_len) { + error = sodelayed_copy_list(so, msgarray, uiocnt, + &free_list, &delayed_copy_len); + } +out: + /* + * Amortize the cost of freeing the mbufs + */ + if (free_list != NULL) { + m_freem_list(free_list); + } + if (free_others != NULL) { + m_freem_list(free_others); + } + + KERNEL_DEBUG(DBG_FNC_SORECEIVE_LIST | DBG_FUNC_END, error, + 0, 0, 0, 0); + return error; +} + +static int +so_statistics_event_to_nstat_event(int64_t *input_options, + uint64_t *nstat_event) +{ + int error = 0; + switch (*input_options) { + case SO_STATISTICS_EVENT_ENTER_CELLFALLBACK: + *nstat_event = NSTAT_EVENT_SRC_ENTER_CELLFALLBACK; + break; + case SO_STATISTICS_EVENT_EXIT_CELLFALLBACK: + *nstat_event = NSTAT_EVENT_SRC_EXIT_CELLFALLBACK; + break; +#if (DEBUG || DEVELOPMENT) + case SO_STATISTICS_EVENT_RESERVED_1: + *nstat_event = NSTAT_EVENT_SRC_RESERVED_1; + break; + case SO_STATISTICS_EVENT_RESERVED_2: + *nstat_event = NSTAT_EVENT_SRC_RESERVED_2; + break; +#endif /* (DEBUG || DEVELOPMENT) */ + default: + error = EINVAL; + break; + } + 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); + } + 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); + } +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; + + /* + * 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 + * copyin:EFAULT + */ +int +sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) +{ + size_t valsize; + + /* + * If the user gives us more than we wanted, we ignore it, + * but if we don't get the minimum length the caller + * wants, we return EINVAL. On success, sopt->sopt_valsize + * is set to however much we actually retrieved. + */ + if ((valsize = sopt->sopt_valsize) < minlen) { + return EINVAL; + } + if (valsize > len) { + sopt->sopt_valsize = valsize = len; + } + + if (sopt->sopt_p != kernproc) { + return copyin(sopt->sopt_val, buf, valsize); + } + + bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), buf, valsize); + return 0; +} + +/* + * sooptcopyin_timeval + * Copy in a timeval value into tv_p, and take into account whether the + * the calling process is 64-bit or 32-bit. Moved the sanity checking + * code here so that we can verify the 64-bit tv_sec value before we lose + * the top 32-bits assigning tv64.tv_sec to tv_p->tv_sec. + */ +static int +sooptcopyin_timeval(struct sockopt *sopt, struct timeval *tv_p) +{ + int error; + + if (proc_is64bit(sopt->sopt_p)) { + struct user64_timeval tv64; + + if (sopt->sopt_valsize < sizeof(tv64)) { + return EINVAL; + } + + sopt->sopt_valsize = sizeof(tv64); + if (sopt->sopt_p != kernproc) { + error = copyin(sopt->sopt_val, &tv64, sizeof(tv64)); + if (error != 0) { + return error; + } + } else { + bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), &tv64, + sizeof(tv64)); + } + if (tv64.tv_sec < 0 || tv64.tv_sec > LONG_MAX || + tv64.tv_usec < 0 || tv64.tv_usec >= 1000000) { + return EDOM; + } + + tv_p->tv_sec = tv64.tv_sec; + tv_p->tv_usec = tv64.tv_usec; + } else { + struct user32_timeval tv32; + + if (sopt->sopt_valsize < sizeof(tv32)) { + return EINVAL; + } + + sopt->sopt_valsize = sizeof(tv32); + if (sopt->sopt_p != kernproc) { + error = copyin(sopt->sopt_val, &tv32, sizeof(tv32)); + if (error != 0) { + return error; + } + } else { + bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), &tv32, + sizeof(tv32)); + } +#ifndef __LP64__ + /* + * K64todo "comparison is always false due to + * limited range of data type" + */ + if (tv32.tv_sec < 0 || tv32.tv_sec > LONG_MAX || + tv32.tv_usec < 0 || tv32.tv_usec >= 1000000) { + return EDOM; + } +#endif + tv_p->tv_sec = tv32.tv_sec; + tv_p->tv_usec = tv32.tv_usec; + } + return 0; +} + +int +soopt_cred_check(struct socket *so, int priv, boolean_t allow_root, + boolean_t ignore_delegate) +{ + kauth_cred_t cred = NULL; + proc_t ep = PROC_NULL; + uid_t uid; + int error = 0; + + if (ignore_delegate == false && so->so_flags & SOF_DELEGATED) { + ep = proc_find(so->e_pid); + if (ep) { + cred = kauth_cred_proc_ref(ep); + } + } + + uid = kauth_cred_getuid(cred ? cred : so->so_cred); + + /* uid is 0 for root */ + if (uid != 0 || !allow_root) { + error = priv_check_cred(cred ? cred : so->so_cred, priv, 0); + } + if (cred) { + kauth_cred_unref(&cred); + } + if (ep != PROC_NULL) { + proc_rele(ep); + } + + return error; +} + +/* + * Returns: 0 Success + * EINVAL + * ENOPROTOOPT + * ENOBUFS + * EDOM + * sooptcopyin:EINVAL + * sooptcopyin:EFAULT + * sooptcopyin_timeval:EINVAL + * sooptcopyin_timeval:EFAULT + * sooptcopyin_timeval:EDOM + * :EOPNOTSUPP[AF_UNIX] + * :???w + * sflt_attach_private:??? [whatever a filter author chooses] + * :??? [whatever a filter author chooses] + * + * Notes: Other returns depend on the protocol family; all + * returns depend on what the filter author causes + * their filter to return. + */ +int +sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) +{ + int error, optval; + int64_t long_optval; + struct linger l; + struct timeval tv; + + if (sopt->sopt_dir != SOPT_SET) { + sopt->sopt_dir = SOPT_SET; + } + + if (dolock) { + socket_lock(so, 1); + } + + if ((so->so_state & (SS_CANTRCVMORE | SS_CANTSENDMORE)) == + (SS_CANTRCVMORE | SS_CANTSENDMORE) && + (so->so_flags & SOF_NPX_SETOPTSHUT) == 0) { + /* the socket has been shutdown, no more sockopt's */ + error = EINVAL; + goto out; + } + + error = sflt_setsockopt(so, sopt); + if (error != 0) { + if (error == EJUSTRETURN) { + error = 0; + } + goto out; + } + + if (sopt->sopt_level != SOL_SOCKET) { + if (so->so_proto != NULL && + so->so_proto->pr_ctloutput != NULL) { + error = (*so->so_proto->pr_ctloutput)(so, sopt); + goto out; + } + error = ENOPROTOOPT; + } else { + /* + * Allow socket-level (SOL_SOCKET) options to be filtered by + * the protocol layer, if needed. A zero value returned from + * the handler means use default socket-level processing as + * done by the rest of this routine. Otherwise, any other + * return value indicates that the option is unsupported. + */ + if (so->so_proto != NULL && (error = so->so_proto->pr_usrreqs-> + pru_socheckopt(so, sopt)) != 0) { + goto out; + } + + error = 0; + switch (sopt->sopt_name) { + case SO_LINGER: + case SO_LINGER_SEC: + error = sooptcopyin(sopt, &l, sizeof(l), sizeof(l)); + if (error != 0) { + goto out; + } + + so->so_linger = (sopt->sopt_name == SO_LINGER) ? + l.l_linger : l.l_linger * hz; + if (l.l_onoff != 0) { + so->so_options |= SO_LINGER; + } else { + so->so_options &= ~SO_LINGER; + } + break; + + case SO_DEBUG: + case SO_KEEPALIVE: + case SO_DONTROUTE: + case SO_USELOOPBACK: + case SO_BROADCAST: + case SO_REUSEADDR: + case SO_REUSEPORT: + case SO_OOBINLINE: + case SO_TIMESTAMP: + case SO_TIMESTAMP_MONOTONIC: + case SO_TIMESTAMP_CONTINUOUS: + case SO_DONTTRUNC: + case SO_WANTMORE: + case SO_WANTOOBFLAG: + case SO_NOWAKEFROMSLEEP: + case SO_NOAPNFALLBK: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) { + goto out; + } + if (optval) { + so->so_options |= sopt->sopt_name; + } else { + so->so_options &= ~sopt->sopt_name; + } + break; + + case SO_SNDBUF: + case SO_RCVBUF: + case SO_SNDLOWAT: + case SO_RCVLOWAT: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) { + goto out; + } + + /* + * Values < 1 make no sense for any of these + * options, so disallow them. + */ + if (optval < 1) { + error = EINVAL; + goto out; + } + + switch (sopt->sopt_name) { + case SO_SNDBUF: + 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 out; + } + sb->sb_flags |= SB_USRSIZE; + sb->sb_flags &= ~SB_AUTOSIZE; + sb->sb_idealsize = (u_int32_t)optval; + break; + } + /* + * Make sure the low-water is never greater than + * the high-water. + */ + 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 > hiwat) ? + hiwat : optval; + + if (space >= so->so_snd.sb_lowat) { + sowwakeup(so); + } + break; + } + 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 != 0) { + goto out; + } + + switch (sopt->sopt_name) { + case SO_SNDTIMEO: + so->so_snd.sb_timeo = tv; + break; + case SO_RCVTIMEO: + so->so_rcv.sb_timeo = tv; + break; + } + break; + + case SO_NKE: { + struct so_nke nke; + + error = sooptcopyin(sopt, &nke, sizeof(nke), + sizeof(nke)); + if (error != 0) { + goto out; + } + + error = sflt_attach_internal(so, nke.nke_handle); + break; + } + + case SO_NOSIGPIPE: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) { + goto out; + } + if (optval != 0) { + so->so_flags |= SOF_NOSIGPIPE; + } else { + so->so_flags &= ~SOF_NOSIGPIPE; + } + break; + + case SO_NOADDRERR: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) { + goto out; + } + if (optval != 0) { + so->so_flags |= SOF_NOADDRAVAIL; + } else { + so->so_flags &= ~SOF_NOADDRAVAIL; + } + break; + + case SO_REUSESHAREUID: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) { + goto out; + } + if (optval != 0) { + so->so_flags |= SOF_REUSESHAREUID; + } else { + so->so_flags &= ~SOF_REUSESHAREUID; + } + break; + + case SO_NOTIFYCONFLICT: + if (kauth_cred_issuser(kauth_cred_get()) == 0) { + error = EPERM; + goto out; + } + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) { + goto out; + } + if (optval != 0) { + so->so_flags |= SOF_NOTIFYCONFLICT; + } else { + so->so_flags &= ~SOF_NOTIFYCONFLICT; + } + break; + + case SO_RESTRICTIONS: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) { + goto out; + } + + error = so_set_restrictions(so, optval); + break; + + case SO_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) { + error = soopt_cred_check(so, + PRIV_NET_RESTRICTED_AWDL, false, false); + if (error == 0) { + inp_set_awdl_unrestricted( + sotoinpcb(so)); + } + } else { + inp_clear_awdl_unrestricted(sotoinpcb(so)); + } + break; + case SO_INTCOPROC_ALLOW: + if (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 && + inp_get_intcoproc_allowed(sotoinpcb(so)) == FALSE) { + error = soopt_cred_check(so, + PRIV_NET_RESTRICTED_INTCOPROC, false, false); + if (error == 0) { + inp_set_intcoproc_allowed( + sotoinpcb(so)); + } + } else if (optval == 0) { + inp_clear_intcoproc_allowed(sotoinpcb(so)); + } + break; + + case SO_LABEL: + error = EOPNOTSUPP; + 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; + } + if (optval >= SO_TC_NET_SERVICE_OFFSET) { + int netsvc = optval - SO_TC_NET_SERVICE_OFFSET; + error = so_set_net_service_type(so, netsvc); + goto out; + } + error = so_set_traffic_class(so, optval); + if (error != 0) { + goto out; + } + so->so_flags1 &= ~SOF1_TC_NET_SERV_TYPE; + so->so_netsvctype = _NET_SERVICE_TYPE_UNSPEC; + 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; + } + +#if (DEVELOPMENT || DEBUG) + 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; + } +#endif /* (DEVELOPMENT || DEBUG) */ + + 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; + +#if (DEVELOPMENT || DEBUG) + case SO_DEFUNCTIT: + error = sosetdefunct(current_proc(), so, 0, FALSE); + if (error == 0) { + error = sodefunct(current_proc(), so, 0); + } + + break; +#endif /* (DEVELOPMENT || DEBUG) */ + + case SO_DEFUNCTOK: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0 || (so->so_flags & SOF_DEFUNCT)) { + if (error == 0) { + error = EBADF; + } + goto out; + } + /* + * Any process can set SO_DEFUNCTOK (clear + * SOF_NODEFUNCT), but only root can clear + * SO_DEFUNCTOK (set SOF_NODEFUNCT). + */ + if (optval == 0 && + kauth_cred_issuser(kauth_cred_get()) == 0) { + error = EPERM; + goto out; + } + if (optval) { + so->so_flags &= ~SOF_NODEFUNCT; + } else { + so->so_flags |= SOF_NODEFUNCT; + } + + if (SOCK_DOM(so) == PF_INET || + SOCK_DOM(so) == PF_INET6) { + char s[MAX_IPv6_STR_LEN]; + char d[MAX_IPv6_STR_LEN]; + struct inpcb *inp = sotoinpcb(so); + + SODEFUNCTLOG("%s[%d, %s]: so 0x%llx " + "[%s %s:%d -> %s:%d] is now marked " + "as %seligible for " + "defunct\n", __func__, proc_selfpid(), + proc_best_name(current_proc()), + (uint64_t)DEBUG_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, %s]: so 0x%llx [%d,%d] " + "is now marked as %seligible for " + "defunct\n", + __func__, proc_selfpid(), + proc_best_name(current_proc()), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), + (so->so_flags & SOF_NODEFUNCT) ? + "not " : ""); + } + break; + + case SO_ISDEFUNCT: + /* This option is not settable */ + error = EINVAL; + break; + + case SO_OPPORTUNISTIC: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error == 0) { + error = so_set_opportunistic(so, optval); + } + break; + + case SO_FLUSH: + /* This option is handled by lower layer(s) */ + error = 0; + break; + + case SO_RECV_ANYIF: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error == 0) { + error = so_set_recv_anyif(so, optval); + } + break; + + case SO_TRAFFIC_MGT_BACKGROUND: { + /* This option is handled by lower layer(s) */ + error = 0; + break; + } + +#if FLOW_DIVERT + case SO_FLOW_DIVERT_TOKEN: + error = flow_divert_token_set(so, sopt); + break; +#endif /* FLOW_DIVERT */ + + + case SO_DELEGATED: + if ((error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval))) != 0) { + break; + } + + error = so_set_effective_pid(so, optval, sopt->sopt_p, true); + 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, true); + break; + } + +#if NECP + case SO_NECP_ATTRIBUTES: + error = necp_set_socket_attributes(so, sopt); + break; + + case SO_NECP_CLIENTUUID: { + if (SOCK_DOM(so) == PF_MULTIPATH) { + /* Handled by MPTCP itself */ + break; + } + + if (SOCK_DOM(so) != PF_INET && SOCK_DOM(so) != PF_INET6) { + error = EINVAL; + goto out; + } + + struct inpcb *inp = sotoinpcb(so); + if (!uuid_is_null(inp->necp_client_uuid)) { + // Clear out the old client UUID if present + necp_inpcb_remove_cb(inp); + } + + error = sooptcopyin(sopt, &inp->necp_client_uuid, + sizeof(uuid_t), sizeof(uuid_t)); + if (error != 0) { + goto out; + } + + if (uuid_is_null(inp->necp_client_uuid)) { + error = EINVAL; + goto out; + } + + pid_t current_pid = proc_pid(current_proc()); + error = necp_client_register_socket_flow(current_pid, + inp->necp_client_uuid, inp); + if (error != 0) { + uuid_clear(inp->necp_client_uuid); + goto out; + } + + if (inp->inp_lport != 0) { + // There is a bound local port, so this is not + // a fresh socket. Assign to the client. + necp_client_assign_from_socket(current_pid, inp->necp_client_uuid, inp); + } + + break; + } + case SO_NECP_LISTENUUID: { + if (SOCK_DOM(so) != PF_INET && SOCK_DOM(so) != PF_INET6) { + error = EINVAL; + goto out; + } + + struct inpcb *inp = sotoinpcb(so); + if (!uuid_is_null(inp->necp_client_uuid)) { + error = EINVAL; + goto out; + } + + error = sooptcopyin(sopt, &inp->necp_client_uuid, + sizeof(uuid_t), sizeof(uuid_t)); + if (error != 0) { + goto out; + } + + if (uuid_is_null(inp->necp_client_uuid)) { + error = EINVAL; + goto out; + } + + error = necp_client_register_socket_listener(proc_pid(current_proc()), + inp->necp_client_uuid, inp); + if (error != 0) { + uuid_clear(inp->necp_client_uuid); + goto out; + } + + // Mark that the port registration is held by NECP + inp->inp_flags2 |= INP2_EXTERNAL_PORT; + + break; + } +#endif /* NECP */ + + case SO_EXTENDED_BK_IDLE: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error == 0) { + error = so_set_extended_bk_idle(so, optval); + } + break; + + case SO_MARK_CELLFALLBACK: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) { + goto out; + } + if (optval < 0) { + error = EINVAL; + goto out; + } + if (optval == 0) { + so->so_flags1 &= ~SOF1_CELLFALLBACK; + } else { + so->so_flags1 |= SOF1_CELLFALLBACK; + } + break; + + case SO_STATISTICS_EVENT: + error = sooptcopyin(sopt, &long_optval, + sizeof(long_optval), sizeof(long_optval)); + if (error != 0) { + goto out; + } + u_int64_t nstat_event = 0; + error = so_statistics_event_to_nstat_event( + &long_optval, &nstat_event); + if (error != 0) { + goto out; + } + nstat_pcb_event(sotoinpcb(so), nstat_event); + break; + + case SO_NET_SERVICE_TYPE: { + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) { + goto out; + } + error = so_set_net_service_type(so, optval); + break; + } + + case SO_QOSMARKING_POLICY_OVERRIDE: + error = priv_check_cred(kauth_cred_get(), + PRIV_NET_QOSMARKING_POLICY_OVERRIDE, 0); + if (error != 0) { + goto out; + } + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) { + goto out; + } + if (optval == 0) { + so->so_flags1 &= ~SOF1_QOSMARKING_POLICY_OVERRIDE; + } else { + so->so_flags1 |= SOF1_QOSMARKING_POLICY_OVERRIDE; + } + break; + + case SO_MPKL_SEND_INFO: { + struct so_mpkl_send_info so_mpkl_send_info; + + error = sooptcopyin(sopt, &so_mpkl_send_info, + sizeof(struct so_mpkl_send_info), sizeof(struct so_mpkl_send_info)); + if (error != 0) { + goto out; + } + uuid_copy(so->so_mpkl_send_uuid, so_mpkl_send_info.mpkl_uuid); + so->so_mpkl_send_proto = so_mpkl_send_info.mpkl_proto; + + if (uuid_is_null(so->so_mpkl_send_uuid) && so->so_mpkl_send_proto == 0) { + so->so_flags1 &= ~SOF1_MPKL_SEND_INFO; + } else { + so->so_flags1 |= SOF1_MPKL_SEND_INFO; + } + break; + } + case SO_WANT_KEV_SOCKET_CLOSED: { + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error != 0) { + goto out; + } + if (optval == 0) { + so->so_flags1 &= ~SOF1_WANT_KEV_SOCK_CLOSED; + } else { + so->so_flags1 |= SOF1_WANT_KEV_SOCK_CLOSED; + } + break; + } + default: + error = ENOPROTOOPT; + break; + } + if (error == 0 && so->so_proto != NULL && + so->so_proto->pr_ctloutput != NULL) { + (void) so->so_proto->pr_ctloutput(so, sopt); + } + } +out: + if (dolock) { + socket_unlock(so, 1); + } + return error; +} + +/* Helper routines for getsockopt */ +int +sooptcopyout(struct sockopt *sopt, void *buf, size_t len) +{ + int error; + size_t valsize; + + error = 0; + + /* + * Documented get behavior is that we always return a value, + * possibly truncated to fit in the user's buffer. + * Traditional behavior is that we always tell the user + * precisely how much we copied, rather than something useful + * like the total amount we had available for her. + * Note that this interface is not idempotent; the entire answer must + * generated ahead of time. + */ + valsize = min(len, sopt->sopt_valsize); + sopt->sopt_valsize = valsize; + if (sopt->sopt_val != USER_ADDR_NULL) { + if (sopt->sopt_p != kernproc) { + error = copyout(buf, sopt->sopt_val, valsize); + } else { + bcopy(buf, CAST_DOWN(caddr_t, sopt->sopt_val), valsize); + } + } + return error; +} + +static int +sooptcopyout_timeval(struct sockopt *sopt, const struct timeval *tv_p) +{ + int error; + size_t len; + struct user64_timeval tv64 = {}; + struct user32_timeval tv32 = {}; + const void * val; + size_t valsize; + + error = 0; + if (proc_is64bit(sopt->sopt_p)) { + len = sizeof(tv64); + tv64.tv_sec = tv_p->tv_sec; + tv64.tv_usec = tv_p->tv_usec; + val = &tv64; + } else { + len = sizeof(tv32); + tv32.tv_sec = tv_p->tv_sec; + tv32.tv_usec = tv_p->tv_usec; + val = &tv32; + } + valsize = min(len, sopt->sopt_valsize); + sopt->sopt_valsize = valsize; + if (sopt->sopt_val != USER_ADDR_NULL) { + if (sopt->sopt_p != kernproc) { + error = copyout(val, sopt->sopt_val, valsize); + } else { + bcopy(val, CAST_DOWN(caddr_t, sopt->sopt_val), valsize); + } + } + return error; +} + +/* + * Return: 0 Success + * ENOPROTOOPT + * :EOPNOTSUPP[AF_UNIX] + * :??? + * :??? + */ +int +sogetoptlock(struct socket *so, struct sockopt *sopt, int dolock) +{ + int error, optval; + struct linger l; + struct timeval tv; + + if (sopt->sopt_dir != SOPT_GET) { + sopt->sopt_dir = SOPT_GET; + } + + if (dolock) { + socket_lock(so, 1); + } + + error = sflt_getsockopt(so, sopt); + if (error != 0) { + if (error == EJUSTRETURN) { + error = 0; + } + goto out; + } + + if (sopt->sopt_level != SOL_SOCKET) { + if (so->so_proto != NULL && + so->so_proto->pr_ctloutput != NULL) { + error = (*so->so_proto->pr_ctloutput)(so, sopt); + goto out; + } + error = ENOPROTOOPT; + } else { + /* + * Allow socket-level (SOL_SOCKET) options to be filtered by + * the protocol layer, if needed. A zero value returned from + * the handler means use default socket-level processing as + * done by the rest of this routine. Otherwise, any other + * return value indicates that the option is unsupported. + */ + if (so->so_proto != NULL && (error = so->so_proto->pr_usrreqs-> + pru_socheckopt(so, sopt)) != 0) { + goto out; + } + + error = 0; + switch (sopt->sopt_name) { + case SO_LINGER: + case SO_LINGER_SEC: + l.l_onoff = ((so->so_options & SO_LINGER) ? 1 : 0); + l.l_linger = (sopt->sopt_name == SO_LINGER) ? + so->so_linger : so->so_linger / hz; + error = sooptcopyout(sopt, &l, sizeof(l)); + break; + + case SO_USELOOPBACK: + case SO_DONTROUTE: + case SO_DEBUG: + case SO_KEEPALIVE: + case SO_REUSEADDR: + case SO_REUSEPORT: + case SO_BROADCAST: + case SO_OOBINLINE: + case SO_TIMESTAMP: + case SO_TIMESTAMP_MONOTONIC: + case SO_TIMESTAMP_CONTINUOUS: + case SO_DONTTRUNC: + case SO_WANTMORE: + case SO_WANTOOBFLAG: + case SO_NOWAKEFROMSLEEP: + case SO_NOAPNFALLBK: + optval = so->so_options & sopt->sopt_name; +integer: + error = sooptcopyout(sopt, &optval, sizeof(optval)); + break; + + case SO_TYPE: + optval = so->so_type; + goto integer; + + case SO_NREAD: + if (so->so_proto->pr_flags & PR_ATOMIC) { + int pkt_total; + struct mbuf *m1; + + pkt_total = 0; + m1 = so->so_rcv.sb_mb; + while (m1 != NULL) { + if (m1->m_type == MT_DATA || + m1->m_type == MT_HEADER || + m1->m_type == MT_OOBDATA) { + pkt_total += m1->m_len; + } + m1 = m1->m_next; + } + optval = pkt_total; + } else { + 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) { + cnt += 1; + m1 = m1->m_nextpkt; + } + optval = cnt; + goto integer; + } else { + error = ENOPROTOOPT; + break; + } + + case SO_NWRITE: + optval = so->so_snd.sb_cc; + goto integer; + + case SO_ERROR: + optval = so->so_error; + so->so_error = 0; + 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; + + case SO_SNDLOWAT: + optval = so->so_snd.sb_lowat; + goto integer; + + case SO_RCVLOWAT: + optval = so->so_rcv.sb_lowat; + goto integer; + + case SO_SNDTIMEO: + case SO_RCVTIMEO: + tv = (sopt->sopt_name == SO_SNDTIMEO ? + so->so_snd.sb_timeo : so->so_rcv.sb_timeo); + + error = sooptcopyout_timeval(sopt, &tv); + break; + + case SO_NOSIGPIPE: + optval = (so->so_flags & SOF_NOSIGPIPE); + goto integer; + + case SO_NOADDRERR: + optval = (so->so_flags & SOF_NOADDRAVAIL); + goto integer; + + case SO_REUSESHAREUID: + optval = (so->so_flags & SOF_REUSESHAREUID); + goto integer; + + + case SO_NOTIFYCONFLICT: + optval = (so->so_flags & SOF_NOTIFYCONFLICT); + goto integer; + + case SO_RESTRICTIONS: + optval = so_get_restrictions(so); + goto integer; + + case SO_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_INTCOPROC_ALLOW: + if (SOCK_DOM(so) == PF_INET6) { + optval = inp_get_intcoproc_allowed( + sotoinpcb(so)); + goto integer; + } else { + error = EOPNOTSUPP; + } + break; + + case SO_LABEL: + error = EOPNOTSUPP; + break; + + case SO_PEERLABEL: + error = EOPNOTSUPP; + break; + +#ifdef __APPLE_API_PRIVATE + case SO_UPCALLCLOSEWAIT: + optval = (so->so_flags & SOF_UPCALLCLOSEWAIT); + goto integer; +#endif + case SO_RANDOMPORT: + optval = (so->so_flags & SOF_BINDRANDOMPORT); + goto integer; + + case SO_NP_EXTENSIONS: { + struct so_np_extensions sonpx = {}; + + sonpx.npx_flags = (so->so_flags & SOF_NPX_SETOPTSHUT) ? + SONPX_SETOPTSHUT : 0; + sonpx.npx_mask = SONPX_MASK_VALID; + + error = sooptcopyout(sopt, &sonpx, + sizeof(struct so_np_extensions)); + break; + } + + case SO_TRAFFIC_CLASS: + optval = so->so_traffic_class; + goto integer; + + case SO_RECV_TRAFFIC_CLASS: + optval = (so->so_flags & SOF_RECV_TRAFFIC_CLASS); + goto integer; + +#if (DEVELOPMENT || DEBUG) + case SO_TRAFFIC_CLASS_DBG: + error = sogetopt_tcdbg(so, sopt); + break; +#endif /* (DEVELOPMENT || DEBUG) */ + + 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; + + case SO_NECP_CLIENTUUID: { + uuid_t *ncu; + + if (SOCK_DOM(so) == PF_MULTIPATH) { + ncu = &mpsotomppcb(so)->necp_client_uuid; + } else if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) { + ncu = &sotoinpcb(so)->necp_client_uuid; + } else { + error = EINVAL; + goto out; + } + + error = sooptcopyout(sopt, ncu, sizeof(uuid_t)); + break; + } + + case SO_NECP_LISTENUUID: { + uuid_t *nlu; + + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) { + if (sotoinpcb(so)->inp_flags2 & INP2_EXTERNAL_PORT) { + nlu = &sotoinpcb(so)->necp_client_uuid; + } else { + error = ENOENT; + goto out; + } + } else { + error = EINVAL; + goto out; + } + + error = sooptcopyout(sopt, nlu, sizeof(uuid_t)); + 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 */ + + case SO_EXTENDED_BK_IDLE: + optval = (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED); + goto integer; + case SO_MARK_CELLFALLBACK: + optval = ((so->so_flags1 & SOF1_CELLFALLBACK) > 0) + ? 1 : 0; + goto integer; + case SO_NET_SERVICE_TYPE: { + if ((so->so_flags1 & SOF1_TC_NET_SERV_TYPE)) { + optval = so->so_netsvctype; + } else { + optval = NET_SERVICE_TYPE_BE; + } + goto integer; + } + case SO_NETSVC_MARKING_LEVEL: + optval = so_get_netsvc_marking_level(so); + goto integer; + + case SO_MPKL_SEND_INFO: { + struct so_mpkl_send_info so_mpkl_send_info; + + uuid_copy(so_mpkl_send_info.mpkl_uuid, so->so_mpkl_send_uuid); + so_mpkl_send_info.mpkl_proto = so->so_mpkl_send_proto; + error = sooptcopyout(sopt, &so_mpkl_send_info, + sizeof(struct so_mpkl_send_info)); + break; + } + default: + error = ENOPROTOOPT; + break; + } + } +out: + if (dolock) { + socket_unlock(so, 1); + } + return error; +} + +/* + * 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) +{ + struct mbuf *m, *m_prev; + int sopt_size = sopt->sopt_valsize; + int how; + + 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 == NULL) { + return ENOBUFS; + } + if (sopt_size > MLEN) { + MCLGET(m, how); + if ((m->m_flags & M_EXT) == 0) { + m_free(m); + return ENOBUFS; + } + m->m_len = min(MCLBYTES, sopt_size); + } else { + m->m_len = min(MLEN, sopt_size); + } + sopt_size -= m->m_len; + *mp = m; + m_prev = m; + + while (sopt_size > 0) { + MGET(m, how, MT_DATA); + if (m == NULL) { + m_freem(*mp); + return ENOBUFS; + } + if (sopt_size > MLEN) { + 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); + } else { + m->m_len = min(MLEN, sopt_size); + } + sopt_size -= m->m_len; + m_prev->m_next = m; + m_prev = m; + } + return 0; +} + +/* copyin sopt data into mbuf chain */ +int +soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) +{ + struct mbuf *m0 = m; + + if (sopt->sopt_val == USER_ADDR_NULL) { + return 0; + } + while (m != NULL && sopt->sopt_valsize >= m->m_len) { + if (sopt->sopt_p != kernproc) { + int error; + + error = copyin(sopt->sopt_val, mtod(m, char *), + m->m_len); + if (error != 0) { + m_freem(m0); + return error; + } + } else { + bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), + mtod(m, char *), m->m_len); + } + sopt->sopt_valsize -= m->m_len; + sopt->sopt_val += m->m_len; + m = m->m_next; + } + /* should be allocated enoughly at ip6_sooptmcopyin() */ + if (m != NULL) { + panic("soopt_mcopyin"); + /* NOTREACHED */ + } + return 0; +} + +/* copyout mbuf chain data into soopt */ +int +soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) +{ + struct mbuf *m0 = m; + size_t valsize = 0; + + if (sopt->sopt_val == USER_ADDR_NULL) { + return 0; + } + while (m != NULL && sopt->sopt_valsize >= m->m_len) { + if (sopt->sopt_p != kernproc) { + int error; + + error = copyout(mtod(m, char *), sopt->sopt_val, + m->m_len); + if (error != 0) { + m_freem(m0); + return error; + } + } else { + bcopy(mtod(m, char *), + CAST_DOWN(caddr_t, sopt->sopt_val), m->m_len); + } + sopt->sopt_valsize -= m->m_len; + sopt->sopt_val += m->m_len; + valsize += m->m_len; + m = m->m_next; + } + if (m != NULL) { + /* enough soopt buffer should be given from user-land */ + m_freem(m0); + return EINVAL; + } + sopt->sopt_valsize = valsize; + return 0; +} + +void +sohasoutofband(struct socket *so) +{ + if (so->so_pgid < 0) { + gsignal(-so->so_pgid, SIGURG); + } else if (so->so_pgid > 0) { + proc_signal(so->so_pgid, SIGURG); + } + selwakeup(&so->so_rcv.sb_sel); + if (so->so_rcv.sb_flags & SB_KNOTE) { + KNOTE(&so->so_rcv.sb_sel.si_note, + (NOTE_OOB | SO_FILT_HINT_LOCKED)); + } +} + +int +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)) { + revents |= events & (POLLIN | POLLRDNORM); + } + } + + if (events & (POLLOUT | POLLWRNORM)) { + if (sowriteable(so)) { + revents |= events & (POLLOUT | POLLWRNORM); + } + } + + if (events & (POLLPRI | POLLRDBAND)) { + if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) { + revents |= events & (POLLPRI | POLLRDBAND); + } + } + + if (revents == 0) { + if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { + /* + * Darwin sets the flag first, + * BSD calls selrecord first + */ + so->so_rcv.sb_flags |= SB_SEL; + selrecord(p, &so->so_rcv.sb_sel, wql); + } + + if (events & (POLLOUT | POLLWRNORM)) { + /* + * Darwin sets the flag first, + * BSD calls selrecord first + */ + so->so_snd.sb_flags |= SB_SEL; + selrecord(p, &so->so_snd.sb_sel, wql); + } + } + + socket_unlock(so, 1); + return revents; +} + +int +soo_kqfilter(struct fileproc *fp, struct knote *kn, struct kevent_qos_s *kev) +{ + struct socket *so = (struct socket *)fp->fp_glob->fg_data; + int result; + + socket_lock(so, 1); + so_update_last_owner_locked(so, PROC_NULL); + so_update_policy(so); + + switch (kn->kn_filter) { + case EVFILT_READ: + kn->kn_filtid = EVFILTID_SOREAD; + break; + case EVFILT_WRITE: + kn->kn_filtid = EVFILTID_SOWRITE; + break; + case EVFILT_SOCK: + kn->kn_filtid = EVFILTID_SCK; + break; + case EVFILT_EXCEPT: + kn->kn_filtid = EVFILTID_SOEXCEPT; + break; + default: + socket_unlock(so, 1); + knote_set_error(kn, EINVAL); + return 0; + } + + /* + * call the appropriate sub-filter attach + * with the socket still locked + */ + result = knote_fops(kn)->f_attach(kn, kev); + + socket_unlock(so, 1); + + return result; +} + +static int +filt_soread_common(struct knote *kn, struct kevent_qos_s *kev, struct socket *so) +{ + int retval = 0; + int64_t data = 0; + + if (so->so_options & SO_ACCEPTCONN) { + /* + * Radar 6615193 handle the listen case dynamically + * for kqueue read filter. This allows to call listen() + * after registering the kqueue EVFILT_READ. + */ + + retval = !TAILQ_EMPTY(&so->so_comp); + data = so->so_qlen; + goto out; + } + + /* socket isn't a listener */ + /* + * NOTE_LOWAT specifies new low water mark in data, i.e. + * the bytes of protocol data. We therefore exclude any + * control bytes. + */ + data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl; + + if (kn->kn_sfflags & NOTE_OOB) { + if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) { + kn->kn_fflags |= NOTE_OOB; + data -= so->so_oobmark; + retval = 1; + goto out; + } + } + + 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; + retval = 1; + goto out; + } + + if (so->so_error) { /* temporary udp error */ + retval = 1; + goto out; + } + + int64_t lowwat = so->so_rcv.sb_lowat; + /* + * Ensure that when NOTE_LOWAT is used, the derived + * low water mark is bounded by socket's rcv buf's + * high and low water mark values. + */ + 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; + } + } + + /* + * While the `data` field is the amount of data to read, + * 0-sized packets need to wake up the kqueue, see 58140856, + * so we need to take control bytes into account too. + */ + retval = (so->so_rcv.sb_cc >= lowwat); + +out: + if (retval && kev) { + knote_fill_kevent(kn, kev, data); + } + return retval; +} + +static int +filt_sorattach(struct knote *kn, __unused struct kevent_qos_s *kev) +{ + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + + /* socket locked */ + + /* + * If the caller explicitly asked for OOB results (e.g. poll()) + * from EVFILT_READ, then save that off in the hookid field + * and reserve the kn_flags EV_OOBAND bit for output only. + */ + if (kn->kn_filter == EVFILT_READ && + kn->kn_flags & EV_OOBAND) { + kn->kn_flags &= ~EV_OOBAND; + kn->kn_hook32 = EV_OOBAND; + } else { + kn->kn_hook32 = 0; + } + if (KNOTE_ATTACH(&so->so_rcv.sb_sel.si_note, kn)) { + so->so_rcv.sb_flags |= SB_KNOTE; + } + + /* indicate if event is already fired */ + return filt_soread_common(kn, NULL, so); +} + +static void +filt_sordetach(struct knote *kn) +{ + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + + socket_lock(so, 1); + if (so->so_rcv.sb_flags & SB_KNOTE) { + if (KNOTE_DETACH(&so->so_rcv.sb_sel.si_note, kn)) { + so->so_rcv.sb_flags &= ~SB_KNOTE; + } + } + socket_unlock(so, 1); +} + +/*ARGSUSED*/ +static int +filt_soread(struct knote *kn, long hint) +{ + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + int retval; + + if ((hint & SO_FILT_HINT_LOCKED) == 0) { + socket_lock(so, 1); + } + + retval = filt_soread_common(kn, NULL, so); + + if ((hint & SO_FILT_HINT_LOCKED) == 0) { + socket_unlock(so, 1); + } + + return retval; +} + +static int +filt_sortouch(struct knote *kn, struct kevent_qos_s *kev) +{ + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + int retval; + + socket_lock(so, 1); + + /* save off the new input fflags and data */ + kn->kn_sfflags = kev->fflags; + kn->kn_sdata = kev->data; + + /* determine if changes result in fired events */ + retval = filt_soread_common(kn, NULL, so); + + socket_unlock(so, 1); + + return retval; +} + +static int +filt_sorprocess(struct knote *kn, struct kevent_qos_s *kev) +{ + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + int retval; + + socket_lock(so, 1); + retval = filt_soread_common(kn, kev, so); + socket_unlock(so, 1); + + return retval; +} + +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; +} + +static int +filt_sowrite_common(struct knote *kn, struct kevent_qos_s *kev, struct socket *so) +{ + int ret = 0; + int64_t data = sbspace(&so->so_snd); + + if (so->so_state & SS_CANTSENDMORE) { + kn->kn_flags |= EV_EOF; + kn->kn_fflags = so->so_error; + ret = 1; + goto out; + } - if (error) - goto release; + if (so->so_error) { /* temporary udp error */ + ret = 1; + goto out; } - if (free_list) { - m_freem_list((struct mbuf *)free_list); - free_list = (struct mbuf *)0; + + if (!socanwrite(so)) { + ret = 0; + goto out; } - 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); - goto restart; + + if (so->so_flags1 & SOF1_PRECONNECT_DATA) { + ret = 1; + goto out; } - if (flagsp) - *flagsp |= flags; -release: -#ifdef MORE_LOCKING_DEBUG - if (so->so_usecount <= 1) - panic("soreceive: release so=%x ref=%d on socket\n", so, so->so_usecount); + 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 (data >= lowwat) { + if ((so->so_flags & SOF_NOTSENT_LOWAT) +#if (DEBUG || DEVELOPMENT) + && so_notsent_lowat_check == 1 +#endif /* DEBUG || DEVELOPMENT */ + ) { + 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 - if (delayed_copy_len) { - error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len); + else { + ret = 1; + goto out; + } + } else { + ret = 1; + } } - if (free_list) { - m_freem_list((struct mbuf *)free_list); + if (so_wait_for_if_feedback(so)) { + ret = 0; } - sbunlock(&so->so_rcv, 0); /* will unlock socket */ - - // LP64todo - fix this! - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, - so, - uio_resid(uio), - so->so_rcv.sb_cc, - 0, - error); - return (error); +out: + if (ret && kev) { + knote_fill_kevent(kn, kev, data); + } + return ret; } - -static int sodelayed_copy(struct socket *so, struct uio *uio, struct mbuf **free_list, int *resid) +static int +filt_sowattach(struct knote *kn, __unused struct kevent_qos_s *kev) { - int error = 0; - struct mbuf *m; + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; - m = *free_list; + /* socket locked */ + if (KNOTE_ATTACH(&so->so_snd.sb_sel.si_note, kn)) { + so->so_snd.sb_flags |= SB_KNOTE; + } - socket_unlock(so, 0); + /* determine if its already fired */ + return filt_sowrite_common(kn, NULL, so); +} - while (m && error == 0) { +static void +filt_sowdetach(struct knote *kn) +{ + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + socket_lock(so, 1); - error = uiomove(mtod(m, caddr_t), (int)m->m_len, uio); + if (so->so_snd.sb_flags & SB_KNOTE) { + if (KNOTE_DETACH(&so->so_snd.sb_sel.si_note, kn)) { + so->so_snd.sb_flags &= ~SB_KNOTE; + } + } + socket_unlock(so, 1); +} - m = m->m_next; +/*ARGSUSED*/ +static int +filt_sowrite(struct knote *kn, long hint) +{ + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + int ret; + + if ((hint & SO_FILT_HINT_LOCKED) == 0) { + socket_lock(so, 1); } - m_freem_list(*free_list); - *free_list = (struct mbuf *)NULL; - *resid = 0; + ret = filt_sowrite_common(kn, NULL, so); - socket_lock(so, 0); + if ((hint & SO_FILT_HINT_LOCKED) == 0) { + socket_unlock(so, 1); + } - return (error); + return ret; } - -int -soshutdown(so, how) - register struct socket *so; - int how; +static int +filt_sowtouch(struct knote *kn, struct kevent_qos_s *kev) { - register struct protosw *pr = so->so_proto; + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; int ret; socket_lock(so, 1); - - sflt_notify(so, sock_evt_shutdown, &how); - if (how != SHUT_WR) { - sorflush(so); - postevent(so, 0, EV_RCLOSED); - } - if (how != SHUT_RD) { - ret = ((*pr->pr_usrreqs->pru_shutdown)(so)); - postevent(so, 0, EV_WCLOSED); - KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_END, 0,0,0,0,0); - socket_unlock(so, 1); - return(ret); - } + /*save off the new input fflags and data */ + kn->kn_sfflags = kev->fflags; + kn->kn_sdata = kev->data; + + /* determine if these changes result in a triggered event */ + ret = filt_sowrite_common(kn, NULL, so); - KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_END, 0,0,0,0,0); socket_unlock(so, 1); - return (0); + + return ret; } -void -sorflush(so) - register struct socket *so; +static int +filt_sowprocess(struct knote *kn, struct kevent_qos_s *kev) { - register struct sockbuf *sb = &so->so_rcv; - register struct protosw *pr = so->so_proto; - struct sockbuf asb; - -#ifdef MORE_LOCKING_DEBUG - lck_mtx_t * mutex_held; - - if (so->so_proto->pr_getlock != NULL) - mutex_held = (*so->so_proto->pr_getlock)(so, 0); - else - mutex_held = so->so_proto->pr_domain->dom_mtx; - lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); -#endif + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + int ret; - sflt_notify(so, sock_evt_flush_read, NULL); + socket_lock(so, 1); + ret = filt_sowrite_common(kn, kev, so); + socket_unlock(so, 1); - 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 (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) - (*pr->pr_domain->dom_dispose)(asb.sb_mb); - sbrelease(&asb); + return ret; } -/* - * 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. - */ -int -sooptcopyin(sopt, buf, len, minlen) - struct sockopt *sopt; - void *buf; - size_t len; - size_t minlen; +static int +filt_sockev_common(struct knote *kn, struct kevent_qos_s *kev, + struct socket *so, long ev_hint) { - size_t valsize; + int ret = 0; + int64_t data = 0; + uint32_t level_trigger = 0; - /* - * If the user gives us more than we wanted, we ignore it, - * but if we don't get the minimum length the caller - * wants, we return EINVAL. On success, sopt->sopt_valsize - * is set to however much we actually retrieved. - */ - if ((valsize = sopt->sopt_valsize) < minlen) - return EINVAL; - if (valsize > len) - sopt->sopt_valsize = valsize = len; + if (ev_hint & SO_FILT_HINT_CONNRESET) { + kn->kn_fflags |= NOTE_CONNRESET; + } + if (ev_hint & SO_FILT_HINT_TIMEOUT) { + kn->kn_fflags |= NOTE_TIMEOUT; + } + if (ev_hint & SO_FILT_HINT_NOSRCADDR) { + kn->kn_fflags |= NOTE_NOSRCADDR; + } + if (ev_hint & SO_FILT_HINT_IFDENIED) { + kn->kn_fflags |= NOTE_IFDENIED; + } + if (ev_hint & SO_FILT_HINT_KEEPALIVE) { + kn->kn_fflags |= NOTE_KEEPALIVE; + } + if (ev_hint & SO_FILT_HINT_ADAPTIVE_WTIMO) { + kn->kn_fflags |= NOTE_ADAPTIVE_WTIMO; + } + if (ev_hint & SO_FILT_HINT_ADAPTIVE_RTIMO) { + kn->kn_fflags |= NOTE_ADAPTIVE_RTIMO; + } + if ((ev_hint & SO_FILT_HINT_CONNECTED) || + (so->so_state & SS_ISCONNECTED)) { + kn->kn_fflags |= NOTE_CONNECTED; + level_trigger |= NOTE_CONNECTED; + } + if ((ev_hint & SO_FILT_HINT_DISCONNECTED) || + (so->so_state & SS_ISDISCONNECTED)) { + kn->kn_fflags |= NOTE_DISCONNECTED; + level_trigger |= NOTE_DISCONNECTED; + } + if (ev_hint & SO_FILT_HINT_CONNINFO_UPDATED) { + if (so->so_proto != NULL && + (so->so_proto->pr_flags & PR_EVCONNINFO)) { + kn->kn_fflags |= NOTE_CONNINFO_UPDATED; + } + } - if (sopt->sopt_p != 0) - return (copyin(sopt->sopt_val, buf, valsize)); + if ((ev_hint & SO_FILT_HINT_NOTIFY_ACK) || + tcp_notify_ack_active(so)) { + kn->kn_fflags |= NOTE_NOTIFY_ACK; + } - bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), buf, valsize); - return 0; -} + if ((so->so_state & SS_CANTRCVMORE) +#if CONTENT_FILTER + && cfil_sock_data_pending(&so->so_rcv) == 0 +#endif /* CONTENT_FILTER */ + ) { + kn->kn_fflags |= NOTE_READCLOSED; + level_trigger |= NOTE_READCLOSED; + } -int -sosetopt(so, sopt) - struct socket *so; - struct sockopt *sopt; -{ - int error, optval; - struct linger l; - struct timeval tv; - short val; + if (so->so_state & SS_CANTSENDMORE) { + kn->kn_fflags |= NOTE_WRITECLOSED; + level_trigger |= NOTE_WRITECLOSED; + } - socket_lock(so, 1); + if ((ev_hint & SO_FILT_HINT_SUSPEND) || + (so->so_flags & SOF_SUSPENDED)) { + kn->kn_fflags &= ~(NOTE_SUSPEND | NOTE_RESUME); - if (sopt->sopt_dir != SOPT_SET) { - sopt->sopt_dir = SOPT_SET; + /* If resume event was delivered before, reset it */ + kn->kn_hook32 &= ~NOTE_RESUME; + + kn->kn_fflags |= NOTE_SUSPEND; + level_trigger |= NOTE_SUSPEND; } - { - 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_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); - } + if ((ev_hint & SO_FILT_HINT_RESUME) || + (so->so_flags & SOF_SUSPENDED) == 0) { + kn->kn_fflags &= ~(NOTE_SUSPEND | NOTE_RESUME); + + /* If suspend event was delivered before, reset it */ + kn->kn_hook32 &= ~NOTE_SUSPEND; + + kn->kn_fflags |= NOTE_RESUME; + level_trigger |= NOTE_RESUME; + } + + if (so->so_error != 0) { + ret = 1; + data = so->so_error; + kn->kn_flags |= EV_EOF; + } else { + u_int32_t data32 = 0; + get_sockev_state(so, &data32); + data = data32; + } + + /* Reset any events that are not requested on this knote */ + kn->kn_fflags &= (kn->kn_sfflags & EVFILT_SOCK_ALL_MASK); + level_trigger &= (kn->kn_sfflags & EVFILT_SOCK_ALL_MASK); + + /* Find the level triggerred events that are already delivered */ + level_trigger &= kn->kn_hook32; + level_trigger &= EVFILT_SOCK_LEVEL_TRIGGER_MASK; + + /* Do not deliver level triggerred events more than once */ + if ((kn->kn_fflags & ~level_trigger) != 0) { + ret = 1; + } + + if (ret && kev) { + /* + * Store the state of the events being delivered. This + * state can be used to deliver level triggered events + * ateast once and still avoid waking up the application + * multiple times as long as the event is active. + */ + if (kn->kn_fflags != 0) { + kn->kn_hook32 |= (kn->kn_fflags & + EVFILT_SOCK_LEVEL_TRIGGER_MASK); } - - if (filtered != 0) { - socket_lock(so, 0); - sflt_unuse(so); - - if (error) { - if (error == EJUSTRETURN) - error = 0; - goto bad; - } + + /* + * NOTE_RESUME and NOTE_SUSPEND are an exception, deliver + * only one of them and remember the last one that was + * delivered last + */ + if (kn->kn_fflags & NOTE_SUSPEND) { + kn->kn_hook32 &= ~NOTE_RESUME; + } + if (kn->kn_fflags & NOTE_RESUME) { + kn->kn_hook32 &= ~NOTE_SUSPEND; } + + knote_fill_kevent(kn, kev, data); } + return ret; +} - error = 0; - if (sopt->sopt_level != SOL_SOCKET) { - if (so->so_proto && so->so_proto->pr_ctloutput) { - error = (*so->so_proto->pr_ctloutput) - (so, sopt); - socket_unlock(so, 1); - return (error); +static int +filt_sockattach(struct knote *kn, __unused struct kevent_qos_s *kev) +{ + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + + /* socket locked */ + kn->kn_hook32 = 0; + if (KNOTE_ATTACH(&so->so_klist, kn)) { + so->so_flags |= SOF_KNOTE; + } + + /* determine if event already fired */ + return filt_sockev_common(kn, NULL, so, 0); +} + +static void +filt_sockdetach(struct knote *kn) +{ + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->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; } - error = ENOPROTOOPT; - } else { - switch (sopt->sopt_name) { - case SO_LINGER: - case SO_LINGER_SEC: - error = sooptcopyin(sopt, &l, sizeof l, sizeof l); - if (error) - goto bad; + } + socket_unlock(so, 1); +} - so->so_linger = (sopt->sopt_name == SO_LINGER) ? l.l_linger : l.l_linger * hz; - if (l.l_onoff) - so->so_options |= SO_LINGER; - else - so->so_options &= ~SO_LINGER; - break; +static int +filt_sockev(struct knote *kn, long hint) +{ + int ret = 0, locked = 0; + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + long ev_hint = (hint & SO_FILT_HINT_EV); - case SO_DEBUG: - case SO_KEEPALIVE: - case SO_DONTROUTE: - case SO_USELOOPBACK: - case SO_BROADCAST: - case SO_REUSEADDR: - case SO_REUSEPORT: - case SO_OOBINLINE: - case SO_TIMESTAMP: -#ifdef __APPLE__ - case SO_DONTTRUNC: - case SO_WANTMORE: - case SO_WANTOOBFLAG: -#endif - error = sooptcopyin(sopt, &optval, sizeof optval, - sizeof optval); - if (error) - goto bad; - if (optval) - so->so_options |= sopt->sopt_name; - else - so->so_options &= ~sopt->sopt_name; - break; + if ((hint & SO_FILT_HINT_LOCKED) == 0) { + socket_lock(so, 1); + locked = 1; + } - case SO_SNDBUF: - case SO_RCVBUF: - case SO_SNDLOWAT: - case SO_RCVLOWAT: - error = sooptcopyin(sopt, &optval, sizeof optval, - sizeof optval); - if (error) - goto bad; + ret = filt_sockev_common(kn, NULL, so, ev_hint); - /* - * Values < 1 make no sense for any of these - * options, so disallow them. - */ - if (optval < 1) { - error = EINVAL; - goto bad; - } + if (locked) { + socket_unlock(so, 1); + } - switch (sopt->sopt_name) { - case SO_SNDBUF: - case SO_RCVBUF: - if (sbreserve(sopt->sopt_name == SO_SNDBUF ? - &so->so_snd : &so->so_rcv, - (u_long) optval) == 0) { - error = ENOBUFS; - goto bad; - } - break; + return ret; +} - /* - * Make sure the low-water is never greater than - * the high-water. - */ - case SO_SNDLOWAT: - so->so_snd.sb_lowat = - (optval > so->so_snd.sb_hiwat) ? - so->so_snd.sb_hiwat : optval; - break; - case SO_RCVLOWAT: - so->so_rcv.sb_lowat = - (optval > so->so_rcv.sb_hiwat) ? - so->so_rcv.sb_hiwat : optval; - break; - } - break; - case SO_SNDTIMEO: - case SO_RCVTIMEO: - error = sooptcopyin(sopt, &tv, sizeof tv, - sizeof tv); - if (error) - goto bad; - if (tv.tv_sec < 0 || tv.tv_sec > LONG_MAX || - tv.tv_usec < 0 || tv.tv_usec >= 1000000) { - error = EDOM; - goto bad; - } - - switch (sopt->sopt_name) { - case SO_SNDTIMEO: - so->so_snd.sb_timeo = tv; - break; - case SO_RCVTIMEO: - so->so_rcv.sb_timeo = tv; - break; - } - break; +/* + * filt_socktouch - update event state + */ +static int +filt_socktouch( + struct knote *kn, + struct kevent_qos_s *kev) +{ + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + uint32_t changed_flags; + int ret; + + socket_lock(so, 1); + + /* save off the [result] data and fflags */ + changed_flags = (kn->kn_sfflags ^ kn->kn_hook32); + + /* save off the new input fflags and data */ + kn->kn_sfflags = kev->fflags; + kn->kn_sdata = kev->data; + + /* restrict the current results to the (smaller?) set of new interest */ + /* + * For compatibility with previous implementations, we leave kn_fflags + * as they were before. + */ + //kn->kn_fflags &= kev->fflags; + + /* + * Since we keep track of events that are already + * delivered, if any of those events are not requested + * anymore the state related to them can be reset + */ + kn->kn_hook32 &= ~(changed_flags & EVFILT_SOCK_LEVEL_TRIGGER_MASK); - case SO_NKE: - { - struct so_nke nke; + /* determine if we have events to deliver */ + ret = filt_sockev_common(kn, NULL, so, 0); - error = sooptcopyin(sopt, &nke, - sizeof nke, sizeof nke); - if (error) - goto bad; + socket_unlock(so, 1); - error = sflt_attach_private(so, NULL, nke.nke_handle, 1); - break; - } + return ret; +} - case SO_NOSIGPIPE: - error = sooptcopyin(sopt, &optval, sizeof optval, - sizeof optval); - if (error) - goto bad; - if (optval) - so->so_flags |= SOF_NOSIGPIPE; - else - so->so_flags &= ~SOF_NOSIGPIPE; - - break; +/* + * filt_sockprocess - query event fired state and return data + */ +static int +filt_sockprocess(struct knote *kn, struct kevent_qos_s *kev) +{ + struct socket *so = (struct socket *)kn->kn_fp->fp_glob->fg_data; + int ret = 0; - case SO_NOADDRERR: - error = sooptcopyin(sopt, &optval, sizeof optval, - sizeof optval); - if (error) - goto bad; - if (optval) - so->so_flags |= SOF_NOADDRAVAIL; - else - so->so_flags &= ~SOF_NOADDRAVAIL; - - break; + socket_lock(so, 1); + + ret = filt_sockev_common(kn, kev, so, 0); - default: - error = ENOPROTOOPT; - break; - } - if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { - (void) ((*so->so_proto->pr_ctloutput) - (so, sopt)); - } - } -bad: socket_unlock(so, 1); - return (error); + + return ret; } -/* Helper routine for getsockopt */ -int -sooptcopyout(sopt, buf, len) - struct sockopt *sopt; - void *buf; - size_t len; +void +get_sockev_state(struct socket *so, u_int32_t *statep) { - int error; - size_t valsize; - - error = 0; + u_int32_t state = *(statep); /* - * Documented get behavior is that we always return a value, - * possibly truncated to fit in the user's buffer. - * Traditional behavior is that we always tell the user - * precisely how much we copied, rather than something useful - * like the total amount we had available for her. - * Note that this interface is not idempotent; the entire answer must - * generated ahead of time. + * If the state variable is already used by a previous event, + * reset it. */ - valsize = min(len, sopt->sopt_valsize); - sopt->sopt_valsize = valsize; - if (sopt->sopt_val != USER_ADDR_NULL) { - if (sopt->sopt_p != 0) - error = copyout(buf, sopt->sopt_val, valsize); - else - bcopy(buf, CAST_DOWN(caddr_t, sopt->sopt_val), valsize); + if (state != 0) { + return; } - return error; + + if (so->so_state & SS_ISCONNECTED) { + state |= SOCKEV_CONNECTED; + } else { + state &= ~(SOCKEV_CONNECTED); + } + state |= ((so->so_state & SS_ISDISCONNECTED) ? SOCKEV_DISCONNECTED : 0); + *(statep) = state; } -int -sogetopt(so, sopt) - struct socket *so; - struct sockopt *sopt; +#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) { - int error, optval; - struct linger l; - struct timeval tv; + 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 += scnprintf(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; +} - if (sopt->sopt_dir != SOPT_GET) { - sopt->sopt_dir = SOPT_GET; - } +lck_mtx_t * +socket_getlock(struct socket *so, int flags) +{ + if (so->so_proto->pr_getlock != NULL) { + return (*so->so_proto->pr_getlock)(so, flags); + } else { + return so->so_proto->pr_domain->dom_mtx; + } +} - socket_lock(so, 1); - - { - 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_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 (error) { - if (error == EJUSTRETURN) - error = 0; - socket_unlock(so, 1); - return error; - } +void +socket_lock(struct socket *so, int refcount) +{ + void *lr_saved; + + lr_saved = __builtin_return_address(0); + + if (so->so_proto->pr_lock) { + (*so->so_proto->pr_lock)(so, refcount, lr_saved); + } else { +#ifdef MORE_LOCKING_DEBUG + LCK_MTX_ASSERT(so->so_proto->pr_domain->dom_mtx, + LCK_MTX_ASSERT_NOTOWNED); +#endif + lck_mtx_lock(so->so_proto->pr_domain->dom_mtx); + if (refcount) { + so->so_usecount++; } + so->lock_lr[so->next_lock_lr] = lr_saved; + so->next_lock_lr = (so->next_lock_lr + 1) % SO_LCKDBG_MAX; } +} - error = 0; - if (sopt->sopt_level != SOL_SOCKET) { - if (so->so_proto && so->so_proto->pr_ctloutput) { - error = (*so->so_proto->pr_ctloutput) - (so, sopt); - socket_unlock(so, 1); - return (error); - } else { - socket_unlock(so, 1); - return (ENOPROTOOPT); - } +void +socket_lock_assert_owned(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 { - switch (sopt->sopt_name) { - case SO_LINGER: - case SO_LINGER_SEC: - l.l_onoff = so->so_options & SO_LINGER; - l.l_linger = (sopt->sopt_name == SO_LINGER) ? so->so_linger : - so->so_linger / hz; - error = sooptcopyout(sopt, &l, sizeof l); - break; + mutex_held = so->so_proto->pr_domain->dom_mtx; + } - case SO_USELOOPBACK: - case SO_DONTROUTE: - case SO_DEBUG: - case SO_KEEPALIVE: - case SO_REUSEADDR: - case SO_REUSEPORT: - case SO_BROADCAST: - case SO_OOBINLINE: - case SO_TIMESTAMP: -#ifdef __APPLE__ - case SO_DONTTRUNC: - case SO_WANTMORE: - case SO_WANTOOBFLAG: + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); +} + +int +socket_try_lock(struct socket *so) +{ + lck_mtx_t *mtx; + + if (so->so_proto->pr_getlock != NULL) { + mtx = (*so->so_proto->pr_getlock)(so, 0); + } else { + mtx = so->so_proto->pr_domain->dom_mtx; + } + + return lck_mtx_try_lock(mtx); +} + +void +socket_unlock(struct socket *so, int refcount) +{ + void *lr_saved; + lck_mtx_t *mutex_held; + + lr_saved = __builtin_return_address(0); + + if (so == NULL || so->so_proto == NULL) { + panic("%s: null so_proto so=%p\n", __func__, so); + /* NOTREACHED */ + } + + if (so->so_proto->pr_unlock) { + (*so->so_proto->pr_unlock)(so, refcount, lr_saved); + } else { + mutex_held = so->so_proto->pr_domain->dom_mtx; +#ifdef MORE_LOCKING_DEBUG + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); #endif - optval = so->so_options & sopt->sopt_name; -integer: - error = sooptcopyout(sopt, &optval, sizeof optval); - break; + so->unlock_lr[so->next_unlock_lr] = lr_saved; + so->next_unlock_lr = (so->next_unlock_lr + 1) % SO_LCKDBG_MAX; - case SO_TYPE: - optval = so->so_type; - goto integer; + if (refcount) { + if (so->so_usecount <= 0) { + panic("%s: bad refcount=%d so=%p (%d, %d, %d) " + "lrh=%s", __func__, so->so_usecount, so, + SOCK_DOM(so), so->so_type, + SOCK_PROTO(so), solockhistory_nr(so)); + /* NOTREACHED */ + } -#ifdef __APPLE__ - case SO_NREAD: - { - int pkt_total; - struct mbuf *m1; - - pkt_total = 0; - m1 = so->so_rcv.sb_mb; - if (so->so_proto->pr_flags & PR_ATOMIC) - { - while (m1) { - if (m1->m_type == MT_DATA) - pkt_total += m1->m_len; - m1 = m1->m_next; - } - optval = pkt_total; - } else - optval = so->so_rcv.sb_cc; - goto integer; + so->so_usecount--; + if (so->so_usecount == 0) { + sofreelastref(so, 1); + } } - 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; + lck_mtx_unlock(mutex_held); + } +} - case SO_SNDBUF: - optval = so->so_snd.sb_hiwat; - goto integer; +/* Called with socket locked, will unlock socket */ +void +sofree(struct socket *so) +{ + lck_mtx_t *mutex_held; - case SO_RCVBUF: - optval = so->so_rcv.sb_hiwat; - goto integer; + 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); - case SO_SNDLOWAT: - optval = so->so_snd.sb_lowat; - goto integer; + sofreelastref(so, 0); +} - case SO_RCVLOWAT: - optval = so->so_rcv.sb_lowat; - goto integer; +void +soreference(struct socket *so) +{ + socket_lock(so, 1); /* locks & take one reference on socket */ + socket_unlock(so, 0); /* unlock only */ +} - case SO_SNDTIMEO: - case SO_RCVTIMEO: - tv = (sopt->sopt_name == SO_SNDTIMEO ? - so->so_snd.sb_timeo : so->so_rcv.sb_timeo); +void +sodereference(struct socket *so) +{ + socket_lock(so, 0); + socket_unlock(so, 1); +} - error = sooptcopyout(sopt, &tv, sizeof tv); - break; +/* + * Set or clear SOF_MULTIPAGES on the socket to enable or disable the + * possibility of using jumbo clusters. Caller must ensure to hold + * the socket lock. + */ +void +somultipages(struct socket *so, boolean_t set) +{ + if (set) { + so->so_flags |= SOF_MULTIPAGES; + } else { + so->so_flags &= ~SOF_MULTIPAGES; + } +} - case SO_NOSIGPIPE: - optval = (so->so_flags & SOF_NOSIGPIPE); - goto integer; +void +soif2kcl(struct socket *so, boolean_t set) +{ + if (set) { + so->so_flags1 |= SOF1_IF_2KCL; + } else { + so->so_flags1 &= ~SOF1_IF_2KCL; + } +} - case SO_NOADDRERR: - optval = (so->so_flags & SOF_NOADDRAVAIL); - goto integer; +int +so_isdstlocal(struct socket *so) +{ + struct inpcb *inp = (struct inpcb *)so->so_pcb; - default: - error = ENOPROTOOPT; - break; - } - socket_unlock(so, 1); - return (error); + 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; } -/* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ int -soopt_getm(struct sockopt *sopt, struct mbuf **mp) +sosetdefunct(struct proc *p, struct socket *so, int level, boolean_t noforce) { - struct mbuf *m, *m_prev; - int sopt_size = sopt->sopt_valsize; + struct sockbuf *rcv, *snd; + int err = 0, defunct; - if (sopt_size > MAX_SOOPTGETM_SIZE) - return EMSGSIZE; + rcv = &so->so_rcv; + snd = &so->so_snd; - MGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT, MT_DATA); - if (m == 0) - return ENOBUFS; - if (sopt_size > MLEN) { - MCLGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT); - if ((m->m_flags & M_EXT) == 0) { - m_free(m); - return ENOBUFS; + 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 */ } - m->m_len = min(MCLBYTES, sopt_size); - } else { - m->m_len = min(MLEN, sopt_size); + goto done; } - sopt_size -= m->m_len; - *mp = m; - m_prev = m; - while (sopt_size) { - MGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT, MT_DATA); - if (m == 0) { - m_freem(*mp); - return ENOBUFS; - } - if (sopt_size > MLEN) { - MCLGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT); - if ((m->m_flags & M_EXT) == 0) { - m_freem(*mp); - return ENOBUFS; + if (so->so_flags & SOF_NODEFUNCT) { + if (noforce) { + err = EOPNOTSUPP; + if (p != PROC_NULL) { + SODEFUNCTLOG("%s[%d, %s]: (target pid %d " + "name %s level %d) so 0x%llx [%d,%d] " + "is not eligible for defunct " + "(%d)\n", __func__, proc_selfpid(), + proc_best_name(current_proc()), proc_pid(p), + proc_best_name(p), level, + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), err); } - m->m_len = min(MCLBYTES, sopt_size); + return err; + } + so->so_flags &= ~SOF_NODEFUNCT; + if (p != PROC_NULL) { + SODEFUNCTLOG("%s[%d, %s]: (target pid %d " + "name %s level %d) so 0x%llx [%d,%d] " + "defunct by force " + "(%d)\n", __func__, proc_selfpid(), + proc_best_name(current_proc()), proc_pid(p), + proc_best_name(p), level, + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), err); + } + } else if (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) { + struct inpcb *inp = (struct inpcb *)so->so_pcb; + struct ifnet *ifp = inp->inp_last_outifp; + + if (ifp && IFNET_IS_CELLULAR(ifp)) { + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_nocell); + } else if (so->so_flags & SOF_DELEGATED) { + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_nodlgtd); + } else if (soextbkidlestat.so_xbkidle_time == 0) { + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_notime); + } else if (noforce && p != PROC_NULL) { + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_active); + + so->so_flags1 |= SOF1_EXTEND_BK_IDLE_INPROG; + so->so_extended_bk_start = net_uptime(); + OSBitOrAtomic(P_LXBKIDLEINPROG, &p->p_ladvflag); + + inpcb_timer_sched(inp->inp_pcbinfo, INPCB_TIMER_LAZY); + + err = EOPNOTSUPP; + SODEFUNCTLOG("%s[%d, %s]: (target pid %d " + "name %s level %d) so 0x%llx [%d,%d] " + "extend bk idle " + "(%d)\n", __func__, proc_selfpid(), + proc_best_name(current_proc()), proc_pid(p), + proc_best_name(p), level, + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), err); + return err; } else { - m->m_len = min(MLEN, sopt_size); + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_forced); } - sopt_size -= m->m_len; - m_prev->m_next = m; - m_prev = m; } - return 0; -} -/* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ -int -soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) -{ - struct mbuf *m0 = m; + so->so_flags |= SOF_DEFUNCT; - if (sopt->sopt_val == USER_ADDR_NULL) - return 0; - while (m != NULL && sopt->sopt_valsize >= m->m_len) { - if (sopt->sopt_p != NULL) { - int error; + /* Prevent further data from being appended to the socket buffers */ + snd->sb_flags |= SB_DROP; + rcv->sb_flags |= SB_DROP; - error = copyin(sopt->sopt_val, mtod(m, char *), m->m_len); - if (error != 0) { - m_freem(m0); - return(error); - } - } else - bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), mtod(m, char *), m->m_len); - sopt->sopt_valsize -= m->m_len; - sopt->sopt_val += m->m_len; - m = m->m_next; + /* 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: + if (p != PROC_NULL) { + SODEFUNCTLOG("%s[%d, %s]: (target pid %d name %s level %d) " + "so 0x%llx [%d,%d] %s defunct%s\n", __func__, + proc_selfpid(), proc_best_name(current_proc()), + proc_pid(p), proc_best_name(p), level, + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), SOCK_DOM(so), + SOCK_TYPE(so), defunct ? "is already" : "marked as", + (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) ? + " extbkidle" : ""); } - if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ - panic("soopt_mcopyin"); - return 0; + return err; } -/* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ int -soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) +sodefunct(struct proc *p, struct socket *so, int level) { - struct mbuf *m0 = m; - size_t valsize = 0; + struct sockbuf *rcv, *snd; - if (sopt->sopt_val == USER_ADDR_NULL) - return 0; - while (m != NULL && sopt->sopt_valsize >= m->m_len) { - if (sopt->sopt_p != NULL) { - int error; + if (!(so->so_flags & SOF_DEFUNCT)) { + panic("%s improperly called", __func__); + /* NOTREACHED */ + } + if (so->so_state & SS_DEFUNCT) { + goto done; + } - error = copyout(mtod(m, char *), sopt->sopt_val, m->m_len); - if (error != 0) { - m_freem(m0); - return(error); - } - } else - bcopy(mtod(m, char *), CAST_DOWN(caddr_t, sopt->sopt_val), m->m_len); - sopt->sopt_valsize -= m->m_len; - sopt->sopt_val += m->m_len; - valsize += m->m_len; - m = m->m_next; + 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); + + if (p != PROC_NULL) { + SODEFUNCTLOG( + "%s[%d, %s]: (target pid %d name %s 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_best_name(current_proc()), + proc_pid(p), proc_best_name(p), level, + (uint64_t)DEBUG_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 if (p != PROC_NULL) { + SODEFUNCTLOG("%s[%d, %s]: (target pid %d name %s 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_best_name(current_proc()), + proc_pid(p), proc_best_name(p), level, + (uint64_t)DEBUG_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); } - if (m != NULL) { - /* enough soopt buffer should be given from user-land */ - m_freem(m0); - return(EINVAL); + + /* + * 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 */ } - sopt->sopt_valsize = valsize; + 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_state & 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; + OSIncrementAtomicLong((volatile long *)&sodefunct_calls); + +done: return 0; } -void -sohasoutofband(so) - register struct socket *so; +int +soresume(struct proc *p, struct socket *so, int locked) { - struct proc *p; + if (locked == 0) { + socket_lock(so, 1); + } - if (so->so_pgid < 0) - gsignal(-so->so_pgid, SIGURG); - else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0) - psignal(p, SIGURG); - selwakeup(&so->so_rcv.sb_sel); + if (so->so_flags1 & SOF1_EXTEND_BK_IDLE_INPROG) { + SODEFUNCTLOG("%s[%d, %s]: (target pid %d name %s) so 0x%llx " + "[%d,%d] resumed from bk idle\n", + __func__, proc_selfpid(), proc_best_name(current_proc()), + proc_pid(p), proc_best_name(p), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so)); + + so->so_flags1 &= ~SOF1_EXTEND_BK_IDLE_INPROG; + so->so_extended_bk_start = 0; + OSBitAndAtomic(~P_LXBKIDLEINPROG, &p->p_ladvflag); + + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_resumed); + OSDecrementAtomic(&soextbkidlestat.so_xbkidle_active); + VERIFY(soextbkidlestat.so_xbkidle_active >= 0); + } + if (locked == 0) { + socket_unlock(so, 1); + } + + return 0; } +/* + * Does not attempt to account for sockets that are delegated from + * the current process + */ int -sopoll(struct socket *so, int events, __unused kauth_cred_t cred, void * wql) +so_set_extended_bk_idle(struct socket *so, int optval) { - struct proc *p = current_proc(); - int revents = 0; + int error = 0; - socket_lock(so, 1); + if ((SOCK_DOM(so) != PF_INET && SOCK_DOM(so) != PF_INET6) || + SOCK_PROTO(so) != IPPROTO_TCP) { + OSDecrementAtomic(&soextbkidlestat.so_xbkidle_notsupp); + error = EOPNOTSUPP; + } else if (optval == 0) { + so->so_flags1 &= ~SOF1_EXTEND_BK_IDLE_WANTED; - if (events & (POLLIN | POLLRDNORM)) - if (soreadable(so)) - revents |= events & (POLLIN | POLLRDNORM); + soresume(current_proc(), so, 1); + } else { + struct proc *p = current_proc(); + struct fileproc *fp; + int count = 0; - if (events & (POLLOUT | POLLWRNORM)) - if (sowriteable(so)) - revents |= events & (POLLOUT | POLLWRNORM); + /* + * Unlock socket to avoid lock ordering issue with + * the proc fd table lock + */ + socket_unlock(so, 0); - if (events & (POLLPRI | POLLRDBAND)) - if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) - revents |= events & (POLLPRI | POLLRDBAND); + proc_fdlock(p); + fdt_foreach(fp, p) { + struct socket *so2; - if (revents == 0) { - if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { - /* Darwin sets the flag first, BSD calls selrecord first */ - so->so_rcv.sb_flags |= SB_SEL; - selrecord(p, &so->so_rcv.sb_sel, wql); + if (FILEGLOB_DTYPE(fp->fp_glob) != DTYPE_SOCKET) { + continue; + } + + so2 = (struct socket *)fp->fp_glob->fg_data; + if (so != so2 && + so2->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) { + count++; + } + if (count >= soextbkidlestat.so_xbkidle_maxperproc) { + break; + } } + proc_fdunlock(p); - if (events & (POLLOUT | POLLWRNORM)) { - /* Darwin sets the flag first, BSD calls selrecord first */ - so->so_snd.sb_flags |= SB_SEL; - selrecord(p, &so->so_snd.sb_sel, wql); + socket_lock(so, 0); + + if (count >= soextbkidlestat.so_xbkidle_maxperproc) { + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_toomany); + error = EBUSY; + } else if (so->so_flags & SOF_DELEGATED) { + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_nodlgtd); + error = EBUSY; + } else { + so->so_flags1 |= SOF1_EXTEND_BK_IDLE_WANTED; + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_wantok); } + SODEFUNCTLOG("%s[%d, %s]: so 0x%llx [%d,%d] " + "%s marked for extended bk idle\n", + __func__, proc_selfpid(), proc_best_name(current_proc()), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), + (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) ? + "is" : "not"); } - socket_unlock(so, 1); - return (revents); + return error; } -int soo_kqfilter(struct fileproc *fp, struct knote *kn, struct proc *p); - -int -soo_kqfilter(__unused struct fileproc *fp, struct knote *kn, __unused struct proc *p) +static void +so_stop_extended_bk_idle(struct socket *so) { - struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; - struct sockbuf *sb; - socket_lock(so, 1); + so->so_flags1 &= ~SOF1_EXTEND_BK_IDLE_INPROG; + so->so_extended_bk_start = 0; - switch (kn->kn_filter) { - case EVFILT_READ: - if (so->so_options & SO_ACCEPTCONN) - kn->kn_fop = &solisten_filtops; - else - kn->kn_fop = &soread_filtops; - sb = &so->so_rcv; - break; - case EVFILT_WRITE: - kn->kn_fop = &sowrite_filtops; - sb = &so->so_snd; - break; - default: - socket_unlock(so, 1); - return (1); + OSDecrementAtomic(&soextbkidlestat.so_xbkidle_active); + VERIFY(soextbkidlestat.so_xbkidle_active >= 0); + /* + * Force defunct + */ + sosetdefunct(current_proc(), so, + SHUTDOWN_SOCKET_LEVEL_DISCONNECT_INTERNAL, FALSE); + if (so->so_flags & SOF_DEFUNCT) { + sodefunct(current_proc(), so, + SHUTDOWN_SOCKET_LEVEL_DISCONNECT_INTERNAL); } - - if (KNOTE_ATTACH(&sb->sb_sel.si_note, kn)) - sb->sb_flags |= SB_KNOTE; - socket_unlock(so, 1); - return (0); } -static void -filt_sordetach(struct knote *kn) +void +so_drain_extended_bk_idle(struct socket *so) { - struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + if (so && (so->so_flags1 & SOF1_EXTEND_BK_IDLE_INPROG)) { + /* + * Only penalize sockets that have outstanding data + */ + if (so->so_rcv.sb_cc || so->so_snd.sb_cc) { + so_stop_extended_bk_idle(so); - socket_lock(so, 1); - if (so->so_rcv.sb_flags & SB_KNOTE) - if (KNOTE_DETACH(&so->so_rcv.sb_sel.si_note, kn)) - so->so_rcv.sb_flags &= ~SB_KNOTE; - socket_unlock(so, 1); + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_drained); + } + } } -/*ARGSUSED*/ -static int -filt_soread(struct knote *kn, long hint) +/* + * Return values tells if socket is still in extended background idle + */ +int +so_check_extended_bk_idle_time(struct socket *so) { - struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + int ret = 1; - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_lock(so, 1); + if ((so->so_flags1 & SOF1_EXTEND_BK_IDLE_INPROG)) { + SODEFUNCTLOG("%s[%d, %s]: so 0x%llx [%d,%d]\n", + __func__, proc_selfpid(), proc_best_name(current_proc()), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so)); + if (net_uptime() - so->so_extended_bk_start > + soextbkidlestat.so_xbkidle_time) { + so_stop_extended_bk_idle(so); - if (so->so_oobmark) { - if (kn->kn_flags & EV_OOBAND) { - kn->kn_data = so->so_rcv.sb_cc - so->so_oobmark; - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (1); - } - kn->kn_data = so->so_oobmark; - kn->kn_flags |= EV_OOBAND; - } else { - kn->kn_data = so->so_rcv.sb_cc; - if (so->so_state & SS_CANTRCVMORE) { - kn->kn_flags |= EV_EOF; - kn->kn_fflags = so->so_error; - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (1); + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_expired); + + ret = 0; + } else { + struct inpcb *inp = (struct inpcb *)so->so_pcb; + + inpcb_timer_sched(inp->inp_pcbinfo, INPCB_TIMER_LAZY); + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_resched); } } - if (so->so_state & SS_RCVATMARK) { - if (kn->kn_flags & EV_OOBAND) { - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (1); + return ret; +} + +void +resume_proc_sockets(proc_t p) +{ + if (p->p_ladvflag & P_LXBKIDLEINPROG) { + struct fileproc *fp; + struct socket *so; + + proc_fdlock(p); + fdt_foreach(fp, p) { + if (FILEGLOB_DTYPE(fp->fp_glob) != DTYPE_SOCKET) { + continue; + } + + so = (struct socket *)fp->fp_glob->fg_data; + (void) soresume(p, so, 0); } - kn->kn_flags |= EV_OOBAND; - } else if (kn->kn_flags & EV_OOBAND) { - kn->kn_data = 0; - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (0); + proc_fdunlock(p); + + OSBitAndAtomic(~P_LXBKIDLEINPROG, &p->p_ladvflag); } +} - if (so->so_error) { /* temporary udp error */ - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (1); +__private_extern__ int +so_set_recv_anyif(struct socket *so, int optval) +{ + int ret = 0; + + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) { + if (optval) { + sotoinpcb(so)->inp_flags |= INP_RECV_ANYIF; + } else { + sotoinpcb(so)->inp_flags &= ~INP_RECV_ANYIF; + } } - 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 ret; } -static void -filt_sowdetach(struct knote *kn) +__private_extern__ int +so_get_recv_anyif(struct socket *so) { - struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; - socket_lock(so, 1); + int ret = 0; - if(so->so_snd.sb_flags & SB_KNOTE) - if (KNOTE_DETACH(&so->so_snd.sb_sel.si_note, kn)) - so->so_snd.sb_flags &= ~SB_KNOTE; - socket_unlock(so, 1); + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) { + ret = (sotoinpcb(so)->inp_flags & INP_RECV_ANYIF) ? 1 : 0; + } + + return ret; } -/*ARGSUSED*/ -static int -filt_sowrite(struct knote *kn, long hint) +int +so_set_restrictions(struct socket *so, uint32_t vals) { - struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; - - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_lock(so, 1); + int nocell_old, nocell_new; + int noexpensive_old, noexpensive_new; + int noconstrained_old, noconstrained_new; - kn->kn_data = sbspace(&so->so_snd); - 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); + /* + * 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); + noconstrained_old = (so->so_restrictions & SO_RESTRICT_DENY_CONSTRAINED); + so->so_restrictions |= (vals & (SO_RESTRICT_DENY_IN | + SO_RESTRICT_DENY_OUT | SO_RESTRICT_DENY_CELLULAR | + SO_RESTRICT_DENY_EXPENSIVE | SO_RESTRICT_DENY_CONSTRAINED)); + nocell_new = (so->so_restrictions & SO_RESTRICT_DENY_CELLULAR); + noexpensive_new = (so->so_restrictions & SO_RESTRICT_DENY_EXPENSIVE); + noconstrained_new = (so->so_restrictions & SO_RESTRICT_DENY_CONSTRAINED); + + /* we can only set, not clear restrictions */ + if ((nocell_new - nocell_old) == 0 && + (noexpensive_new - noexpensive_old) == 0 && + (noconstrained_new - noconstrained_old) == 0) { + return 0; } - if (so->so_error) { /* temporary udp error */ - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (1); + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_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)); + } + if (noconstrained_new - noconstrained_old != 0) { + inp_set_noconstrained(sotoinpcb(so)); + } } - 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); + + if (SOCK_DOM(so) == PF_MULTIPATH) { + mptcp_set_restrictions(so); } - if ((hint & SO_FILT_HINT_LOCKED) == 0) - 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); + + return 0; } -/*ARGSUSED*/ -static int -filt_solisten(struct knote *kn, long hint) +uint32_t +so_get_restrictions(struct socket *so) { - struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; - int isempty; - - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_lock(so, 1); - kn->kn_data = so->so_qlen; - isempty = ! TAILQ_EMPTY(&so->so_comp); - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (isempty); + return so->so_restrictions & (SO_RESTRICT_DENY_IN | + SO_RESTRICT_DENY_OUT | + SO_RESTRICT_DENY_CELLULAR | SO_RESTRICT_DENY_EXPENSIVE); } - int -socket_lock(so, refcount) - struct socket *so; - int refcount; +so_set_effective_pid(struct socket *so, int epid, struct proc *p, boolean_t check_cred) { - int error = 0, lr_saved; + struct proc *ep = PROC_NULL; + int error = 0; - lr_saved = (unsigned int) __builtin_return_address(0); + /* pid 0 is reserved for kernel */ + if (epid == 0) { + error = EINVAL; + goto done; + } - if (so->so_proto->pr_lock) { - error = (*so->so_proto->pr_lock)(so, refcount, lr_saved); + /* + * 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; } - else { -#ifdef MORE_LOCKING_DEBUG - lck_mtx_assert(so->so_proto->pr_domain->dom_mtx, LCK_MTX_ASSERT_NOTOWNED); + + /* + * 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 (check_cred && (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)); + +#if defined(XNU_TARGET_OS_OSX) + if (ep->p_responsible_pid != so->e_pid) { + proc_t rp = proc_find(ep->p_responsible_pid); + if (rp != PROC_NULL) { + proc_getexecutableuuid(rp, so->so_ruuid, sizeof(so->so_ruuid)); + so->so_rpid = ep->p_responsible_pid; + proc_rele(rp); + } else { + uuid_clear(so->so_ruuid); + so->so_rpid = -1; + } + } #endif - lck_mtx_lock(so->so_proto->pr_domain->dom_mtx); - if (refcount) - so->so_usecount++; - so->lock_lr[so->next_lock_lr] = (void *)lr_saved; - so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX; + } + if (so->so_proto != NULL && so->so_proto->pr_update_last_owner != NULL) { + (*so->so_proto->pr_update_last_owner)(so, NULL, ep); + } +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)DEBUG_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)DEBUG_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 */ } - return(error); + if (ep != PROC_NULL) { + proc_rele(ep); + } + return error; } int -socket_unlock(so, refcount) - struct socket *so; - int refcount; +so_set_effective_uuid(struct socket *so, uuid_t euuid, struct proc *p, boolean_t check_cred) { - int error = 0, lr_saved; - lck_mtx_t * mutex_held; + uuid_string_t buf; + uuid_t uuid; + int error = 0; - lr_saved = (unsigned int) __builtin_return_address(0); + /* UUID must not be all-zeroes (reserved for kernel) */ + if (uuid_is_null(euuid)) { + error = EINVAL; + goto done; + } - if (so->so_proto == NULL) - panic("socket_unlock null so_proto so=%x\n", so); + /* + * 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 (so && so->so_proto->pr_unlock) - error = (*so->so_proto->pr_unlock)(so, refcount, lr_saved); - else { - mutex_held = so->so_proto->pr_domain->dom_mtx; -#ifdef MORE_LOCKING_DEBUG - lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); -#endif - so->unlock_lr[so->next_unlock_lr] = (void *)lr_saved; - so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; + /* Get the UUID of the issuing process */ + proc_getexecutableuuid(p, uuid, sizeof(uuid)); - if (refcount) { - if (so->so_usecount <= 0) - panic("socket_unlock: bad refcount so=%x value=%d\n", so, so->so_usecount); - so->so_usecount--; - if (so->so_usecount == 0) { - sofreelastref(so, 1); - } + /* + * 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 (check_cred && + (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; } - lck_mtx_unlock(mutex_held); } - return(error); + /* + * 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); + } + /* + * The following will clear the effective process name as it's the same + * as the real process + */ + if (so->so_proto != NULL && so->so_proto->pr_update_last_owner != NULL) { + (*so->so_proto->pr_update_last_owner)(so, NULL, NULL); + } +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)DEBUG_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)DEBUG_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; } -//### Called with socket locked, will unlock socket + void -sofree(so) - struct socket *so; +netpolicy_post_msg(uint32_t ev_code, struct netpolicy_event_data *ev_data, + uint32_t ev_datalen) { + struct kev_msg ev_msg; - lck_mtx_t * mutex_held; - if (so->so_proto->pr_getlock != NULL) - mutex_held = (*so->so_proto->pr_getlock)(so, 0); - else - mutex_held = so->so_proto->pr_domain->dom_mtx; - lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); - - sofreelastref(so, 0); + /* + * 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 -soreference(so) - struct socket *so; +socket_post_kev_msg(uint32_t ev_code, + struct kev_socket_event_data *ev_data, + uint32_t ev_datalen) { - socket_lock(so, 1); /* locks & take one reference on socket */ - socket_unlock(so, 0); /* unlock only */ + 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 -sodereference(so) - struct socket *so; +socket_post_kev_msg_closed(struct socket *so) { - socket_lock(so, 0); - socket_unlock(so, 1); + struct kev_socket_closed ev = {}; + struct sockaddr *socksa = NULL, *peersa = NULL; + int err; + + if ((so->so_flags1 & SOF1_WANT_KEV_SOCK_CLOSED) == 0) { + return; + } + 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); + } }