X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/935ed37a5c468c8a1c07408573c08b8b7ef80e8b..490019cf9519204c5fb36b2fba54ceb983bb6b72:/bsd/netinet/ip_divert.c diff --git a/bsd/netinet/ip_divert.c b/bsd/netinet/ip_divert.c index 73ab3ea91..614c0c7c1 100644 --- a/bsd/netinet/ip_divert.c +++ b/bsd/netinet/ip_divert.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2007 Apple Inc. All rights reserved. + * Copyright (c) 2000-2013 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -76,6 +76,7 @@ #include #include +#include #include #include @@ -131,50 +132,64 @@ static struct inpcbhead divcb; static struct inpcbinfo divcbinfo; -static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ -static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ +static u_int32_t div_sendspace = DIVSNDQ; /* XXX sysctl ? */ +static u_int32_t div_recvspace = DIVRCVQ; /* XXX sysctl ? */ /* Optimization: have this preinitialized */ static struct sockaddr_in divsrc = { sizeof(divsrc), AF_INET, 0, { 0 }, { 0,0,0,0,0,0,0,0 } }; /* Internal functions */ static int div_output(struct socket *so, - struct mbuf *m, struct sockaddr *addr, struct mbuf *control); + struct mbuf *m, struct sockaddr_in *addr, struct mbuf *control); extern int load_ipfw(void); /* * Initialize divert connection block queue. */ void -div_init(void) +div_init(struct protosw *pp, struct domain *dp) { +#pragma unused(dp) + static int div_initialized = 0; struct inpcbinfo *pcbinfo; + + VERIFY((pp->pr_flags & (PR_INITIALIZED|PR_ATTACHED)) == PR_ATTACHED); + + if (div_initialized) + return; + div_initialized = 1; + LIST_INIT(&divcb); - divcbinfo.listhead = &divcb; + divcbinfo.ipi_listhead = &divcb; /* * XXX We don't use the hash list for divert IP, but it's easier * to allocate a one entry hash list than it is to check all - * over the place for hashbase == NULL. + * over the place for ipi_hashbase == NULL. */ - divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask); - divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask); - divcbinfo.ipi_zone = (void *) zinit(sizeof(struct inpcb),(maxsockets * sizeof(struct inpcb)), + divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &divcbinfo.ipi_hashmask); + divcbinfo.ipi_porthashbase = hashinit(1, M_PCB, &divcbinfo.ipi_porthashmask); + divcbinfo.ipi_zone = zinit(sizeof(struct inpcb),(512 * sizeof(struct inpcb)), 4096, "divzone"); pcbinfo = &divcbinfo; /* * allocate lock group attribute and group for udp pcb mutexes */ - pcbinfo->mtx_grp_attr = lck_grp_attr_alloc_init(); + pcbinfo->ipi_lock_grp_attr = lck_grp_attr_alloc_init(); + + pcbinfo->ipi_lock_grp = lck_grp_alloc_init("divcb", pcbinfo->ipi_lock_grp_attr); - pcbinfo->mtx_grp = lck_grp_alloc_init("divcb", pcbinfo->mtx_grp_attr); - /* * allocate the lock attribute for divert pcb mutexes */ - pcbinfo->mtx_attr = lck_attr_alloc_init(); + pcbinfo->ipi_lock_attr = lck_attr_alloc_init(); + + if ((pcbinfo->ipi_lock = lck_rw_alloc_init(pcbinfo->ipi_lock_grp, + pcbinfo->ipi_lock_attr)) == NULL) { + panic("%s: unable to allocate PCB lock\n", __func__); + /* NOTREACHED */ + } - if ((pcbinfo->mtx = lck_rw_alloc_init(pcbinfo->mtx_grp, pcbinfo->mtx_attr)) == NULL) - return; /* pretty much dead if this fails... */ + in_pcbinfo_attach(&divcbinfo); #if IPFIREWALL if (!IPFW_LOADED) { @@ -190,7 +205,7 @@ div_init(void) void div_input(struct mbuf *m, __unused int off) { - OSAddAtomic(1, (SInt32*)&ipstat.ips_noproto); + OSAddAtomic(1, &ipstat.ips_noproto); m_freem(m); } @@ -199,7 +214,6 @@ div_input(struct mbuf *m, __unused int off) * * Setup generic address and protocol structures for div_input routine, * then pass them along with mbuf chain. - * ###LOCK called in ip_mutex from ip_output/ip_input */ void divert_packet(struct mbuf *m, int incoming, int port, int rule) @@ -235,12 +249,14 @@ divert_packet(struct mbuf *m, int incoming, int port, int rule) /* Find IP address for receive interface */ ifnet_lock_shared(m->m_pkthdr.rcvif); TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) { - if (ifa->ifa_addr == NULL) - continue; - if (ifa->ifa_addr->sa_family != AF_INET) + IFA_LOCK(ifa); + if (ifa->ifa_addr->sa_family != AF_INET) { + IFA_UNLOCK(ifa); continue; + } divsrc.sin_addr = - ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; + ((struct sockaddr_in *)(void *) ifa->ifa_addr)->sin_addr; + IFA_UNLOCK(ifa); break; } ifnet_lock_done(m->m_pkthdr.rcvif); @@ -269,14 +285,13 @@ divert_packet(struct mbuf *m, int incoming, int port, int rule) * (see div_output for the other half of this.) */ snprintf(divsrc.sin_zero, sizeof(divsrc.sin_zero), - "%s%d", m->m_pkthdr.rcvif->if_name, - m->m_pkthdr.rcvif->if_unit); + "%s", if_name(m->m_pkthdr.rcvif)); } /* Put packet on socket queue, if any */ sa = NULL; nport = htons((u_int16_t)port); - lck_rw_lock_shared(divcbinfo.mtx); + lck_rw_lock_shared(divcbinfo.ipi_lock); LIST_FOREACH(inp, &divcb, inp_list) { if (inp->inp_lport == nport) sa = inp->inp_socket; @@ -291,10 +306,10 @@ divert_packet(struct mbuf *m, int incoming, int port, int rule) socket_unlock(sa, 1); } else { m_freem(m); - OSAddAtomic(1, (SInt32*)&ipstat.ips_noproto); - OSAddAtomic(-1, (SInt32*)&ipstat.ips_delivered); + OSAddAtomic(1, &ipstat.ips_noproto); + OSAddAtomic(-1, &ipstat.ips_delivered); } - lck_rw_done(divcbinfo.mtx); + lck_rw_done(divcbinfo.ipi_lock); } /* @@ -306,17 +321,20 @@ divert_packet(struct mbuf *m, int incoming, int port, int rule) * ###LOCK called in inet_proto mutex when from div_send. */ static int -div_output(struct socket *so, struct mbuf *m, struct sockaddr *addr, +div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, struct mbuf *control) { struct inpcb *const inp = sotoinpcb(so); struct ip *const ip = mtod(m, struct ip *); - struct sockaddr_in *sin = (struct sockaddr_in *)addr; int error = 0; + mbuf_svc_class_t msc = MBUF_SC_UNSPEC; - if (control) - m_freem(control); /* XXX */ + if (control != NULL) { + msc = mbuf_service_class_from_control(control); + m_freem(control); /* XXX */ + control = NULL; + } /* Loopback avoidance and state recovery */ if (sin) { struct m_tag *mtag; @@ -324,8 +342,8 @@ div_output(struct socket *so, struct mbuf *m, struct sockaddr *addr, int len = 0; char *c = sin->sin_zero; - mtag = m_tag_alloc(KERNEL_MODULE_TAG_ID, KERNEL_TAG_TYPE_DIVERT, - sizeof(struct divert_tag), M_NOWAIT); + mtag = m_tag_create(KERNEL_MODULE_TAG_ID, KERNEL_TAG_TYPE_DIVERT, + sizeof(struct divert_tag), M_NOWAIT, m); if (mtag == NULL) { error = ENOBUFS; goto cantsend; @@ -351,6 +369,11 @@ div_output(struct socket *so, struct mbuf *m, struct sockaddr *addr, /* Reinject packet into the system as incoming or outgoing */ if (!sin || sin->sin_addr.s_addr == 0) { + struct ip_out_args ipoa = + { IFSCOPE_NONE, { 0 }, IPOAF_SELECT_SRCIF, 0 }; + struct route ro; + struct ip_moptions *imo; + /* * Don't allow both user specified and setsockopt options, * and don't allow packet length sizes that will crash @@ -362,40 +385,55 @@ div_output(struct socket *so, struct mbuf *m, struct sockaddr *addr, } /* Convert fields to host order for ip_output() */ +#if BYTE_ORDER != BIG_ENDIAN NTOHS(ip->ip_len); NTOHS(ip->ip_off); +#endif - /* Send packet to output processing */ - OSAddAtomic(1, (SInt32*)&ipstat.ips_rawout); + OSAddAtomic(1, &ipstat.ips_rawout); + /* Copy the cached route and take an extra reference */ + inp_route_copyout(inp, &ro); + + set_packet_service_class(m, so, msc, 0); + + imo = inp->inp_moptions; + if (imo != NULL) + IMO_ADDREF(imo); socket_unlock(so, 0); #if CONFIG_MACF_NET mac_mbuf_label_associate_inpcb(inp, m); #endif - error = ip_output(m, - inp->inp_options, &inp->inp_route, + /* Send packet to output processing */ + error = ip_output(m, inp->inp_options, &ro, (so->so_options & SO_DONTROUTE) | - IP_ALLOWBROADCAST | IP_RAWOUTPUT, - inp->inp_moptions, NULL); + IP_ALLOWBROADCAST | IP_RAWOUTPUT | IP_OUTARGS, + imo, &ipoa); + socket_lock(so, 0); + if (imo != NULL) + IMO_REMREF(imo); + /* Synchronize cached PCB route */ + inp_route_copyin(inp, &ro); } else { struct ifaddr *ifa; /* If no luck with the name above. check by IP address. */ if (m->m_pkthdr.rcvif == NULL) { + struct sockaddr_in _sin; /* - * Make sure there are no distractions - * for ifa_ifwithaddr. Clear the port and the ifname. - * Maybe zap all 8 bytes at once using a 64bit write? + * Make sure there are no distractions for + * ifa_ifwithaddr; use sanitized version. */ - bzero(sin->sin_zero, sizeof(sin->sin_zero)); - /* *((u_int64_t *)sin->sin_zero) = 0; */ /* XXX ?? */ - sin->sin_port = 0; - if (!(ifa = ifa_ifwithaddr((struct sockaddr *) sin))) { + bzero(&_sin, sizeof (_sin)); + _sin.sin_family = AF_INET; + _sin.sin_len = sizeof (struct sockaddr_in); + _sin.sin_addr.s_addr = sin->sin_addr.s_addr; + if (!(ifa = ifa_ifwithaddr(SA(&_sin)))) { error = EADDRNOTAVAIL; goto cantsend; } m->m_pkthdr.rcvif = ifa->ifa_ifp; - ifafree(ifa); + IFA_REMREF(ifa); } #if CONFIG_MACF_NET mac_mbuf_label_associate_socket(so, m); @@ -421,7 +459,7 @@ div_attach(struct socket *so, int proto, struct proc *p) inp = sotoinpcb(so); if (inp) panic("div_attach"); - if (p && (error = proc_suser(p)) != 0) + if ((error = proc_suser(p)) != 0) return error; error = soreserve(so, div_sendspace, div_recvspace); @@ -439,8 +477,11 @@ div_attach(struct socket *so, int proto, struct proc *p) so->so_state |= SS_ISCONNECTED; #ifdef MORE_DICVLOCK_DEBUG - printf("div_attach: so=%p sopcb=%p lock=%x ref=%x\n", - so, so->so_pcb, ((struct inpcb *)so->so_pcb)->inpcb_mtx, so->so_usecount); + printf("div_attach: so=0x%llx sopcb=0x%llx lock=0x%llx ref=%x\n", + (uint64_t)VM_KERNEL_ADDRPERM(so), + (uint64_t)VM_KERNEL_ADDRPERM(so->so_pcb), + (uint64_t)VM_KERNEL_ADDRPERM(&(sotoinpcb(so)->inpcb_mtx)), + so->so_usecount); #endif return 0; } @@ -451,8 +492,11 @@ div_detach(struct socket *so) struct inpcb *inp; #ifdef MORE_DICVLOCK_DEBUG - printf("div_detach: so=%p sopcb=%p lock=%x ref=%x\n", - so, so->so_pcb, ((struct inpcb *)so->so_pcb)->inpcb_mtx, so->so_usecount); + printf("div_detach: so=0x%llx sopcb=0x%llx lock=0x%llx ref=%x\n", + (uint64_t)VM_KERNEL_ADDRPERM(so), + (uint64_t)VM_KERNEL_ADDRPERM(so->so_pcb), + (uint64_t)VM_KERNEL_ADDRPERM(&(sotoinpcb(so)->inpcb_mtx)), + so->so_usecount); #endif inp = sotoinpcb(so); if (inp == 0) @@ -494,7 +538,7 @@ div_bind(struct socket *so, struct sockaddr *nam, struct proc *p) if (nam->sa_family != AF_INET) { error = EAFNOSUPPORT; } else { - ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; + ((struct sockaddr_in *)(void *)nam)->sin_addr.s_addr = INADDR_ANY; error = in_pcbbind(inp, nam, p); } return error; @@ -514,18 +558,20 @@ div_send(struct socket *so, __unused int flags, struct mbuf *m, struct sockaddr /* Packet must have a header (but that's about it) */ if (m->m_len < sizeof (struct ip) && (m = m_pullup(m, sizeof (struct ip))) == 0) { - OSAddAtomic(1, (SInt32*)&ipstat.ips_toosmall); + OSAddAtomic(1, &ipstat.ips_toosmall); m_freem(m); return EINVAL; } /* Send packet */ - return div_output(so, m, nam, control); + return div_output(so, m, SIN(nam), control); } +#if 0 static int div_pcblist SYSCTL_HANDLER_ARGS { +#pragma unused(oidp, arg1, arg2) int error, i, n; struct inpcb *inp, **inp_list; inp_gen_t gencnt; @@ -535,17 +581,17 @@ div_pcblist SYSCTL_HANDLER_ARGS * The process of preparing the TCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ - lck_rw_lock_exclusive(divcbinfo.mtx); + lck_rw_lock_exclusive(divcbinfo.ipi_lock); if (req->oldptr == USER_ADDR_NULL) { n = divcbinfo.ipi_count; req->oldidx = 2 * (sizeof xig) + (n + n/8) * sizeof(struct xinpcb); - lck_rw_done(divcbinfo.mtx); + lck_rw_done(divcbinfo.ipi_lock); return 0; } if (req->newptr != USER_ADDR_NULL) { - lck_rw_done(divcbinfo.mtx); + lck_rw_done(divcbinfo.ipi_lock); return EPERM; } @@ -562,17 +608,17 @@ div_pcblist SYSCTL_HANDLER_ARGS xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) { - lck_rw_done(divcbinfo.mtx); + lck_rw_done(divcbinfo.ipi_lock); return error; } inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK); if (inp_list == 0) { - lck_rw_done(divcbinfo.mtx); + lck_rw_done(divcbinfo.ipi_lock); return ENOMEM; } - for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n; + for (inp = LIST_FIRST(divcbinfo.ipi_listhead), i = 0; inp && i < n; inp = LIST_NEXT(inp, inp_list)) { #ifdef __APPLE__ if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) @@ -614,85 +660,95 @@ div_pcblist SYSCTL_HANDLER_ARGS error = SYSCTL_OUT(req, &xig, sizeof xig); } FREE(inp_list, M_TEMP); - lck_rw_done(divcbinfo.mtx); + lck_rw_done(divcbinfo.ipi_lock); return error; } +#endif __private_extern__ int -div_lock(struct socket *so, int refcount, int lr) - { - int lr_saved; - if (lr == 0) - lr_saved = (unsigned int) __builtin_return_address(0); - else lr_saved = lr; - +div_lock(struct socket *so, int refcount, void *lr) +{ + void *lr_saved; + + if (lr == NULL) + lr_saved = __builtin_return_address(0); + else + lr_saved = lr; + #ifdef MORE_DICVLOCK_DEBUG - printf("div_lock: so=%p sopcb=%p lock=%x ref=%x lr=%x\n", - so, - so->so_pcb, - so->so_pcb ? ((struct inpcb *)so->so_pcb)->inpcb_mtx : 0, - so->so_usecount, - lr_saved); + printf("div_lock: so=0x%llx sopcb=0x%llx lock=0x%llx ref=%x " + "lr=0x%llx\n", (uint64_t)VM_KERNEL_ADDRPERM(so), + (uint64_t)VM_KERNEL_ADDRPERM(so->so_pcb), so->so_pcb ? + (uint64_t)VM_KERNEL_ADDRPERM(&(sotoinpcb(so)->inpcb_mtx)) : NULL, + so->so_usecount, (uint64_t)VM_KERNEL_ADDRPERM(lr_saved)); #endif if (so->so_pcb) { - lck_mtx_lock(((struct inpcb *)so->so_pcb)->inpcb_mtx); + lck_mtx_lock(&((struct inpcb *)so->so_pcb)->inpcb_mtx); } else { - panic("div_lock: so=%p NO PCB! lr=%x\n", so, lr_saved); - lck_mtx_lock(so->so_proto->pr_domain->dom_mtx); + panic("div_lock: so=%p NO PCB! lr=%p lrh= lrh= %s\n", + so, lr_saved, solockhistory_nr(so)); + /* NOTREACHED */ } - - if (so->so_usecount < 0) - panic("div_lock: so=%p so_pcb=%p lr=%x ref=%x\n", - so, so->so_pcb, lr_saved, so->so_usecount); - + + if (so->so_usecount < 0) { + panic("div_lock: so=%p so_pcb=%p lr=%p ref=%x lrh= %s\n", + so, so->so_pcb, lr_saved, so->so_usecount, + solockhistory_nr(so)); + /* NOTREACHED */ + } + if (refcount) so->so_usecount++; - so->lock_lr[so->next_lock_lr] = (u_int32_t)lr_saved; + so->lock_lr[so->next_lock_lr] = lr_saved; so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX; return (0); } __private_extern__ int -div_unlock(struct socket *so, int refcount, int lr) +div_unlock(struct socket *so, int refcount, void *lr) { - int lr_saved; + void *lr_saved; lck_mtx_t * mutex_held; - struct inpcb *inp = sotoinpcb(so); + struct inpcb *inp = sotoinpcb(so); - if (lr == 0) - lr_saved = (unsigned int) __builtin_return_address(0); - else lr_saved = lr; + if (lr == NULL) + lr_saved = __builtin_return_address(0); + else + lr_saved = lr; - #ifdef MORE_DICVLOCK_DEBUG - printf("div_unlock: so=%p sopcb=%p lock=%x ref=%x lr=%x\n", - so, - so->so_pcb, - so->so_pcb ? ((struct inpcb *)so->so_pcb)->inpcb_mtx : 0, - so->so_usecount, - lr_saved); + printf("div_unlock: so=0x%llx sopcb=0x%llx lock=0x%llx ref=%x " + "lr=0x%llx\n", (uint64_t)VM_KERNEL_ADDRPERM(so), + (uint64_t)VM_KERNEL_ADDRPERM(so->so_pcb), so->so_pcb ? + (uint64_t)VM_KERNEL_ADDRPERM(&(sotoinpcb(so)->inpcb_mtx)) : NULL, + so->so_usecount, lr_saved); #endif if (refcount) so->so_usecount--; - - if (so->so_usecount < 0) - panic("div_unlock: so=%p usecount=%x\n", so, so->so_usecount); + + if (so->so_usecount < 0) { + panic("div_unlock: so=%p usecount=%x lrh= %s\n", + so, so->so_usecount, solockhistory_nr(so)); + /* NOTREACHED */ + } if (so->so_pcb == NULL) { - panic("div_unlock: so=%p NO PCB usecount=%x lr=%x\n", so, so->so_usecount, lr_saved); - mutex_held = so->so_proto->pr_domain->dom_mtx; - } else { - mutex_held = ((struct inpcb *)so->so_pcb)->inpcb_mtx; + panic("div_unlock: so=%p NO PCB usecount=%x lr=%p lrh= %s\n", + so, so->so_usecount, lr_saved, solockhistory_nr(so)); + /* NOTREACHED */ } + mutex_held = &((struct inpcb *)so->so_pcb)->inpcb_mtx; if (so->so_usecount == 0 && (inp->inp_wantcnt == WNT_STOPUSING)) { - lck_rw_lock_exclusive(divcbinfo.mtx); + lck_rw_lock_exclusive(divcbinfo.ipi_lock); + if (inp->inp_state != INPCB_STATE_DEAD) + in_pcbdetach(inp); in_pcbdispose(inp); - lck_rw_done(divcbinfo.mtx); + lck_rw_done(divcbinfo.ipi_lock); return (0); } lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); - so->unlock_lr[so->next_unlock_lr] = (u_int32_t) lr_saved; + so->unlock_lr[so->next_unlock_lr] = lr_saved; so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; lck_mtx_unlock(mutex_held); return (0); @@ -705,20 +761,28 @@ div_getlock(struct socket *so, __unused int locktype) if (so->so_pcb) { if (so->so_usecount < 0) - panic("div_getlock: so=%p usecount=%x\n", so, so->so_usecount); - return(inpcb->inpcb_mtx); + panic("div_getlock: so=%p usecount=%x lrh= %s\n", + so, so->so_usecount, solockhistory_nr(so)); + return(&inpcb->inpcb_mtx); } else { - panic("div_getlock: so=%p NULL so_pcb\n", so); + panic("div_getlock: so=%p NULL NO PCB lrh= %s\n", + so, solockhistory_nr(so)); return (so->so_proto->pr_domain->dom_mtx); } } - struct pr_usrreqs div_usrreqs = { - div_abort, pru_accept_notsupp, div_attach, div_bind, - pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach, - div_disconnect, pru_listen_notsupp, in_setpeeraddr, pru_rcvd_notsupp, - pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown, - in_setsockaddr, sosend, soreceive, pru_sopoll_notsupp + .pru_abort = div_abort, + .pru_attach = div_attach, + .pru_bind = div_bind, + .pru_control = in_control, + .pru_detach = div_detach, + .pru_disconnect = div_disconnect, + .pru_peeraddr = in_getpeeraddr, + .pru_send = div_send, + .pru_shutdown = div_shutdown, + .pru_sockaddr = in_getsockaddr, + .pru_sosend = sosend, + .pru_soreceive = soreceive, };