X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/39236c6e673c41db228275375ab7fdb0f837b292..d9a64523371fa019c4575bb400cbbc3a50ac9903:/bsd/kern/uipc_socket.c?ds=sidebyside diff --git a/bsd/kern/uipc_socket.c b/bsd/kern/uipc_socket.c index f7f3d5202..348afd442 100644 --- a/bsd/kern/uipc_socket.c +++ b/bsd/kern/uipc_socket.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 1998-2013 Apple Inc. All rights reserved. + * Copyright (c) 1998-2018 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -89,6 +89,7 @@ #include #include #include +#include #include #include #include @@ -97,9 +98,13 @@ #include #include #include +#include #include +#include #include #include +#include +#include #include #include #include @@ -110,18 +115,30 @@ #include #include #include +#include + #include #include +#include +#include #if CONFIG_MACF -#include #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 + /* TODO: this should be in a header file somewhere */ extern char *proc_name_address(void *p); @@ -142,36 +159,75 @@ static lck_mtx_t *so_cache_mtx; #include +static int filt_sorattach(struct knote *kn, struct kevent_internal_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_internal_s *kev); +static int filt_sorprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev); + +static int filt_sowattach(struct knote *kn, struct kevent_internal_s *kev); static void filt_sowdetach(struct knote *kn); static int filt_sowrite(struct knote *kn, long hint); +static int filt_sowtouch(struct knote *kn, struct kevent_internal_s *kev); +static int filt_sowprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev); + +static int filt_sockattach(struct knote *kn, struct kevent_internal_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_internal_s *kev); +static int filt_sockprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev); static int sooptcopyin_timeval(struct sockopt *, struct timeval *); static int sooptcopyout_timeval(struct sockopt *, const struct timeval *); -static struct filterops soread_filtops = { +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, }; -static struct filterops sowrite_filtops = { +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, }; -static struct filterops sock_filtops = { +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, }; +SYSCTL_DECL(_kern_ipc); + #define EVEN_MORE_LOCKING_DEBUG 0 + int socket_debug = 0; +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, ""); + static int socket_zone = M_SOCKET; so_gen_t so_gencnt; /* generation count for sockets */ @@ -183,13 +239,13 @@ MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); #define DBG_LAYER_OUT_BEG NETDBG_CODE(DBG_NETSOCK, 1) #define DBG_LAYER_OUT_END NETDBG_CODE(DBG_NETSOCK, 3) #define DBG_FNC_SOSEND NETDBG_CODE(DBG_NETSOCK, (4 << 8) | 1) +#define DBG_FNC_SOSEND_LIST NETDBG_CODE(DBG_NETSOCK, (4 << 8) | 3) #define DBG_FNC_SORECEIVE NETDBG_CODE(DBG_NETSOCK, (8 << 8)) +#define DBG_FNC_SORECEIVE_LIST NETDBG_CODE(DBG_NETSOCK, (8 << 8) | 3) #define DBG_FNC_SOSHUTDOWN NETDBG_CODE(DBG_NETSOCK, (9 << 8)) #define MAX_SOOPTGETM_SIZE (128 * MCLBYTES) -SYSCTL_DECL(_kern_ipc); - int somaxconn = SOMAXCONN; SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW | CTLFLAG_LOCKED, &somaxconn, 0, ""); @@ -226,6 +282,19 @@ 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, ""); @@ -238,48 +307,99 @@ int sorestrictrecv = 1; SYSCTL_INT(_kern_ipc, OID_AUTO, sorestrictrecv, CTLFLAG_RW | CTLFLAG_LOCKED, &sorestrictrecv, 0, "Enable inbound interface restrictions"); -/* - * Socket operation routines. - * These routines are called by the routines in - * sys_socket.c or from a system process, and - * implement the semantics of socket operations by - * switching out to the protocol specific routines. - */ +int 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"); + +#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 */ -/* sys_generic.c */ -extern void postevent(struct socket *, struct sockbuf *, int); -extern void evsofree(struct socket *); -extern int tcp_notsent_lowat_check(struct socket *so); extern struct inpcbinfo tcbinfo; /* TODO: these should be in header file */ extern int get_inpcb_str_size(void); extern int get_tcp_str_size(void); -static unsigned int sl_zone_size; /* size of sockaddr_list */ -static struct zone *sl_zone; /* zone for sockaddr_list */ - -static unsigned int se_zone_size; /* size of sockaddr_entry */ -static struct zone *se_zone; /* zone for sockaddr_entry */ - vm_size_t so_cache_zone_element_size; -static int sodelayed_copy(struct socket *, struct uio *, struct mbuf **, user_ssize_t *); +static int sodelayed_copy(struct socket *, struct uio *, struct mbuf **, + user_ssize_t *); static void cached_sock_alloc(struct socket **, int); static void cached_sock_free(struct socket *); +/* + * Maximum of extended background idle sockets per process + * Set to zero to disable further setting of the option + */ + +#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"); + +SYSCTL_UINT(_kern_ipc, OID_AUTO, extbkidlercvhiwat, CTLFLAG_RW | CTLFLAG_LOCKED, + &soextbkidlestat.so_xbkidle_rcvhiwat, 0, + "High water mark for extended background idle sockets"); + +SYSCTL_STRUCT(_kern_ipc, OID_AUTO, extbkidlestat, CTLFLAG_RD | CTLFLAG_LOCKED, + &soextbkidlestat, soextbkidlestat, ""); + +int so_set_extended_bk_idle(struct socket *, int); + + /* * SOTCDB_NO_DSCP is set by default, to prevent the networking stack from * setting the DSCP code on the packet based on the service class; see * for details. */ -__private_extern__ u_int32_t sotcdb = SOTCDB_NO_DSCP; +__private_extern__ u_int32_t sotcdb = 0; SYSCTL_INT(_kern_ipc, OID_AUTO, sotcdb, CTLFLAG_RW | CTLFLAG_LOCKED, &sotcdb, 0, ""); void socketinit(void) { + _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 (socketinit_done) { printf("socketinit: already called...\n"); return; @@ -312,29 +432,15 @@ socketinit(void) so_cache_zone_element_size = (vm_size_t)(sizeof (struct socket) + 4 + get_inpcb_str_size() + 4 + get_tcp_str_size()); - so_cache_zone = zinit(so_cache_zone_element_size, + so_cache_zone = zinit(so_cache_zone_element_size, (120000 * so_cache_zone_element_size), 8192, "socache zone"); zone_change(so_cache_zone, Z_CALLERACCT, FALSE); zone_change(so_cache_zone, Z_NOENCRYPT, TRUE); - sl_zone_size = sizeof (struct sockaddr_list); - if ((sl_zone = zinit(sl_zone_size, 1024 * sl_zone_size, 1024, - "sockaddr_list")) == NULL) { - panic("%s: unable to allocate sockaddr_list zone\n", __func__); - /* NOTREACHED */ - } - zone_change(sl_zone, Z_CALLERACCT, FALSE); - zone_change(sl_zone, Z_EXPAND, TRUE); - - se_zone_size = sizeof (struct sockaddr_entry); - if ((se_zone = zinit(se_zone_size, 1024 * se_zone_size, 1024, - "sockaddr_entry")) == NULL) { - panic("%s: unable to allocate sockaddr_entry zone\n", __func__); - /* NOTREACHED */ - } - zone_change(se_zone, Z_CALLERACCT, FALSE); - zone_change(se_zone, Z_EXPAND, TRUE); - + 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(); @@ -381,8 +487,8 @@ cached_sock_alloc(struct socket **so, int waitok) bzero((caddr_t)*so, sizeof (struct socket)); /* - * Define offsets for extra structures into our - * single block of memory. Align extra structures + * Define offsets for extra structures into our + * single block of memory. Align extra structures * on longword boundaries. */ @@ -400,7 +506,7 @@ cached_sock_alloc(struct socket **so, int waitok) (caddr_t)offset; } - (*so)->cached_in_sock_layer = true; + OSBitOrAtomic(SOF1_CACHED_IN_SOCK_LAYER, &(*so)->so_flags1); } static void @@ -443,6 +549,7 @@ so_update_last_owner_locked(struct socket *so, proc_t self) proc_getexecutableuuid(self, so->last_uuid, sizeof (so->last_uuid)); } + proc_pidoriginatoruuid(so->so_vuuid, sizeof(so->so_vuuid)); } } @@ -453,6 +560,17 @@ so_update_policy(struct socket *so) (void) inp_update_policy(sotoinpcb(so)); } +#if NECP +static void +so_update_necp_policy(struct socket *so, struct sockaddr *override_local_addr, + struct sockaddr *override_remote_addr) +{ + if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) + inp_update_necp_policy(sotoinpcb(so), override_local_addr, + override_remote_addr, 0); +} +#endif /* NECP */ + boolean_t so_cache_timer(void) { @@ -467,7 +585,7 @@ so_cache_timer(void) while (!STAILQ_EMPTY(&so_cache_head)) { VERIFY(cached_sock_count > 0); p = STAILQ_FIRST(&so_cache_head); - if ((so_cache_time - p->cache_timestamp) < + if ((so_cache_time - p->cache_timestamp) < SO_CACHE_TIME_LIMIT) break; @@ -511,8 +629,14 @@ soalloc(int waitok, int dom, int type) bzero(so, sizeof (*so)); } if (so != NULL) { - so->so_gencnt = ++so_gencnt; + so->so_gencnt = OSIncrementAtomic64((SInt64 *)&so_gencnt); so->so_zone = socket_zone; + + /* + * Increment the socket allocation statistics + */ + INC_ATOMIC_INT64_LIM(net_api_stats.nas_socket_alloc_total); + #if CONFIG_MACF_SOCKET /* Convert waitok to M_WAITOK/M_NOWAIT for MAC Framework. */ if (mac_socket_label_init(so, !waitok) != 0) { @@ -560,19 +684,48 @@ socreate_internal(int dom, struct socket **aso, int type, int proto, if (so == NULL) return (ENOBUFS); + 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 (flags & SOCF_ASYNC) so->so_state |= SS_NBIO; -#if MULTIPATH - if (flags & SOCF_MP_SUBFLOW) { - /* - * A multipath subflow socket is used internally in the kernel, - * therefore it does not have a file desciptor associated by - * default. - */ - so->so_state |= SS_NOFDREF; - so->so_flags |= SOF_MP_SUBFLOW; - } -#endif /* MULTIPATH */ TAILQ_INIT(&so->so_incomp); TAILQ_INIT(&so->so_comp); @@ -580,6 +733,7 @@ socreate_internal(int dom, struct socket **aso, int type, int proto, 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); @@ -617,6 +771,7 @@ socreate_internal(int dom, struct socket **aso, int type, int proto, * 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); @@ -637,7 +792,8 @@ socreate_internal(int dom, struct socket **aso, int type, int proto, * If this thread or task is marked to create backgrounded sockets, * mark the socket as background. */ - if (proc_get_effective_thread_policy(current_thread(), TASK_POLICY_NEW_SOCKETS_BG)) { + if (proc_get_effective_thread_policy(current_thread(), + TASK_POLICY_NEW_SOCKETS_BG)) { socket_set_traffic_mgt_flags(so, TRAFFIC_MGT_SO_BACKGROUND); so->so_background_thread = current_thread(); } @@ -656,6 +812,17 @@ socreate_internal(int dom, struct socket **aso, int type, int proto, 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); @@ -733,20 +900,24 @@ sobindlock(struct socket *so, struct sockaddr *nam, int dolock) if (dolock) socket_lock(so, 1); - VERIFY(so->so_usecount > 1); so_update_last_owner_locked(so, p); so_update_policy(so); +#if NECP + so_update_necp_policy(so, nam, NULL); +#endif /* NECP */ + /* * If this is a bind request on a socket that has been marked * as inactive, reject it now before we go any further. */ if (so->so_flags & SOF_DEFUNCT) { error = EINVAL; - SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] (%d)\n", - __func__, proc_pid(p), (uint64_t)VM_KERNEL_ADDRPERM(so), - SOCK_DOM(so), SOCK_TYPE(so), error)); + 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; } @@ -773,6 +944,10 @@ sodealloc(struct socket *so) /* Remove any filters */ sflt_termsock(so); +#if CONTENT_FILTER + cfil_sock_detach(so); +#endif /* CONTENT_FILTER */ + /* Delete the state allocated for msg queues on a socket */ if (so->so_flags & SOF_ENABLE_MSGS) { FREE(so->so_msg_state, M_TEMP); @@ -780,13 +955,13 @@ sodealloc(struct socket *so) } VERIFY(so->so_msg_state == NULL); - so->so_gencnt = ++so_gencnt; + so->so_gencnt = OSIncrementAtomic64((SInt64 *)&so_gencnt); #if CONFIG_MACF_SOCKET mac_socket_label_destroy(so); #endif /* MAC_SOCKET */ - if (so->cached_in_sock_layer) { + if (so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) { cached_sock_free(so); } else { FREE_ZONE(so, sizeof (*so), so->so_zone); @@ -823,6 +998,10 @@ solisten(struct socket *so, int backlog) 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; @@ -842,10 +1021,11 @@ solisten(struct socket *so, int backlog) (so->so_flags & SOF_DEFUNCT)) { error = EINVAL; if (so->so_flags & SOF_DEFUNCT) { - SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] " + SODEFUNCTLOG("%s[%d, %s]: defunct so 0x%llx [%d,%d] " "(%d)\n", __func__, proc_pid(p), - (uint64_t)VM_KERNEL_ADDRPERM(so), - SOCK_DOM(so), SOCK_TYPE(so), error)); + proc_best_name(p), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error); } goto out; } @@ -890,6 +1070,75 @@ out: 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) { @@ -902,15 +1151,34 @@ sofreelastref(struct socket *so, int dealloc) selthreadclear(&so->so_rcv.sb_sel); so->so_rcv.sb_flags &= ~(SB_SEL|SB_UPCALL); so->so_snd.sb_flags &= ~(SB_SEL|SB_UPCALL); - so->so_event = NULL; + 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 @@ -921,16 +1189,15 @@ sofreelastref(struct socket *so, int dealloc) selthreadclear(&so->so_rcv.sb_sel); so->so_rcv.sb_flags &= ~(SB_SEL|SB_UPCALL); so->so_snd.sb_flags &= ~(SB_SEL|SB_UPCALL); - so->so_event = NULL; - socket_unlock(head, 1); + 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); } sowflush(so); sorflush(so); @@ -943,8 +1210,8 @@ sofreelastref(struct socket *so, int dealloc) /* 3932268: disable upcall */ so->so_rcv.sb_flags &= ~SB_UPCALL; - so->so_snd.sb_flags &= ~SB_UPCALL; - so->so_event = NULL; + so->so_snd.sb_flags &= ~(SB_UPCALL|SB_SNDBYTE_CNT); + so->so_event = sonullevent; if (dealloc) sodealloc(so); @@ -956,10 +1223,10 @@ soclose_wait_locked(struct socket *so) lck_mtx_t *mutex_held; if (so->so_proto->pr_getlock != NULL) - mutex_held = (*so->so_proto->pr_getlock)(so, 0); + 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); + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); /* * Double check here and return if there's no outstanding upcall; @@ -970,9 +1237,10 @@ soclose_wait_locked(struct socket *so) 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); + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); so->so_flags &= ~SOF_CLOSEWAIT; } @@ -985,7 +1253,6 @@ int soclose_locked(struct socket *so) { int error = 0; - lck_mtx_t *mutex_held; struct timespec ts; if (so->so_usecount == 0) { @@ -998,9 +1265,26 @@ soclose_locked(struct socket *so) 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, *sonext; - int socklock = 0; + int persocklock = 0; + int incomp_overflow_only; /* * We do not want new connection to be added @@ -1008,10 +1292,19 @@ soclose_locked(struct socket *so) */ so->so_options &= ~SO_ACCEPTCONN; - for (sp = TAILQ_FIRST(&so->so_incomp); - sp != NULL; sp = sonext) { - sonext = TAILQ_NEXT(sp, so_list); + /* + * 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 @@ -1022,53 +1315,72 @@ soclose_locked(struct socket *so) if (sp->so_flags & SOF_OVERFLOW) continue; - if (so->so_proto->pr_getlock != NULL) { - /* - * Lock ordering for consistency with the - * rest of the stack, we lock the socket - * first and then grabb the head. - */ - socket_unlock(so, 0); + if (persocklock != 0) socket_lock(sp, 1); - socket_lock(so, 0); - socklock = 1; - } - - TAILQ_REMOVE(&so->so_incomp, sp, so_list); - so->so_incqlen--; + /* + * 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 (socklock) + if (persocklock != 0) socket_unlock(sp, 1); } - 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); - } 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); + } else { + panic("%s sp %p in so_comp but !SS_COMP", + __func__, sp); } - if (so->so_proto->pr_getlock != NULL) { + 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 == NULL) { @@ -1083,11 +1395,13 @@ soclose_locked(struct socket *so) goto drop; } if (so->so_options & SO_LINGER) { + lck_mtx_t *mutex_held; + if ((so->so_state & SS_ISDISCONNECTING) && (so->so_state & SS_NBIO)) goto drop; if (so->so_proto->pr_getlock != NULL) - mutex_held = (*so->so_proto->pr_getlock)(so, 0); + 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) { @@ -1114,14 +1428,6 @@ drop: /* NOTREACHED */ } if (so->so_pcb != NULL && !(so->so_flags & SOF_PCBCLEARING)) { - /* - * Let NetworkStatistics know this PCB is going away - * before we detach it. - */ - if (nstat_collect && - (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6)) - nstat_pcb_detach(so->so_pcb); - int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so); if (error == 0) error = error2; @@ -1138,15 +1444,13 @@ discard: } so->so_state |= SS_NOFDREF; - if (so->so_flags & SOF_MP_SUBFLOW) - so->so_flags &= ~SOF_MP_SUBFLOW; - if ((so->so_flags & SOF_KNOTE) != 0) KNOTE(&so->so_klist, SO_FILT_HINT_LOCKED); atomic_add_32(&so->so_proto->pr_domain->dom_refs, -1); evsofree(so); + VERIFY(so->so_usecount > 0); so->so_usecount--; sofree(so); return (error); @@ -1190,7 +1494,7 @@ soabort(struct socket *so) 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); + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); #endif if ((so->so_flags & SOF_ABORTED) == 0) { @@ -1214,6 +1518,9 @@ soacceptlock(struct socket *so, struct sockaddr **nam, int dolock) so_update_last_owner_locked(so, PROC_NULL); so_update_policy(so); +#if NECP + so_update_necp_policy(so, NULL, NULL); +#endif /* NECP */ if ((so->so_state & SS_NOFDREF) == 0) panic("soaccept: !NOFDREF"); @@ -1232,11 +1539,10 @@ soaccept(struct socket *so, struct sockaddr **nam) } int -soacceptfilter(struct socket *so) +soacceptfilter(struct socket *so, struct socket *head) { struct sockaddr *local = NULL, *remote = NULL; int error = 0; - struct socket *head = so->so_head; /* * Hold the lock even if this socket has not been made visible @@ -1246,8 +1552,7 @@ soacceptfilter(struct socket *so) socket_lock(so, 1); if (sogetaddr_locked(so, &remote, 1) != 0 || sogetaddr_locked(so, &local, 0) != 0) { - so->so_state &= ~(SS_NOFDREF | SS_COMP); - so->so_head = NULL; + so->so_state &= ~SS_NOFDREF; socket_unlock(so, 1); soclose(so); /* Out of resources; try it again next time */ @@ -1275,8 +1580,7 @@ soacceptfilter(struct socket *so) * the following is done while holding the lock since * the socket has been exposed to the filter(s) earlier. */ - so->so_state &= ~(SS_NOFDREF | SS_COMP); - so->so_head = NULL; + so->so_state &= ~SS_NOFDREF; socket_unlock(so, 1); soclose(so); /* Propagate socket filter's error code to the caller */ @@ -1315,6 +1619,10 @@ soconnectlock(struct socket *so, struct sockaddr *nam, int dolock) 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. @@ -1322,10 +1630,11 @@ soconnectlock(struct socket *so, struct sockaddr *nam, int dolock) if ((so->so_options & SO_ACCEPTCONN) || (so->so_flags & SOF_DEFUNCT)) { error = EOPNOTSUPP; if (so->so_flags & SOF_DEFUNCT) { - SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] " + SODEFUNCTLOG("%s[%d, %s]: defunct so 0x%llx [%d,%d] " "(%d)\n", __func__, proc_pid(p), - (uint64_t)VM_KERNEL_ADDRPERM(so), - SOCK_DOM(so), SOCK_TYPE(so), error)); + proc_best_name(p), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error); } if (dolock) socket_unlock(so, 1); @@ -1399,13 +1708,16 @@ soconnect2(struct socket *so1, struct socket *so2) } int -soconnectxlocked(struct socket *so, struct sockaddr_list **src_sl, - struct sockaddr_list **dst_sl, struct proc *p, uint32_t ifscope, - associd_t aid, connid_t *pcid, uint32_t flags, void *arg, - uint32_t arglen) +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. @@ -1413,10 +1725,11 @@ soconnectxlocked(struct socket *so, struct sockaddr_list **src_sl, if ((so->so_options & SO_ACCEPTCONN) || (so->so_flags & SOF_DEFUNCT)) { error = EOPNOTSUPP; if (so->so_flags & SOF_DEFUNCT) { - SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] " + SODEFUNCTLOG("%s[%d, %s]: defunct so 0x%llx [%d,%d] " "(%d)\n", __func__, proc_pid(p), - (uint64_t)VM_KERNEL_ADDRPERM(so), - SOCK_DOM(so), SOCK_TYPE(so), error)); + proc_best_name(p), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), + SOCK_DOM(so), SOCK_TYPE(so), error); } return (error); } @@ -1440,14 +1753,16 @@ soconnectxlocked(struct socket *so, struct sockaddr_list **src_sl, * Run connect filter before calling protocol: * - non-blocking connect returns before completion; */ - error = sflt_connectxout(so, dst_sl); + 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_sl, dst_sl, p, ifscope, aid, pcid, - flags, arg, arglen); + (so, src, dst, p, ifscope, aid, pcid, + flags, arg, arglen, auio, bytes_written); } } @@ -1489,7 +1804,7 @@ sodisconnect(struct socket *so) } int -sodisconnectxlocked(struct socket *so, associd_t aid, connid_t cid) +sodisconnectxlocked(struct socket *so, sae_associd_t aid, sae_connid_t cid) { int error; @@ -1510,7 +1825,7 @@ sodisconnectxlocked(struct socket *so, associd_t aid, connid_t cid) } int -sodisconnectx(struct socket *so, associd_t aid, connid_t cid) +sodisconnectx(struct socket *so, sae_associd_t aid, sae_connid_t cid) { int error; @@ -1520,12 +1835,6 @@ sodisconnectx(struct socket *so, associd_t aid, connid_t cid) return (error); } -int -sopeelofflocked(struct socket *so, associd_t aid, struct socket **psop) -{ - return ((*so->so_proto->pr_usrreqs->pru_peeloff)(so, aid, psop)); -} - #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) /* @@ -1579,15 +1888,28 @@ restart: if (so->so_flags & SOF_DEFUNCT) { defunct: error = EPIPE; - SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] (%d)\n", - __func__, proc_selfpid(), (uint64_t)VM_KERNEL_ADDRPERM(so), - SOCK_DOM(so), SOCK_TYPE(so), error)); + 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) - return (EPIPE); - + 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; @@ -1596,14 +1918,17 @@ defunct: if ((so->so_state & SS_ISCONNECTED) == 0) { if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { - if ((so->so_state & SS_ISCONFIRMING) == 0 && - !(resid == 0 && clen != 0)) + if (((so->so_state & SS_ISCONFIRMING) == 0) && + (resid != 0 || clen == 0) && + !(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); } } + if (so->so_flags & SOF_ENABLE_MSGS) space = msgq_sbspace(so, control); else @@ -1616,8 +1941,21 @@ defunct: return (EMSGSIZE); if ((space < resid + clen && - (atomic || space < (int32_t)so->so_snd.sb_lowat || space < 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); @@ -1701,12 +2039,14 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, { struct mbuf **mp; struct mbuf *m, *freelist = NULL; - user_ssize_t space, len, resid; + 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(); struct mbuf *control_copy = NULL; + uint16_t headroom = 0; + boolean_t en_tracing = FALSE; if (uio != NULL) resid = uio_resid(uio); @@ -1717,13 +2057,40 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, so->so_snd.sb_cc, so->so_snd.sb_lowat, so->so_snd.sb_hiwat); socket_lock(so, 1); - so_update_last_owner_locked(so, p); - so_update_policy(so); + + /* + * 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; - socket_unlock(so, 1); - goto out; + goto out_locked; } /* @@ -1736,14 +2103,13 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, * Usually, MSG_EOR isn't used on SOCK_STREAM type sockets. * But it will be used by sockets doing message delivery. * - * Note: We limit resid to be a positive 32 bits value as we use + * Note: We limit resid to be a positive int value as we use * imin() to set bytes_to_copy -- radr://14558484 */ - if ((int32_t)resid < 0 || (so->so_type == SOCK_STREAM && + if (resid < 0 || resid > INT_MAX || (so->so_type == SOCK_STREAM && !(so->so_flags & SOF_ENABLE_MSGS) && (flags & MSG_EOR))) { error = EINVAL; - socket_unlock(so, 1); - goto out; + goto out_locked; } dontroute = (flags & MSG_DONTROUTE) && @@ -1754,11 +2120,14 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, if (control != NULL) clen = control->m_len; + if (soreserveheadroom != 0) + headroom = so->so_pktheadroom; + do { error = sosendcheck(so, addr, resid, clen, atomic, flags, &sblocked, control); if (error) - goto release; + goto out_locked; mp = ⊤ if (so->so_flags & SOF_ENABLE_MSGS) @@ -1779,14 +2148,27 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 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 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 @@ -1795,7 +2177,8 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, */ jumbocl = sosendjcl && njcl > 0 && ((so->so_flags & SOF_MULTIPAGES) || - sosendjcl_ignore_capab); + sosendjcl_ignore_capab) && + bigcl; socket_unlock(so, 0); @@ -1819,12 +2202,12 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, * haven't yet consumed. */ if (freelist == NULL && - bytes_to_copy > MBIGCLBYTES && + bytes_to_alloc > MBIGCLBYTES && jumbocl) { num_needed = - bytes_to_copy / M16KCLBYTES; + bytes_to_alloc / M16KCLBYTES; - if ((bytes_to_copy - + if ((bytes_to_alloc - (num_needed * M16KCLBYTES)) >= MINCLSIZE) num_needed++; @@ -1841,11 +2224,12 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, } if (freelist == NULL && - bytes_to_copy > MCLBYTES) { + bytes_to_alloc > MCLBYTES && + bigcl) { num_needed = - bytes_to_copy / MBIGCLBYTES; + bytes_to_alloc / MBIGCLBYTES; - if ((bytes_to_copy - + if ((bytes_to_alloc - (num_needed * MBIGCLBYTES)) >= MINCLSIZE) num_needed++; @@ -1861,12 +2245,34 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, */ } - if (freelist == NULL && - bytes_to_copy > 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_copy / MCLBYTES; + bytes_to_alloc / MCLBYTES; - if ((bytes_to_copy - + if ((bytes_to_alloc - (num_needed * MCLBYTES)) >= MINCLSIZE) num_needed++; @@ -1881,7 +2287,20 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, * 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 == NULL) MGETHDR(freelist, @@ -1893,7 +2312,7 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, if (freelist == NULL) { error = ENOBUFS; socket_lock(so, 0); - goto release; + goto out_locked; } /* * For datagram protocols, @@ -1911,12 +2330,13 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, m->m_next = NULL; if ((m->m_flags & M_EXT)) - mlen = m->m_ext.ext_size; + mlen = m->m_ext.ext_size - + M_LEADINGSPACE(m); else if ((m->m_flags & M_PKTHDR)) mlen = - MHLEN - m_leadingspace(m); + MHLEN - M_LEADINGSPACE(m); else - mlen = MLEN; + mlen = MLEN - M_LEADINGSPACE(m); len = imin(mlen, bytes_to_copy); chainlength += len; @@ -1948,7 +2368,7 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, socket_lock(so, 0); if (error) - goto release; + goto out_locked; } if (flags & (MSG_HOLD|MSG_SEND)) { @@ -1968,46 +2388,60 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, so->so_tail = mb1; if (flags & MSG_HOLD) { top = NULL; - goto release; + goto out_locked; } top = so->so_temp; } if (dontroute) so->so_options |= SO_DONTROUTE; - /* Compute flags here, for pru_send and NKEs */ + /* + * 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. + */ sendflags = (flags & MSG_OOB) ? PRUS_OOB : - /* - * If the user set MSG_EOF, the protocol - * understands this flag and nothing left to - * send then use PRU_SEND_EOF instead of PRU_SEND. - */ ((flags & MSG_EOF) && - (so->so_proto->pr_flags & PR_IMPLOPCL) && - (resid <= 0)) ? PRUS_EOF : - /* If there is more to send set PRUS_MORETOCOME */ - (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; + (so->so_proto->pr_flags & PR_IMPLOPCL) && + (resid <= 0)) ? PRUS_EOF : + /* If there is more to send set PRUS_MORETOCOME */ + (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; - /* - * Socket filter processing - */ - error = sflt_data_out(so, addr, &top, - &control, (sendflags & MSG_OOB) ? - sock_data_filt_flag_oob : 0); - if (error) { - if (error == EJUSTRETURN) { - error = 0; - clen = 0; - control = NULL; - top = NULL; + if ((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; } - - 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 (so->so_flags & SOF_ENABLE_MSGS) { /* * Make a copy of control mbuf, @@ -2031,16 +2465,15 @@ sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, top = NULL; mp = ⊤ if (error) - goto release; + goto out_locked; } while (resid && space > 0); } while (resid); -release: +out_locked: if (sblocked) sbunlock(&so->so_snd, FALSE); /* will unlock socket */ else socket_unlock(so, 1); -out: if (top != NULL) m_freem(top); if (control != NULL) @@ -2050,681 +2483,1158 @@ out: if (control_copy != NULL) m_freem(control_copy); - KERNEL_DEBUG(DBG_FNC_SOSEND | DBG_FUNC_END, so, resid, so->so_snd.sb_cc, - space, error); + 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, *control_end; + + 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)); + + /* + * 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; +} + /* - * 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. + * Supported only connected sockets (no address) without ancillary data + * (control mbuf) for atomic protocols */ int -soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, - struct mbuf **mp0, struct mbuf **controlp, int *flagsp) +sosend_list(struct socket *so, struct uio **uioarray, u_int uiocnt, int flags) { - struct mbuf *m, **mp, *ml = NULL; - struct mbuf *nextrecord, *free_list; - int flags, error, offset; - user_ssize_t len; - struct protosw *pr = so->so_proto; - int moff, type =0; - user_ssize_t orig_resid = uio_resid(uio); - user_ssize_t delayed_copy_len; - int can_delay; - int need_event; + struct 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_SORECEIVE | DBG_FUNC_START, so, uio_resid(uio), - so->so_rcv.sb_cc, so->so_rcv.sb_lowat, so->so_rcv.sb_hiwat); + KERNEL_DEBUG((DBG_FNC_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); + + /* + * 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("%s: so=%x no other reference on socket\n", __func__, so); - /* NOTREACHED */ - } -#endif - mp = mp0; - if (psa != NULL) - *psa = NULL; - if (controlp != NULL) - *controlp = NULL; - if (flagsp != NULL) - flags = *flagsp &~ MSG_EOR; - else - flags = 0; +#if 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); + + error = sosendcheck(so, NULL, resid, 0, atomic, flags, + &sblocked, NULL); + if (error) + goto release; /* - * If a recv attempt is made on a previously-accepted socket - * that has been marked as inactive (disconnected), reject - * the request. + * Use big 4 KB clusters when the outgoing interface does not prefer + * 2 KB clusters */ - if (so->so_flags & SOF_DEFUNCT) { - struct sockbuf *sb = &so->so_rcv; + bigcl = !(so->so_flags1 & SOF1_IF_2KCL) || sosendbigcl_ignore_capab; + + if (soreserveheadroom != 0) + headroom = so->so_pktheadroom; + + do { + int i; + int num_needed = 0; + int chainlength; + size_t maxpktlen = 0; + int bytes_to_alloc; + + if (sosendminchain > 0) + chainlength = 0; + else + chainlength = sosendmaxchain; + + socket_unlock(so, 0); - error = ENOTCONN; - SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] (%d)\n", - __func__, proc_pid(p), (uint64_t)VM_KERNEL_ADDRPERM(so), - SOCK_DOM(so), SOCK_TYPE(so), error)); /* - * This socket should have been disconnected and flushed - * prior to being returned from sodefunct(); there should - * be no data on its receive list, so panic otherwise. + * Find a set of uio that fit in a reasonable number + * of mbuf packets */ - if (so->so_state & SS_DEFUNCT) - sb_empty_assert(sb, __func__); - socket_unlock(so, 1); - return (error); - } + for (i = uiofirst; i < uiocnt; i++) { + struct uio *auio = uioarray[i]; - /* - * 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); + len = uio_resid(auio); + + /* Do nothing for empty messages */ + if (len == 0) + continue; + + num_needed += 1; + uiolast += 1; + + if (len > maxpktlen) + maxpktlen = len; + + chainlength += len; + if (chainlength > sosendmaxchain) + break; } - 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); + /* + * Nothing left to send + */ + if (num_needed == 0) { + socket_lock(so, 0); + break; + } + /* + * Allocate buffer large enough to include headroom space for + * network and link header + * + */ + 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; + } + /* + * Copy each uio of the set into its own mbuf packet + */ + for (i = uiofirst, m = freelist; + i < uiolast && m != NULL; + i++) { + int bytes_to_copy; + struct mbuf *n; + struct uio *auio = uioarray[i]; + + bytes_to_copy = uio_resid(auio); + + /* Do nothing for empty messages */ + if (bytes_to_copy == 0) + continue; + /* + * 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 + */ + error = uiomove(mtod(n, caddr_t), + len, auio); + if (error != 0) + break; + n->m_len = len; + m->m_pkthdr.len += len; + + VERIFY(m->m_pkthdr.len <= maxpktlen); + + bytes_to_copy -= len; + resid -= len; + } + if (m->m_pkthdr.len == 0) { + printf( + "%s:%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); -bad: - if (m != NULL) - m_freem(m); - if ((so->so_options & SO_WANTOOBFLAG) != 0) { - if (error == EWOULDBLOCK || error == EINVAL) { + 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; + /* - * Let's try to get normal data: - * EWOULDBLOCK: out-of-band data not - * receive yet. EINVAL: out-of-band data - * already read. + * Socket filter processing */ - error = 0; - goto nooob; - } else if (error == 0 && flagsp != NULL) { - *flagsp |= MSG_OOB; + 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; + } +#endif /* CONTENT_FILTER */ + /* + * Remove packet from the list when + * swallowed by a filter + */ + if (error == EJUSTRETURN) { + error = 0; + if (prevnextp != NULL) + *prevnextp = nextpkt; + else + top = nextpkt; + } + + m = nextpkt; + if (m != NULL) + prevnextp = &m->m_nextpkt; } } + if (top != NULL) + error = (*so->so_proto->pr_usrreqs->pru_send_list) + (so, 0, top, NULL, NULL, p); + + if (dontroute) + so->so_options &= ~SO_DONTROUTE; + + 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); - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error, - 0, 0, 0, 0); +out: + if (top != NULL) + m_freem(top); + if (freelist != NULL) + m_freem_list(freelist); - return (error); - } -nooob: - if (mp != NULL) - *mp = NULL; - if (so->so_state & SS_ISCONFIRMING && uio_resid(uio)) - (*pr->pr_usrreqs->pru_rcvd)(so, 0); + KERNEL_DEBUG(DBG_FNC_SOSEND_LIST | DBG_FUNC_END, so, resid, + so->so_snd.sb_cc, 0, error); - free_list = NULL; - delayed_copy_len = 0; -restart: -#ifdef MORE_LOCKING_DEBUG - if (so->so_usecount <= 1) - printf("soreceive: sblock so=%p ref=%d on socket\n", - so, so->so_usecount); -#endif - /* - * See if the socket has been closed (SS_NOFDREF|SS_CANTRCVMORE) - * and if so just return to the caller. This could happen when - * soreceive() is called by a socket upcall function during the - * time the socket is freed. The socket buffer would have been - * locked across the upcall, therefore we cannot put this thread - * to sleep (else we will deadlock) or return EWOULDBLOCK (else - * we may livelock), because the lock on the socket buffer will - * only be released when the upcall routine returns to its caller. - * Because the socket has been officially closed, there can be - * no further read on it. - * - * A multipath subflow socket would have its SS_NOFDREF set by - * default, so check for SOF_MP_SUBFLOW socket flag; when the - * socket is closed for real, SOF_MP_SUBFLOW would be cleared. - */ - if ((so->so_state & (SS_NOFDREF | SS_CANTRCVMORE)) == - (SS_NOFDREF | SS_CANTRCVMORE) && !(so->so_flags & SOF_MP_SUBFLOW)) { - socket_unlock(so, 1); - return (0); - } + return (error); +} - 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); - 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; - m = so->so_rcv.sb_mb; + KASSERT(m->m_type == MT_SONAME, ("receive 1a")); +#if CONFIG_MACF_SOCKET_SUBSET /* - * 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. + * Call the MAC framework for policy checking if we're in + * the user process context and the socket isn't connected. */ - if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && - so->so_rcv.sb_cc < uio_resid(uio)) && - (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || - ((flags & MSG_WAITALL) && uio_resid(uio) <= so->so_rcv.sb_hiwat)) && - m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { + if (p != kernproc && !(so->so_state & SS_ISCONNECTED)) { + struct mbuf *m0 = m; /* - * 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. + * 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. */ - SB_MB_CHECK(&so->so_rcv); + 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 (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 (m != NULL) - goto dontblock; - else - goto release; + 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; } - 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)) { + socket_lock(so, 0); + /* + * If the socket has been defunct'd, drop it. + */ + if (so->so_flags & SOF_DEFUNCT) { + m_freem(m); error = ENOTCONN; - goto release; - } - 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 - - 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); - return (error); + goto done; } - 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) { - 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. + * 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. */ - if (p != kernproc && !(so->so_state & SS_ISCONNECTED)) { - struct mbuf *m0 = m; - /* - * Dequeue this record (temporarily) from the receive - * list since we're about to drop the socket's lock - * where a new record may arrive and be appended to - * the list. Upon MAC policy failure, the record - * will be freed. Otherwise, we'll add it back to - * the head of the list. We cannot rely on SB_LOCK - * because append operation uses the socket's lock. - */ - do { - m->m_nextpkt = NULL; - sbfree(&so->so_rcv, m); - m = m->m_next; - } while (m != NULL); - m = m0; - so->so_rcv.sb_mb = nextrecord; - SB_EMPTY_FIXUP(&so->so_rcv); - SBLASTRECORDCHK(&so->so_rcv, "soreceive 1a"); - SBLASTMBUFCHK(&so->so_rcv, "soreceive 1a"); - socket_unlock(so, 0); - if (mac_socket_check_received(proc_ucred(p), so, - mtod(m, struct sockaddr *)) != 0) { - /* - * MAC policy failure; free this record and - * process the next record (or block until - * one is available). We have adjusted sb_cc - * and sb_mbcnt above so there is no need to - * call sbfree() again. - */ - do { - m = m_free(m); - } while (m != NULL); - /* - * Clear SB_LOCK but don't unlock the socket. - * Process the next record or wait for one. - */ - socket_lock(so, 0); - sbunlock(&so->so_rcv, TRUE); /* stay locked */ - goto restart; - } - socket_lock(so, 0); - /* - * If the socket has been defunct'd, drop it. - */ - if (so->so_flags & SOF_DEFUNCT) { - m_freem(m); - error = ENOTCONN; - goto release; - } - /* - * Re-adjust the socket receive list and re-enqueue - * the record in front of any packets which may have - * been appended while we dropped the lock. - */ - for (m = m0; m->m_next != NULL; m = m->m_next) - sballoc(&so->so_rcv, m); + for (m = m0; m->m_next != NULL; m = m->m_next) 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"); + sballoc(&so->so_rcv, m); + if (so->so_rcv.sb_mb == NULL) { + so->so_rcv.sb_lastrecord = m0; + so->so_rcv.sb_mbtail = m; } + m = m0; + nextrecord = m->m_nextpkt = so->so_rcv.sb_mb; + so->so_rcv.sb_mb = m; + SBLASTRECORDCHK(&so->so_rcv, "soreceive 1b"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive 1b"); + } #endif /* CONFIG_MACF_SOCKET_SUBSET */ - orig_resid = 0; - if (psa != NULL) { - *psa = dup_sockaddr(mtod(m, struct sockaddr *), - mp0 == NULL); - if ((*psa == NULL) && (flags & MSG_NEEDSA)) { - error = EWOULDBLOCK; - goto release; - } + 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; + } + 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 { - sbfree(&so->so_rcv, m); - if (m->m_next == NULL && so->so_rcv.sb_cc != 0) { - panic("%s: about to create invalid socketbuf", - __func__); - /* NOTREACHED */ - } - MFREE(m, so->so_rcv.sb_mb); - m = so->so_rcv.sb_mb; - if (m != NULL) { - m->m_nextpkt = nextrecord; - } else { - so->so_rcv.sb_mb = nextrecord; - SB_EMPTY_FIXUP(&so->so_rcv); - } + so->so_rcv.sb_mb = nextrecord; + SB_EMPTY_FIXUP(&so->so_rcv); } } +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; /* - * 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. + * 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. */ - if (m != NULL && m->m_type == MT_CONTROL) { - struct mbuf *cm = NULL, *cmn; - struct mbuf **cme = &cm; - struct sockbuf *sb_rcv = &so->so_rcv; - struct mbuf **msgpcm = NULL; - - /* - * Externalizing the control messages would require us to - * drop the socket's lock below. Once we re-acquire the - * lock, the mbuf chain might change. In order to preserve - * consistency, we unlink all control messages from the - * first mbuf chain in one shot and link them separately - * onto a different chain. - */ - do { - if (flags & MSG_PEEK) { - if (controlp != NULL) { - if (*controlp == NULL) { - msgpcm = controlp; - } - *controlp = m_copy(m, 0, m->m_len); + do { + if (flags & MSG_PEEK) { + if (controlp != NULL) { + if (*controlp == NULL) { + msgpcm = controlp; + } + *controlp = m_copy(m, 0, m->m_len); - /* - * If we failed to allocate an mbuf, - * release any previously allocated - * mbufs for control data. Return - * an error. Keep the mbufs in the - * socket as this is using - * MSG_PEEK flag. - */ - if (*controlp == NULL) { - m_freem(*msgpcm); - error = ENOBUFS; - goto release; - } - controlp = &(*controlp)->m_next; + /* + * 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; } - 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; + controlp = &(*controlp)->m_next; } - } while (m != NULL && m->m_type == MT_CONTROL); + 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; + } + } 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; + 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"); + SBLASTRECORDCHK(&so->so_rcv, "soreceive ctl"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive ctl"); - while (cm != NULL) { - int cmsg_type; + while (cm != NULL) { + int cmsg_type; - cmn = cm->m_next; - cm->m_next = NULL; - cmsg_type = mtod(cm, struct cmsghdr *)->cmsg_type; + cmn = cm->m_next; + cm->m_next = NULL; + cmsg_type = mtod(cm, struct cmsghdr *)->cmsg_type; + /* + * Call the protocol to externalize SCM_RIGHTS message + * and return the modified message to the caller upon + * success. Otherwise, all other control messages are + * returned unmodified to the caller. Note that we + * only get into this loop if MSG_PEEK is not set. + */ + if (pr->pr_domain->dom_externalize != NULL && + cmsg_type == SCM_RIGHTS) { /* - * 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. + * 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 (pr->pr_domain->dom_externalize != NULL && - cmsg_type == SCM_RIGHTS) { - /* - * Release socket lock: see 3903171. This - * would also allow more records to be appended - * to the socket buffer. We still have SB_LOCK - * set on it, so we can be sure that the head - * of the mbuf chain won't change. - */ - socket_unlock(so, 0); - error = (*pr->pr_domain->dom_externalize)(cm); - socket_lock(so, 0); - } else { - error = 0; - } + socket_unlock(so, 0); + error = (*pr->pr_domain->dom_externalize)(cm); + socket_lock(so, 0); + } else { + error = 0; + } - if (controlp != NULL && error == 0) { - *controlp = cm; - controlp = &(*controlp)->m_next; - orig_resid = 0; - } else { - (void) m_free(cm); - } - cm = cmn; + if (controlp != NULL && error == 0) { + *controlp = cm; + controlp = &(*controlp)->m_next; + } else { + (void) m_free(cm); } - /* - * Update the value of nextrecord in case we received new - * records when the socket was unlocked above for - * externalizing SCM_RIGHTS. - */ - if (m != NULL) - nextrecord = sb_rcv->sb_mb->m_nextpkt; - else - nextrecord = sb_rcv->sb_mb; - orig_resid = 0; + cm = cmn; } - /* - * If the socket is a TCP socket with message delivery - * enabled, then create a control msg to deliver the - * relative TCP sequence number for this data. Waiting - * until this point will protect against failures to - * allocate an mbuf for control msgs. + * Update the value of nextrecord in case we received new + * records when the socket was unlocked above for + * externalizing SCM_RIGHTS. */ - if (so->so_type == SOCK_STREAM && SOCK_PROTO(so) == IPPROTO_TCP && - (so->so_flags & SOF_ENABLE_MSGS) && controlp != NULL) { - struct mbuf *seq_cm; + if (m != NULL) + nextrecord = sb_rcv->sb_mb->m_nextpkt; + else + nextrecord = sb_rcv->sb_mb; - seq_cm = sbcreatecontrol((caddr_t)&m->m_pkthdr.msg_seq, - sizeof (uint32_t), SCM_SEQNUM, SOL_SOCKET); - if (seq_cm == NULL) { - /* unable to allocate a control mbuf */ - error = ENOBUFS; - goto release; - } - *controlp = seq_cm; - controlp = &seq_cm->m_next; - } +done: + *mp = m; + *nextrecordp = nextrecord; - if (m != NULL) { - if (!(flags & MSG_PEEK)) { - /* - * We get here because m points to an mbuf following - * any MT_SONAME or MT_CONTROL mbufs which have been - * processed above. In any case, m should be pointing - * to the head of the mbuf chain, and the nextrecord - * should be either NULL or equal to m->m_nextpkt. - * See comments above about SB_LOCK. - */ - if (m != so->so_rcv.sb_mb || - m->m_nextpkt != nextrecord) { - panic("%s: post-control !sync so=%p m=%p " - "nextrecord=%p\n", __func__, so, m, - nextrecord); - /* NOTREACHED */ - } - if (nextrecord == NULL) - so->so_rcv.sb_lastrecord = m; - } - type = m->m_type; - if (type == MT_OOBDATA) - flags |= MSG_OOB; - } else { - if (!(flags & MSG_PEEK)) { - SB_EMPTY_FIXUP(&so->so_rcv); - } - } - SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); - SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); + return (error); +} - moff = 0; - offset = 0; +/* + * Implement receive operations on a socket. + * We depend on the way that records are added to the sockbuf + * by sbappend*. In particular, each record (mbufs linked through m_next) + * must begin with an address if the protocol so specifies, + * followed by an optional mbuf or mbufs containing ancillary data, + * and then zero or more mbufs of data. + * In order to avoid blocking network interrupts for the entire time here, + * we splx() while doing the actual copy to user space. + * Although the sockbuf is locked, new data may still be appended, + * and thus we must maintain consistency of the sockbuf during that time. + * + * The caller may receive the data as a single mbuf chain by supplying + * an mbuf **mp0 for use in returning the chain. The uio is then used + * only for the count in uio_resid. + * + * Returns: 0 Success + * ENOBUFS + * ENOTCONN + * EWOULDBLOCK + * uiomove:EFAULT + * sblock:EWOULDBLOCK + * sblock:EINTR + * sbwait:EBADF + * sbwait:EINTR + * sodelayed_copy:EFAULT + * :EINVAL[TCP] + * :EWOULDBLOCK[TCP] + * :??? + * dom_externalize>:EMSGSIZE[AF_UNIX] + * dom_externalize>:ENOBUFS[AF_UNIX] + * dom_externalize>:??? + * + * Notes: Additional return values from calls through and + * dom_externalize> depend on protocols other than + * TCP or AF_UNIX, which are documented above. + */ +int +soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, + struct mbuf **mp0, struct mbuf **controlp, int *flagsp) +{ + struct mbuf *m, **mp, *ml = NULL; + struct mbuf *nextrecord, *free_list; + int flags, error, offset; + user_ssize_t len; + struct protosw *pr = so->so_proto; + int moff, type = 0; + user_ssize_t orig_resid = uio_resid(uio); + user_ssize_t delayed_copy_len; + int can_delay; + int need_event; + struct proc *p = current_proc(); + boolean_t en_tracing = FALSE; - if (!(flags & MSG_PEEK) && uio_resid(uio) > sorecvmincopy) - can_delay = 1; + /* + * 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 (controlp != NULL) + *controlp = NULL; + if (flagsp != NULL) + flags = *flagsp &~ MSG_EOR; else - can_delay = 0; + flags = 0; - need_event = 0; + /* + * If a recv attempt is made on a previously-accepted socket + * that has been marked as inactive (disconnected), reject + * the request. + */ + if (so->so_flags & SOF_DEFUNCT) { + struct sockbuf *sb = &so->so_rcv; - 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; - } + 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); /* - * Make sure to allways set MSG_OOB event when getting - * out of band data inline. + * 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_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 (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) { /* - * 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. + * 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. */ - 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); + error = (*pr->pr_usrreqs->pru_preconnect)(so); - 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) { + socket_unlock(so, 1); + return (error); + } + } - if (error) - goto release; - } - } else { - uio_setresid(uio, (uio_resid(uio) - len)); + 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); } - 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 this packet is an unordered packet - * (indicated by M_UNORDERED_DATA flag), remove - * the additional bytes added to the - * receive socket buffer size. - */ - if ((so->so_flags & SOF_ENABLE_MSGS) && + /* + * 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 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 == NULL || (((flags & MSG_DONTWAIT) == 0 && + so->so_rcv.sb_cc < uio_resid(uio)) && + (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || + ((flags & MSG_WAITALL) && uio_resid(uio) <= so->so_rcv.sb_hiwat)) && + m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { + /* + * Panic if we notice inconsistencies in the socket's + * receive list; both sb_mb and sb_cc should correctly + * reflect the contents of the list, otherwise we may + * end up with false positives during select() or poll() + * which could put the application in a bad state. + */ + SB_MB_CHECK(&so->so_rcv); + + 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 + + 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 the socket is a TCP socket with message delivery + * enabled, then create a control msg to deliver the + * relative TCP sequence number for this data. Waiting + * until this point will protect against failures to + * allocate an mbuf for control msgs. + */ + if (so->so_type == SOCK_STREAM && SOCK_PROTO(so) == IPPROTO_TCP && + (so->so_flags & SOF_ENABLE_MSGS) && controlp != NULL) { + struct mbuf *seq_cm; + + seq_cm = sbcreatecontrol((caddr_t)&m->m_pkthdr.msg_seq, + sizeof (uint32_t), SCM_SEQNUM, SOL_SOCKET); + if (seq_cm == NULL) { + /* unable to allocate a control mbuf */ + error = ENOBUFS; + goto release; + } + *controlp = seq_cm; + controlp = &seq_cm->m_next; + } + + if (m != NULL) { + 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; + + need_event = 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 this packet is an unordered packet + * (indicated by M_UNORDERED_DATA flag), remove + * the additional bytes added to the + * receive socket buffer size. + */ + if ((so->so_flags & SOF_ENABLE_MSGS) && m->m_len && (m->m_flags & M_UNORDERED_DATA) && sbreserve(&so->so_rcv, @@ -2766,242 +3676,716 @@ dontblock: SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); } } else { - if (flags & MSG_PEEK) { - moff += len; - } else { - if (mp != NULL) { - int copy_flag; + 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; + /* + * delay posting the actual event until + * after any delayed copy processing + * has finished + */ + need_event = 1; + 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 && + (((struct inpcb *)so->so_pcb)->inp_state != + INPCB_STATE_DEAD)) + (*pr->pr_usrreqs->pru_rcvd)(so, flags); + + SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); + + if (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); + } + } - 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; + /* + * 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 (need_event) + postevent(so, 0, EV_OOB); + + if (orig_resid == uio_resid(uio) && orig_resid && + (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { + sbunlock(&so->so_rcv, 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_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; - } - } else { - offset += len; - if (offset == so->so_oobmark) - break; - } + 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; } - 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. + * Do not block if we got some data */ - while (flags & (MSG_WAITALL|MSG_WAITSTREAM) && m == NULL && - (uio_resid(uio) - delayed_copy_len) > 0 && - !sosendallatonce(so) && !nextrecord) { - if (so->so_error || so->so_state & SS_CANTRCVMORE) - goto release; - - /* - * Depending on the protocol (e.g. TCP), the following - * might cause the socket lock to be dropped and later - * be reacquired, and more data could have arrived and - * have been appended to the receive socket buffer by - * the time it returns. Therefore, we only sleep in - * sbwait() below if and only if the socket buffer is - * empty, in order to avoid a false sleep. - */ - if (pr->pr_flags & PR_WANTRCVD && so->so_pcb && - (((struct inpcb *)so->so_pcb)->inp_state != - INPCB_STATE_DEAD)) - (*pr->pr_usrreqs->pru_rcvd)(so, flags); + if (free_list != NULL) { + error = 0; + goto release; + } - SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); - SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); + SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); + SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); - 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); + sbunlock(&so->so_rcv, TRUE); /* keep socket locked */ + sblocked = 0; - if (error) - goto release; - } - m = so->so_rcv.sb_mb; - if (m != NULL) { - nextrecord = m->m_nextpkt; - } - SB_MB_CHECK(&so->so_rcv); + error = sbwait(&so->so_rcv); + if (error) { + goto release; } + goto restart; } -#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 */ + + 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; } -#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); - } + 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); } /* - * 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. + * Loop to copy the mbufs of the current record + * Support zero length packets */ - if ((so->so_options & SO_WANTMORE) && so->so_rcv.sb_cc > 0) - flags |= MSG_HAVEMORE; + 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 ((flags & MSG_PEEK) == 0) { - if (m == NULL) { - so->so_rcv.sb_mb = nextrecord; + if (len == m->m_len) { /* - * 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. + * m was entirely copied */ - 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; + 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; } - 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); - } + /* + * Link current mbuf to last mbuf of current packet + */ + if (ml != NULL) + ml->m_next = m; + ml = m; - 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 (need_event) - postevent(so, 0, EV_OOB); + /* + * Move next buf to head of socket buffer + */ + so->so_rcv.sb_mb = m = ml->m_next; + ml->m_next = 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 (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; + } } - - 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); + panic("%s: after big while so=%llx ref=%d on socket\n", + __func__, + (uint64_t)DEBUG_KERNEL_ADDRPERM(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 */ - - KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, so, uio_resid(uio), - so->so_rcv.sb_cc, 0, error); + /* + * 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); - return (error); -} + if (error == 0) { + error = uiomove(mtod(m, caddr_t), (int)len, + auio); + } + socket_lock(so, 0); + if (error) + goto release; -/* - * 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->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; + } + } - m = *free_list; + 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"); - socket_unlock(so, 0); + /* + * 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; - while (m != NULL && error == 0) { - error = uiomove(mtod(m, caddr_t), (int)m->m_len, uio); - m = m->m_next; + goto next; } - m_freem_list(*free_list); + if (flagsp != NULL) + *flagsp |= flags; - *free_list = NULL; - *resid = 0; +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; - socket_lock(so, 0); + 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); } @@ -3028,6 +4412,8 @@ 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: @@ -3046,11 +4432,13 @@ soshutdown(struct socket *so, int how) break; } + KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_END, how, error, 0, 0, 0); + return (error); } int -soshutdownlock(struct socket *so, int how) +soshutdownlock_final(struct socket *so, int how) { struct protosw *pr = so->so_proto; int error = 0; @@ -3076,28 +4464,41 @@ soshutdownlock(struct socket *so, int how) postevent(so, 0, EV_WCLOSED); } done: - KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_END, 0, 0, 0, 0, 0); + KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN, how, 1, 0, 0, 0); return (error); } -void -sowflush(struct socket *so) +int +soshutdownlock(struct socket *so, int how) { - struct sockbuf *sb = &so->so_snd; -#ifdef notyet - lck_mtx_t *mutex_held; + int error = 0; + +#if CONTENT_FILTER /* - * 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. + * A content filter may delay the actual shutdown until it + * has processed the pending data */ - 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; + 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 */ - lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); -#endif /* notyet */ + 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 @@ -3138,7 +4539,7 @@ sorflush(struct socket *so) 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 /* notyet */ sflt_notify(so, sock_evt_flush_read, NULL); @@ -3317,6 +4718,33 @@ sooptcopyin_timeval(struct sockopt *sopt, struct timeval *tv_p) return (0); } +int +soopt_cred_check(struct socket *so, int priv, boolean_t allow_root) +{ + kauth_cred_t cred = NULL; + proc_t ep = PROC_NULL; + uid_t uid; + int error = 0; + + if (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 @@ -3413,9 +4841,12 @@ sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) 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) @@ -3463,17 +4894,40 @@ sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) * Make sure the low-water is never greater than * the high-water. */ - case SO_SNDLOWAT: + case SO_SNDLOWAT: { + int space = sbspace(&so->so_snd); + u_int32_t hiwat = so->so_snd.sb_hiwat; + + if (so->so_snd.sb_flags & SB_UNIX) { + struct unpcb *unp = + (struct unpcb *)(so->so_pcb); + if (unp != NULL && + unp->unp_conn != NULL) { + hiwat += unp->unp_conn->unp_cc; + } + } + so->so_snd.sb_lowat = - (optval > so->so_snd.sb_hiwat) ? - so->so_snd.sb_hiwat : optval; + (optval > hiwat) ? + hiwat : optval; + + if (space >= so->so_snd.sb_lowat) { + sowwakeup(so); + } break; - case SO_RCVLOWAT: + } + case SO_RCVLOWAT: { + int64_t data_len; so->so_rcv.sb_lowat = (optval > so->so_rcv.sb_hiwat) ? so->so_rcv.sb_hiwat : optval; + data_len = so->so_rcv.sb_cc + - so->so_rcv.sb_ctl; + if (data_len >= so->so_rcv.sb_lowat) + sorwakeup(so); break; } + } break; case SO_SNDTIMEO: @@ -3561,6 +5015,45 @@ sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) 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); + 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); + if (error == 0) + inp_set_intcoproc_allowed( + sotoinpcb(so)); + } else if (optval == 0) + inp_clear_intcoproc_allowed(sotoinpcb(so)); + break; + case SO_LABEL: #if CONFIG_MACF_SOCKET if ((error = sooptcopyin(sopt, &extmac, sizeof (extmac), @@ -3624,9 +5117,16 @@ sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) 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; } @@ -3642,6 +5142,7 @@ sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) break; } +#if (DEVELOPMENT || DEBUG) case SO_TRAFFIC_CLASS_DBG: { struct so_tcdbg so_tcdbg; @@ -3654,6 +5155,7 @@ sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) goto out; break; } +#endif /* (DEVELOPMENT || DEBUG) */ case SO_PRIVILEGED_TRAFFIC_CLASS: error = priv_check_cred(kauth_cred_get(), @@ -3670,6 +5172,15 @@ sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) 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)); @@ -3699,10 +5210,12 @@ sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) char d[MAX_IPv6_STR_LEN]; struct inpcb *inp = sotoinpcb(so); - SODEFUNCTLOG(("%s[%d]: so 0x%llx [%s %s:%d -> " - "%s:%d] is now marked as %seligible for " + SODEFUNCTLOG("%s[%d, %s]: so 0x%llx " + "[%s %s:%d -> %s:%d] is now marked " + "as %seligible for " "defunct\n", __func__, proc_selfpid(), - (uint64_t)VM_KERNEL_ADDRPERM(so), + 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) ? @@ -3715,15 +5228,17 @@ sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) (void *)&inp->in6p_faddr, d, sizeof (d)), ntohs(inp->in6p_fport), (so->so_flags & SOF_NODEFUNCT) ? - "not " : "")); + "not " : ""); } else { - SODEFUNCTLOG(("%s[%d]: so 0x%llx [%d,%d] is " - "now marked as %seligible for defunct\n", + SODEFUNCTLOG("%s[%d, %s]: so 0x%llx [%d,%d] " + "is now marked as %seligible for " + "defunct\n", __func__, proc_selfpid(), - (uint64_t)VM_KERNEL_ADDRPERM(so), + proc_best_name(current_proc()), + (uint64_t)DEBUG_KERNEL_ADDRPERM(so), SOCK_DOM(so), SOCK_TYPE(so), (so->so_flags & SOF_NODEFUNCT) ? - "not " : "")); + "not " : ""); } break; @@ -3783,6 +5298,101 @@ sosetoptlock(struct socket *so, struct sockopt *sopt, int dolock) 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; + } + + error = necp_client_register_socket_flow(so->last_pid, + inp->necp_client_uuid, inp); + if (error != 0) { + uuid_clear(inp->necp_client_uuid); + goto out; + } + + if (inp->inp_lport != 0) { + // There is bound local port, so this is not + // a fresh socket. Assign to the client. + necp_client_assign_from_socket(so->last_pid, inp->necp_client_uuid, inp); + } + + 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_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; + default: error = ENOPROTOOPT; break; @@ -3832,8 +5442,8 @@ sooptcopyout_timeval(struct sockopt *sopt, const struct timeval *tv_p) { int error; size_t len; - struct user64_timeval tv64; - struct user32_timeval tv32; + struct user64_timeval tv64 = {}; + struct user32_timeval tv32 = {}; const void * val; size_t valsize; @@ -3929,9 +5539,12 @@ sogetoptlock(struct socket *so, struct sockopt *sopt, int dolock) 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)); @@ -3961,6 +5574,26 @@ integer: } goto integer; + case SO_NUMRCVPKT: + if (so->so_proto->pr_flags & PR_ATOMIC) { + int cnt = 0; + struct mbuf *m1; + + m1 = so->so_rcv.sb_mb; + while (m1 != NULL) { + if (m1->m_type == MT_DATA || + m1->m_type == MT_HEADER || + m1->m_type == MT_OOBDATA) + cnt += 1; + m1 = m1->m_nextpkt; + } + optval = cnt; + goto integer; + } else { + error = EINVAL; + break; + } + case SO_NWRITE: optval = so->so_snd.sb_cc; goto integer; @@ -3970,10 +5603,20 @@ integer: so->so_error = 0; goto integer; - case SO_SNDBUF: - optval = so->so_snd.sb_hiwat; - goto integer; + case SO_SNDBUF: { + u_int32_t hiwat = so->so_snd.sb_hiwat; + if (so->so_snd.sb_flags & SB_UNIX) { + struct unpcb *unp = + (struct unpcb *)(so->so_pcb); + if (unp != NULL && unp->unp_conn != NULL) { + hiwat += unp->unp_conn->unp_cc; + } + } + + optval = hiwat; + goto integer; + } case SO_RCVBUF: optval = so->so_rcv.sb_hiwat; goto integer; @@ -4015,6 +5658,25 @@ integer: 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: #if CONFIG_MACF_SOCKET if ((error = sooptcopyin(sopt, &extmac, sizeof (extmac), @@ -4053,7 +5715,7 @@ integer: goto integer; case SO_NP_EXTENSIONS: { - struct so_np_extensions sonpx; + struct so_np_extensions sonpx = {}; sonpx.npx_flags = (so->so_flags & SOF_NPX_SETOPTSHUT) ? SONPX_SETOPTSHUT : 0; @@ -4077,9 +5739,11 @@ integer: sizeof (so->so_tc_stats)); break; +#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); @@ -4120,6 +5784,59 @@ integer: 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; + } +#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; + default: error = ENOPROTOOPT; break; @@ -4267,6 +5984,10 @@ sohasoutofband(struct socket *so) 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 @@ -4317,14 +6038,15 @@ sopoll(struct socket *so, int events, kauth_cred_t cred, void * wql) } int -soo_kqfilter(struct fileproc *fp, struct knote *kn, vfs_context_t ctx) +soo_kqfilter(struct fileproc *fp, struct knote *kn, + struct kevent_internal_s *kev, vfs_context_t ctx) { #pragma unused(fp) #if !CONFIG_MACF_SOCKET #pragma unused(ctx) #endif /* MAC_SOCKET */ struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; - struct klist *skl; + int result; socket_lock(so, 1); so_update_last_owner_locked(so, PROC_NULL); @@ -4334,71 +6056,48 @@ soo_kqfilter(struct fileproc *fp, struct knote *kn, vfs_context_t ctx) if (mac_socket_check_kqfilter(proc_ucred(vfs_context_proc(ctx)), kn, so) != 0) { socket_unlock(so, 1); - return (1); + kn->kn_flags = EV_ERROR; + kn->kn_data = EPERM; + return 0; } #endif /* MAC_SOCKET */ switch (kn->kn_filter) { case EVFILT_READ: - kn->kn_fop = &soread_filtops; - skl = &so->so_rcv.sb_sel.si_note; + kn->kn_filtid = EVFILTID_SOREAD; break; case EVFILT_WRITE: - kn->kn_fop = &sowrite_filtops; - skl = &so->so_snd.sb_sel.si_note; + kn->kn_filtid = EVFILTID_SOWRITE; break; case EVFILT_SOCK: - kn->kn_fop = &sock_filtops; - skl = &so->so_klist; + kn->kn_filtid = EVFILTID_SCK; + break; + case EVFILT_EXCEPT: + kn->kn_filtid = EVFILTID_SOEXCEPT; break; default: socket_unlock(so, 1); - return (1); - } - - if (KNOTE_ATTACH(skl, kn)) { - switch (kn->kn_filter) { - case EVFILT_READ: - so->so_rcv.sb_flags |= SB_KNOTE; - break; - case EVFILT_WRITE: - so->so_snd.sb_flags |= SB_KNOTE; - break; - case EVFILT_SOCK: - so->so_flags |= SOF_KNOTE; - break; - default: - socket_unlock(so, 1); - return (1); - } + kn->kn_flags = EV_ERROR; + kn->kn_data = EINVAL; + return 0; } - socket_unlock(so, 1); - return (0); -} - -static void -filt_sordetach(struct knote *kn) -{ - struct socket *so = (struct socket *)kn->kn_fp->f_fglob->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; + /* + * 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; } -/*ARGSUSED*/ static int -filt_soread(struct knote *kn, long hint) +filt_soread_common(struct knote *kn, struct socket *so) { - struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; - - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_lock(so, 1); - if (so->so_options & SO_ACCEPTCONN) { - int isempty; + int is_not_empty; /* * Radar 6615193 handle the listen case dynamically @@ -4407,58 +6106,47 @@ filt_soread(struct knote *kn, long hint) */ kn->kn_data = so->so_qlen; - isempty = ! TAILQ_EMPTY(&so->so_comp); - - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); + is_not_empty = ! TAILQ_EMPTY(&so->so_comp); - return (isempty); + return (is_not_empty); } /* 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. + */ kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl; - if (so->so_oobmark) { - if (kn->kn_flags & EV_OOBAND) { + if (kn->kn_sfflags & NOTE_OOB) { + if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) { + kn->kn_fflags |= NOTE_OOB; kn->kn_data -= so->so_oobmark; - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (1); - } - kn->kn_data = so->so_oobmark; - kn->kn_flags |= EV_OOBAND; - } else { - if (so->so_state & SS_CANTRCVMORE) { - kn->kn_flags |= EV_EOF; - kn->kn_fflags = so->so_error; - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); return (1); } } - if (so->so_state & SS_RCVATMARK) { - if (kn->kn_flags & EV_OOBAND) { - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (1); - } - kn->kn_flags |= EV_OOBAND; - } else if (kn->kn_flags & EV_OOBAND) { - kn->kn_data = 0; - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); - return (0); + if ((so->so_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; + return (1); } if (so->so_error) { /* temporary udp error */ - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); return (1); } 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; @@ -4466,22 +6154,112 @@ filt_soread(struct knote *kn, long hint) lowwat = kn->kn_sdata; } + /* + * The order below is important. Since NOTE_LOWAT + * overrides sb_lowat, check for NOTE_LOWAT case + * first. + */ + if (kn->kn_sfflags & NOTE_LOWAT) + return (kn->kn_data >= lowwat); + + return (so->so_rcv.sb_cc >= lowwat); +} + +static int +filt_sorattach(struct knote *kn, __unused struct kevent_internal_s *kev) +{ + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->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_hookid = EV_OOBAND; + } else { + kn->kn_hookid = 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, so); +} + +static void +filt_sordetach(struct knote *kn) +{ + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->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->f_fglob->fg_data; + int retval; + + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_lock(so, 1); + + retval = filt_soread_common(kn, so); + if ((hint & SO_FILT_HINT_LOCKED) == 0) socket_unlock(so, 1); - return ((kn->kn_flags & EV_OOBAND) || kn->kn_data >= lowwat); + return retval; } -static void -filt_sowdetach(struct knote *kn) +static int +filt_sortouch(struct knote *kn, struct kevent_internal_s *kev) { struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + int retval; + socket_lock(so, 1); - 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; + /* 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, so); + + socket_unlock(so, 1); + + return retval; +} + +static int +filt_sorprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev) +{ +#pragma unused(data) + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + int retval; + + socket_lock(so, 1); + retval = filt_soread_common(kn, so); + if (retval) { + *kev = kn->kn_kevent; + if (kn->kn_flags & EV_CLEAR) { + kn->kn_fflags = 0; + kn->kn_data = 0; + } + } socket_unlock(so, 1); + + return retval; } int @@ -4496,31 +6274,25 @@ so_wait_for_if_feedback(struct socket *so) return (0); } -/*ARGSUSED*/ static int -filt_sowrite(struct knote *kn, long hint) +filt_sowrite_common(struct knote *kn, struct socket *so) { - struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; int ret = 0; - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_lock(so, 1); - kn->kn_data = sbspace(&so->so_snd); if (so->so_state & SS_CANTSENDMORE) { kn->kn_flags |= EV_EOF; kn->kn_fflags = so->so_error; - ret = 1; - goto out; + return 1; } if (so->so_error) { /* temporary udp error */ - ret = 1; - goto out; + return 1; } - if (((so->so_state & SS_ISCONNECTED) == 0) && - (so->so_proto->pr_flags & PR_CONNREQUIRED)) { - ret = 0; - goto out; + if (!socanwrite(so)) { + return 0; + } + if (so->so_flags1 & SOF1_PRECONNECT_DATA) { + return 1; } int64_t lowwat = so->so_snd.sb_lowat; if (kn->kn_sfflags & NOTE_LOWAT) { @@ -4530,124 +6302,365 @@ filt_sowrite(struct knote *kn, long hint) lowwat = kn->kn_sdata; } if (kn->kn_data >= lowwat) { - if ((so->so_flags & SOF_NOTSENT_LOWAT) != 0) { - ret = tcp_notsent_lowat_check(so); + 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 + else { + return 1; + } } else { ret = 1; } } if (so_wait_for_if_feedback(so)) ret = 0; -out: - if ((hint & SO_FILT_HINT_LOCKED) == 0) - socket_unlock(so, 1); return (ret); } +static int +filt_sowattach(struct knote *kn, __unused struct kevent_internal_s *kev) +{ + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + + /* socket locked */ + if (KNOTE_ATTACH(&so->so_snd.sb_sel.si_note, kn)) + so->so_snd.sb_flags |= SB_KNOTE; + + /* determine if its already fired */ + return filt_sowrite_common(kn, so); +} + static void -filt_sockdetach(struct knote *kn) +filt_sowdetach(struct knote *kn) { struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; socket_lock(so, 1); - if ((so->so_flags & SOF_KNOTE) != 0) - if (KNOTE_DETACH(&so->so_klist, kn)) - so->so_flags &= ~SOF_KNOTE; + 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); } +/*ARGSUSED*/ static int -filt_sockev(struct knote *kn, long hint) +filt_sowrite(struct knote *kn, long hint) { - int ret = 0, locked = 0; struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; - long ev_hint = (hint & SO_FILT_HINT_EV); + int ret; - if ((hint & SO_FILT_HINT_LOCKED) == 0) { + if ((hint & SO_FILT_HINT_LOCKED) == 0) socket_lock(so, 1); - locked = 1; + + ret = filt_sowrite_common(kn, so); + + if ((hint & SO_FILT_HINT_LOCKED) == 0) + socket_unlock(so, 1); + + return ret; +} + +static int +filt_sowtouch(struct knote *kn, struct kevent_internal_s *kev) +{ + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + int ret; + + socket_lock(so, 1); + + /*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, so); + + socket_unlock(so, 1); + + return ret; +} + +static int +filt_sowprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev) +{ +#pragma unused(data) + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + int ret; + + socket_lock(so, 1); + ret = filt_sowrite_common(kn, so); + if (ret) { + *kev = kn->kn_kevent; + if (kn->kn_flags & EV_CLEAR) { + kn->kn_fflags = 0; + kn->kn_data = 0; + } } + socket_unlock(so, 1); + return ret; +} + +static int +filt_sockev_common(struct knote *kn, struct socket *so, long ev_hint) +{ + int ret = 0; + uint32_t level_trigger = 0; if (ev_hint & SO_FILT_HINT_CONNRESET) { - if (kn->kn_sfflags & NOTE_CONNRESET) - kn->kn_fflags |= NOTE_CONNRESET; + kn->kn_fflags |= NOTE_CONNRESET; } if (ev_hint & SO_FILT_HINT_TIMEOUT) { - if (kn->kn_sfflags & NOTE_TIMEOUT) - kn->kn_fflags |= NOTE_TIMEOUT; + kn->kn_fflags |= NOTE_TIMEOUT; } if (ev_hint & SO_FILT_HINT_NOSRCADDR) { - if (kn->kn_sfflags & NOTE_NOSRCADDR) - kn->kn_fflags |= NOTE_NOSRCADDR; + kn->kn_fflags |= NOTE_NOSRCADDR; } if (ev_hint & SO_FILT_HINT_IFDENIED) { - if ((kn->kn_sfflags & NOTE_IFDENIED)) - kn->kn_fflags |= NOTE_IFDENIED; + kn->kn_fflags |= NOTE_IFDENIED; } if (ev_hint & SO_FILT_HINT_KEEPALIVE) { - if (kn->kn_sfflags & NOTE_KEEPALIVE) - kn->kn_fflags |= NOTE_KEEPALIVE; + kn->kn_fflags |= NOTE_KEEPALIVE; } if (ev_hint & SO_FILT_HINT_ADAPTIVE_WTIMO) { - if (kn->kn_sfflags & NOTE_ADAPTIVE_WTIMO) - kn->kn_fflags |= NOTE_ADAPTIVE_WTIMO; + kn->kn_fflags |= NOTE_ADAPTIVE_WTIMO; } if (ev_hint & SO_FILT_HINT_ADAPTIVE_RTIMO) { - if (kn->kn_sfflags & NOTE_ADAPTIVE_RTIMO) - kn->kn_fflags |= NOTE_ADAPTIVE_RTIMO; + kn->kn_fflags |= NOTE_ADAPTIVE_RTIMO; } - if (ev_hint & SO_FILT_HINT_CONNECTED) { - if (kn->kn_sfflags & NOTE_CONNECTED) - kn->kn_fflags |= NOTE_CONNECTED; + if ((ev_hint & SO_FILT_HINT_CONNECTED) || + (so->so_state & SS_ISCONNECTED)) { + kn->kn_fflags |= NOTE_CONNECTED; + level_trigger |= NOTE_CONNECTED; } - if (ev_hint & SO_FILT_HINT_DISCONNECTED) { - if (kn->kn_sfflags & NOTE_DISCONNECTED) - kn->kn_fflags |= NOTE_DISCONNECTED; + if ((ev_hint & SO_FILT_HINT_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_sfflags & NOTE_CONNINFO_UPDATED)) + (so->so_proto->pr_flags & PR_EVCONNINFO)) kn->kn_fflags |= NOTE_CONNINFO_UPDATED; } - if ((kn->kn_sfflags & NOTE_READCLOSED) && - (so->so_state & SS_CANTRCVMORE)) - kn->kn_fflags |= NOTE_READCLOSED; + if ((ev_hint & SO_FILT_HINT_NOTIFY_ACK) || + tcp_notify_ack_active(so)) { + kn->kn_fflags |= NOTE_NOTIFY_ACK; + } + + 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; + } + + if (so->so_state & SS_CANTSENDMORE) { + kn->kn_fflags |= NOTE_WRITECLOSED; + level_trigger |= NOTE_WRITECLOSED; + } + + if ((ev_hint & SO_FILT_HINT_SUSPEND) || + (so->so_flags & SOF_SUSPENDED)) { + kn->kn_fflags &= ~(NOTE_SUSPEND | NOTE_RESUME); + + /* If resume event was delivered before, reset it */ + kn->kn_hookid &= ~NOTE_RESUME; + + kn->kn_fflags |= NOTE_SUSPEND; + level_trigger |= NOTE_SUSPEND; + } + + 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_hookid &= ~NOTE_SUSPEND; + + kn->kn_fflags |= NOTE_RESUME; + level_trigger |= NOTE_RESUME; + } + + if (so->so_error != 0) { + ret = 1; + kn->kn_data = so->so_error; + kn->kn_flags |= EV_EOF; + } else { + get_sockev_state(so, (u_int32_t *)&(kn->kn_data)); + } + + /* 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_hookid; + 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; + + return (ret); +} + +static int +filt_sockattach(struct knote *kn, __unused struct kevent_internal_s *kev) +{ + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + + /* socket locked */ + kn->kn_hookid = 0; + if (KNOTE_ATTACH(&so->so_klist, kn)) + so->so_flags |= SOF_KNOTE; + + /* determine if event already fired */ + return filt_sockev_common(kn, so, 0); +} + +static void +filt_sockdetach(struct knote *kn) +{ + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + socket_lock(so, 1); + + if ((so->so_flags & SOF_KNOTE) != 0) + if (KNOTE_DETACH(&so->so_klist, kn)) + so->so_flags &= ~SOF_KNOTE; + socket_unlock(so, 1); +} + +static int +filt_sockev(struct knote *kn, long hint) +{ + int ret = 0, locked = 0; + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + long ev_hint = (hint & SO_FILT_HINT_EV); + + if ((hint & SO_FILT_HINT_LOCKED) == 0) { + socket_lock(so, 1); + locked = 1; + } + + ret = filt_sockev_common(kn, so, ev_hint); + + if (locked) + socket_unlock(so, 1); + + return ret; +} + + + +/* + * filt_socktouch - update event state + */ +static int +filt_socktouch( + struct knote *kn, + struct kevent_internal_s *kev) +{ + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->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_hookid); + + /* 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_hookid &= + ~(changed_flags & EVFILT_SOCK_LEVEL_TRIGGER_MASK); + + /* determine if we have events to deliver */ + ret = filt_sockev_common(kn, so, 0); - if ((kn->kn_sfflags & NOTE_WRITECLOSED) && - (so->so_state & SS_CANTSENDMORE)) - kn->kn_fflags |= NOTE_WRITECLOSED; + socket_unlock(so, 1); - if ((kn->kn_sfflags & NOTE_SUSPEND) && - ((ev_hint & SO_FILT_HINT_SUSPEND) || - (so->so_flags & SOF_SUSPENDED))) { - kn->kn_fflags &= ~(NOTE_SUSPEND | NOTE_RESUME); - kn->kn_fflags |= NOTE_SUSPEND; - } + return ret; +} - if ((kn->kn_sfflags & NOTE_RESUME) && - ((ev_hint & SO_FILT_HINT_RESUME) || - (so->so_flags & SOF_SUSPENDED) == 0)) { - kn->kn_fflags &= ~(NOTE_SUSPEND | NOTE_RESUME); - kn->kn_fflags |= NOTE_RESUME; - } +/* + * filt_sockprocess - query event fired state and return data + */ +static int +filt_sockprocess( + struct knote *kn, + struct filt_process_s *data, + struct kevent_internal_s *kev) +{ +#pragma unused(data) - if (so->so_error != 0) { - ret = 1; - kn->kn_data = so->so_error; - kn->kn_flags |= EV_EOF; - } else { - get_sockev_state(so, (u_int32_t *)&(kn->kn_data)); - } + struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data; + int ret = 0; - if (kn->kn_fflags != 0) - ret = 1; + socket_lock(so, 1); - if (locked) - socket_unlock(so, 1); + ret = filt_sockev_common(kn, so, 0); + if (ret) { + *kev = kn->kn_kevent; - return (ret); + /* + * 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_hookid |= (kn->kn_fflags & + EVFILT_SOCK_LEVEL_TRIGGER_MASK); + + /* + * 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_hookid &= ~NOTE_RESUME; + if (kn->kn_fflags & NOTE_RESUME) + kn->kn_hookid &= ~NOTE_SUSPEND; + + if (kn->kn_flags & EV_CLEAR) { + kn->kn_data = 0; + kn->kn_fflags = 0; + } + } + + socket_unlock(so, 1); + + return ret; } void @@ -4655,6 +6668,13 @@ get_sockev_state(struct socket *so, u_int32_t *statep) { u_int32_t state = *(statep); + /* + * If the state variable is already used by a previous event, + * reset it. + */ + if (state != 0) + return; + if (so->so_state & SS_ISCONNECTED) state |= SOCKEV_CONNECTED; else @@ -4683,19 +6703,18 @@ solockhistory_nr(struct socket *so) return (lock_history_str); } -int +void socket_lock(struct socket *so, int refcount) { - int error = 0; void *lr_saved; lr_saved = __builtin_return_address(0); if (so->so_proto->pr_lock) { - error = (*so->so_proto->pr_lock)(so, refcount, lr_saved); + (*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(so->so_proto->pr_domain->dom_mtx, LCK_MTX_ASSERT_NOTOWNED); #endif lck_mtx_lock(so->so_proto->pr_domain->dom_mtx); @@ -4704,14 +6723,37 @@ socket_lock(struct socket *so, int refcount) so->lock_lr[so->next_lock_lr] = lr_saved; so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX; } +} - return (error); +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 + mutex_held = so->so_proto->pr_domain->dom_mtx; + + 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) { - int error = 0; void *lr_saved; lck_mtx_t *mutex_held; @@ -4723,11 +6765,11 @@ socket_unlock(struct socket *so, int refcount) } if (so && so->so_proto->pr_unlock) { - error = (*so->so_proto->pr_unlock)(so, refcount, lr_saved); + (*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); + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); #endif so->unlock_lr[so->next_unlock_lr] = lr_saved; so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; @@ -4747,8 +6789,6 @@ socket_unlock(struct socket *so, int refcount) } lck_mtx_unlock(mutex_held); } - - return (error); } /* Called with socket locked, will unlock socket */ @@ -4761,7 +6801,7 @@ sofree(struct socket *so) 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); + LCK_MTX_ASSERT(mutex_held, LCK_MTX_ASSERT_OWNED); sofreelastref(so, 0); } @@ -4794,6 +6834,15 @@ somultipages(struct socket *so, boolean_t set) so->so_flags &= ~SOF_MULTIPAGES; } +void +soif2kcl(struct socket *so, boolean_t set) +{ + if (set) + so->so_flags1 |= SOF1_IF_2KCL; + else + so->so_flags1 &= ~SOF1_IF_2KCL; +} + int so_isdstlocal(struct socket *so) { @@ -4828,18 +6877,61 @@ sosetdefunct(struct proc *p, struct socket *so, int level, boolean_t noforce) if (so->so_flags & SOF_NODEFUNCT) { if (noforce) { err = EOPNOTSUPP; - SODEFUNCTLOG(("%s[%d]: (target pid %d level %d) " - "so 0x%llx [%d,%d] is not eligible for defunct " - "(%d)\n", __func__, proc_selfpid(), proc_pid(p), - level, (uint64_t)VM_KERNEL_ADDRPERM(so), - SOCK_DOM(so), SOCK_TYPE(so), err)); + 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); + } return (err); } so->so_flags &= ~SOF_NODEFUNCT; - SODEFUNCTLOG(("%s[%d]: (target pid %d level %d) so 0x%llx " - "[%d,%d] defunct by force\n", __func__, proc_selfpid(), - proc_pid(p), level, (uint64_t)VM_KERNEL_ADDRPERM(so), - SOCK_DOM(so), SOCK_TYPE(so))); + 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 { + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_forced); + } } so->so_flags |= SOF_DEFUNCT; @@ -4861,11 +6953,16 @@ sosetdefunct(struct proc *p, struct socket *so, int level, boolean_t noforce) } done: - SODEFUNCTLOG(("%s[%d]: (target pid %d level %d) so 0x%llx [%d,%d] %s " - "defunct\n", __func__, proc_selfpid(), proc_pid(p), level, - (uint64_t)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), SOCK_TYPE(so), - defunct ? "is already" : "marked as")); - + 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" : ""); + } return (err); } @@ -4889,28 +6986,39 @@ sodefunct(struct proc *p, struct socket *so, int level) char d[MAX_IPv6_STR_LEN]; struct inpcb *inp = sotoinpcb(so); - SODEFUNCTLOG(("%s[%d]: (target pid %d level %d) so 0x%llx [%s " - "%s:%d -> %s:%d] is now defunct [rcv_si 0x%x, snd_si 0x%x, " - "rcv_fl 0x%x, snd_fl 0x%x]\n", __func__, proc_selfpid(), - proc_pid(p), level, (uint64_t)VM_KERNEL_ADDRPERM(so), - (SOCK_TYPE(so) == SOCK_STREAM) ? "TCP" : "UDP", - inet_ntop(SOCK_DOM(so), ((SOCK_DOM(so) == PF_INET) ? - (void *)&inp->inp_laddr.s_addr : (void *)&inp->in6p_laddr), - s, sizeof (s)), ntohs(inp->in6p_lport), - inet_ntop(SOCK_DOM(so), (SOCK_DOM(so) == PF_INET) ? - (void *)&inp->inp_faddr.s_addr : (void *)&inp->in6p_faddr, - d, sizeof (d)), ntohs(inp->in6p_fport), + 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)); - } else { - SODEFUNCTLOG(("%s[%d]: (target pid %d level %d) so 0x%llx " - "[%d,%d] is now defunct [rcv_si 0x%x, snd_si 0x%x, " - "rcv_fl 0x%x, snd_fl 0x%x]\n", __func__, proc_selfpid(), - proc_pid(p), level, (uint64_t)VM_KERNEL_ADDRPERM(so), - SOCK_DOM(so), SOCK_TYPE(so), (uint32_t)rcv->sb_sel.si_flags, (uint32_t)snd->sb_sel.si_flags, rcv->sb_flags, - snd->sb_flags)); + snd->sb_flags); } /* @@ -4919,6 +7027,7 @@ sodefunct(struct proc *p, struct socket *so, int level) sbwakeup(rcv); sbwakeup(snd); + so->so_flags1 |= SOF1_DEFUNCTINPROG; if (rcv->sb_flags & SB_LOCK) sbunlock(rcv, TRUE); /* keep socket locked */ if (snd->sb_flags & SB_LOCK) @@ -4930,15 +7039,15 @@ sodefunct(struct proc *p, struct socket *so, int level) * states are set for the socket. This would also flush out data * hanging off the receive list of this socket. */ - (void) soshutdownlock(so, SHUT_RD); - (void) soshutdownlock(so, SHUT_WR); + (void) soshutdownlock_final(so, SHUT_RD); + (void) soshutdownlock_final(so, SHUT_WR); (void) sodisconnectlocked(so); /* * Explicitly handle connectionless-protocol disconnection * and release any remaining data in the socket buffers. */ - if (!(so->so_flags & SS_ISDISCONNECTED)) + if (!(so->so_state & SS_ISDISCONNECTED)) (void) soisdisconnected(so); if (so->so_error == 0) @@ -4955,11 +7064,207 @@ sodefunct(struct proc *p, struct socket *so, int level) sbrelease(snd); } so->so_state |= SS_DEFUNCT; + OSIncrementAtomicLong((volatile long *)&sodefunct_calls); done: return (0); } +int +soresume(struct proc *p, struct socket *so, int locked) +{ + if (locked == 0) + socket_lock(so, 1); + + 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 +so_set_extended_bk_idle(struct socket *so, int optval) +{ + int error = 0; + + 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; + + soresume(current_proc(), so, 1); + } else { + struct proc *p = current_proc(); + int i; + struct filedesc *fdp; + int count = 0; + + /* + * Unlock socket to avoid lock ordering issue with + * the proc fd table lock + */ + socket_unlock(so, 0); + + proc_fdlock(p); + + fdp = p->p_fd; + for (i = 0; i < fdp->fd_nfiles; i++) { + struct fileproc *fp = fdp->fd_ofiles[i]; + struct socket *so2; + + if (fp == NULL || + (fdp->fd_ofileflags[i] & UF_RESERVED) != 0 || + FILEGLOB_DTYPE(fp->f_fglob) != DTYPE_SOCKET) + continue; + + so2 = (struct socket *)fp->f_fglob->fg_data; + if (so != so2 && + so2->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) + count++; + if (count >= soextbkidlestat.so_xbkidle_maxperproc) + break; + } + proc_fdunlock(p); + + 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"); + } + + return (error); +} + +static void +so_stop_extended_bk_idle(struct socket *so) +{ + so->so_flags1 &= ~SOF1_EXTEND_BK_IDLE_INPROG; + so->so_extended_bk_start = 0; + + 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); + } +} + +void +so_drain_extended_bk_idle(struct socket *so) +{ + 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); + + OSIncrementAtomic(&soextbkidlestat.so_xbkidle_drained); + } + } +} + +/* + * Return values tells if socket is still in extended background idle + */ +int +so_check_extended_bk_idle_time(struct socket *so) +{ + int ret = 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); + + 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); + } + } + + return (ret); +} + +void +resume_proc_sockets(proc_t p) +{ + if (p->p_ladvflag & P_LXBKIDLEINPROG) { + struct filedesc *fdp; + int i; + + proc_fdlock(p); + fdp = p->p_fd; + for (i = 0; i < fdp->fd_nfiles; i++) { + struct fileproc *fp; + struct socket *so; + + fp = fdp->fd_ofiles[i]; + if (fp == NULL || + (fdp->fd_ofileflags[i] & UF_RESERVED) != 0 || + FILEGLOB_DTYPE(fp->f_fglob) != DTYPE_SOCKET) + continue; + + so = (struct socket *)fp->f_fglob->fg_data; + (void) soresume(p, so, 0); + } + proc_fdunlock(p); + + OSBitAndAtomic(~P_LXBKIDLEINPROG, &p->p_ladvflag); + } +} + __private_extern__ int so_set_recv_anyif(struct socket *so, int optval) { @@ -4976,6 +7281,7 @@ so_set_recv_anyif(struct socket *so, int optval) sotoinpcb(so)->inp_flags &= ~INP_RECV_ANYIF; } + return (ret); } @@ -4999,7 +7305,7 @@ int so_set_restrictions(struct socket *so, uint32_t vals) { int nocell_old, nocell_new; - int ret = 0; + int noexpensive_old, noexpensive_new; /* * Deny-type restrictions are trapdoors; once set they cannot be @@ -5015,153 +7321,46 @@ so_set_restrictions(struct socket *so, uint32_t vals) * i.e. when SO_RESTRICT_DENY_CELLULAR has not been issued. */ nocell_old = (so->so_restrictions & SO_RESTRICT_DENY_CELLULAR); + noexpensive_old = (so->so_restrictions & SO_RESTRICT_DENY_EXPENSIVE); so->so_restrictions |= (vals & (SO_RESTRICT_DENY_IN | - SO_RESTRICT_DENY_OUT | SO_RESTRICT_DENY_CELLULAR)); + SO_RESTRICT_DENY_OUT | SO_RESTRICT_DENY_CELLULAR | + SO_RESTRICT_DENY_EXPENSIVE)); nocell_new = (so->so_restrictions & SO_RESTRICT_DENY_CELLULAR); - - /* other than deny cellular, there's nothing more to do */ - if ((nocell_new - nocell_old) == 0) - return (ret); + noexpensive_new = (so->so_restrictions & SO_RESTRICT_DENY_EXPENSIVE); /* we can only set, not clear restrictions */ - VERIFY((nocell_new - nocell_old) > 0); - + if ((nocell_new - nocell_old) == 0 && + (noexpensive_new - noexpensive_old) == 0) + return (0); #if INET6 if (SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) { #else if (SOCK_DOM(so) == PF_INET) { #endif /* !INET6 */ - /* if deny cellular is now set, do what's needed for INPCB */ - inp_set_nocellular(sotoinpcb(so)); + if (nocell_new - nocell_old != 0) { + /* + * if deny cellular is now set, do what's needed + * for INPCB + */ + inp_set_nocellular(sotoinpcb(so)); + } + if (noexpensive_new - noexpensive_old != 0) { + inp_set_noexpensive(sotoinpcb(so)); + } } - return (ret); + if (SOCK_DOM(so) == PF_MULTIPATH) + mptcp_set_restrictions(so); + + return (0); } uint32_t so_get_restrictions(struct socket *so) { return (so->so_restrictions & (SO_RESTRICT_DENY_IN | - SO_RESTRICT_DENY_OUT | SO_RESTRICT_DENY_CELLULAR)); -} - -struct sockaddr_entry * -sockaddrentry_alloc(int how) -{ - struct sockaddr_entry *se; - - se = (how == M_WAITOK) ? zalloc(se_zone) : zalloc_noblock(se_zone); - if (se != NULL) - bzero(se, se_zone_size); - - return (se); -} - -void -sockaddrentry_free(struct sockaddr_entry *se) -{ - if (se->se_addr != NULL) { - FREE(se->se_addr, M_SONAME); - se->se_addr = NULL; - } - zfree(se_zone, se); -} - -struct sockaddr_entry * -sockaddrentry_dup(const struct sockaddr_entry *src_se, int how) -{ - struct sockaddr_entry *dst_se; - - dst_se = sockaddrentry_alloc(how); - if (dst_se != NULL) { - int len = src_se->se_addr->sa_len; - - MALLOC(dst_se->se_addr, struct sockaddr *, - len, M_SONAME, how | M_ZERO); - if (dst_se->se_addr != NULL) { - bcopy(src_se->se_addr, dst_se->se_addr, len); - } else { - sockaddrentry_free(dst_se); - dst_se = NULL; - } - } - - return (dst_se); -} - -struct sockaddr_list * -sockaddrlist_alloc(int how) -{ - struct sockaddr_list *sl; - - sl = (how == M_WAITOK) ? zalloc(sl_zone) : zalloc_noblock(sl_zone); - if (sl != NULL) { - bzero(sl, sl_zone_size); - TAILQ_INIT(&sl->sl_head); - } - return (sl); -} - -void -sockaddrlist_free(struct sockaddr_list *sl) -{ - struct sockaddr_entry *se, *tse; - - TAILQ_FOREACH_SAFE(se, &sl->sl_head, se_link, tse) { - sockaddrlist_remove(sl, se); - sockaddrentry_free(se); - } - VERIFY(sl->sl_cnt == 0 && TAILQ_EMPTY(&sl->sl_head)); - zfree(sl_zone, sl); -} - -void -sockaddrlist_insert(struct sockaddr_list *sl, struct sockaddr_entry *se) -{ - VERIFY(!(se->se_flags & SEF_ATTACHED)); - se->se_flags |= SEF_ATTACHED; - TAILQ_INSERT_TAIL(&sl->sl_head, se, se_link); - sl->sl_cnt++; - VERIFY(sl->sl_cnt != 0); -} - -void -sockaddrlist_remove(struct sockaddr_list *sl, struct sockaddr_entry *se) -{ - VERIFY(se->se_flags & SEF_ATTACHED); - se->se_flags &= ~SEF_ATTACHED; - VERIFY(sl->sl_cnt != 0); - sl->sl_cnt--; - TAILQ_REMOVE(&sl->sl_head, se, se_link); -} - -struct sockaddr_list * -sockaddrlist_dup(const struct sockaddr_list *src_sl, int how) -{ - struct sockaddr_entry *src_se, *tse; - struct sockaddr_list *dst_sl; - - dst_sl = sockaddrlist_alloc(how); - if (dst_sl == NULL) - return (NULL); - - TAILQ_FOREACH_SAFE(src_se, &src_sl->sl_head, se_link, tse) { - struct sockaddr_entry *dst_se; - - if (src_se->se_addr == NULL) - continue; - - dst_se = sockaddrentry_dup(src_se, how); - if (dst_se == NULL) { - sockaddrlist_free(dst_sl); - return (NULL); - } - - sockaddrlist_insert(dst_sl, dst_se); - } - VERIFY(src_sl->sl_cnt == dst_sl->sl_cnt); - - return (dst_sl); + SO_RESTRICT_DENY_OUT | + SO_RESTRICT_DENY_CELLULAR | SO_RESTRICT_DENY_EXPENSIVE)); } int @@ -5227,7 +7426,6 @@ so_set_effective_pid(struct socket *so, int epid, struct proc *p) so->e_pid = proc_pid(ep); proc_getexecutableuuid(ep, so->e_uuid, sizeof (so->e_uuid)); } - done: if (error == 0 && net_io_policy_log) { uuid_string_t buf; @@ -5235,17 +7433,28 @@ done: uuid_unparse(so->e_uuid, buf); log(LOG_DEBUG, "%s[%s,%d]: so 0x%llx [%d,%d] epid %d (%s) " "euuid %s%s\n", __func__, proc_name_address(p), - proc_pid(p), (uint64_t)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), - SOCK_TYPE(so), so->e_pid, proc_name_address(ep), buf, + 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)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), - SOCK_TYPE(so), epid, (ep == PROC_NULL) ? "PROC_NULL" : + 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 */ + } + if (ep != PROC_NULL) proc_rele(ep); @@ -5262,7 +7471,7 @@ so_set_effective_uuid(struct socket *so, uuid_t euuid, struct proc *p) /* UUID must not be all-zeroes (reserved for kernel) */ if (uuid_is_null(euuid)) { error = EINVAL; - goto done;; + goto done; } /* @@ -5326,17 +7535,26 @@ done: uuid_unparse(so->e_uuid, buf); log(LOG_DEBUG, "%s[%s,%d]: so 0x%llx [%d,%d] epid %d " "euuid %s%s\n", __func__, proc_name_address(p), proc_pid(p), - (uint64_t)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), + (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)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), + (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); } @@ -5364,3 +7582,50 @@ netpolicy_post_msg(uint32_t ev_code, struct netpolicy_event_data *ev_data, kev_post_msg(&ev_msg); } + +void +socket_post_kev_msg(uint32_t ev_code, + struct kev_socket_event_data *ev_data, + uint32_t ev_datalen) +{ + struct kev_msg ev_msg; + + bzero(&ev_msg, sizeof(ev_msg)); + ev_msg.vendor_code = KEV_VENDOR_APPLE; + ev_msg.kev_class = KEV_NETWORK_CLASS; + ev_msg.kev_subclass = KEV_SOCKET_SUBCLASS; + ev_msg.event_code = ev_code; + + ev_msg.dv[0].data_ptr = ev_data; + ev_msg.dv[0]. data_length = ev_datalen; + + kev_post_msg(&ev_msg); +} + +void +socket_post_kev_msg_closed(struct socket *so) +{ + struct kev_socket_closed ev; + struct sockaddr *socksa = NULL, *peersa = NULL; + int err; + bzero(&ev, sizeof(ev)); + err = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, &socksa); + if (err == 0) { + err = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, + &peersa); + if (err == 0) { + memcpy(&ev.ev_data.kev_sockname, socksa, + min(socksa->sa_len, + sizeof (ev.ev_data.kev_sockname))); + memcpy(&ev.ev_data.kev_peername, peersa, + min(peersa->sa_len, + sizeof (ev.ev_data.kev_peername))); + socket_post_kev_msg(KEV_SOCKET_CLOSED, + &ev.ev_data, sizeof (ev)); + } + } + if (socksa != NULL) + FREE(socksa, M_SONAME); + if (peersa != NULL) + FREE(peersa, M_SONAME); +}