diff --git a/sys/kern/uipc_domain.c b/sys/kern/uipc_domain.c index 8c6bd93ae703..3c0a570842b0 100644 --- a/sys/kern/uipc_domain.c +++ b/sys/kern/uipc_domain.c @@ -1,495 +1,485 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)uipc_domain.c 8.2 (Berkeley) 10/18/93 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * System initialization * * Note: domain initialization takes place on a per domain basis * as a result of traversing a SYSINIT linker set. Most likely, * each domain would want to call DOMAIN_SET(9) itself, which * would cause the domain to be added just after domaininit() * is called during startup. * * See DOMAIN_SET(9) for details on its use. */ static void domaininit(void *); SYSINIT(domain, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, domaininit, NULL); static void domainfinalize(void *); SYSINIT(domainfin, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST, domainfinalize, NULL); struct domain *domains; /* registered protocol domains */ int domain_init_status = 0; static struct mtx dom_mtx; /* domain list lock */ MTX_SYSINIT(domain, &dom_mtx, "domain list", MTX_DEF); static int pr_accept_notsupp(struct socket *so, struct sockaddr **nam) { return (EOPNOTSUPP); } static int pr_aio_queue_notsupp(struct socket *so, struct kaiocb *job) { return (EOPNOTSUPP); } static int pr_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) { return (EOPNOTSUPP); } static int pr_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { return (EOPNOTSUPP); } static int pr_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) { return (EOPNOTSUPP); } static int pr_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { return (EOPNOTSUPP); } static int pr_connect2_notsupp(struct socket *so1, struct socket *so2) { return (EOPNOTSUPP); } static int pr_control_notsupp(struct socket *so, u_long cmd, void *data, struct ifnet *ifp, struct thread *td) { return (EOPNOTSUPP); } static int pr_disconnect_notsupp(struct socket *so) { return (EOPNOTSUPP); } static int pr_listen_notsupp(struct socket *so, int backlog, struct thread *td) { return (EOPNOTSUPP); } static int pr_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) { return (EOPNOTSUPP); } static int pr_rcvd_notsupp(struct socket *so, int flags) { return (EOPNOTSUPP); } static int pr_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) { return (EOPNOTSUPP); } static int pr_send_notsupp(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, struct mbuf *control, struct thread *td) { if (control != NULL) m_freem(control); if ((flags & PRUS_NOTREADY) == 0) m_freem(m); return (EOPNOTSUPP); } static int pr_ready_notsupp(struct socket *so, struct mbuf *m, int count) { return (EOPNOTSUPP); } static int pr_shutdown_notsupp(struct socket *so) { return (EOPNOTSUPP); } static int pr_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) { return (EOPNOTSUPP); } static int pr_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *top, struct mbuf *control, int flags, struct thread *td) { return (EOPNOTSUPP); } static int pr_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { return (EOPNOTSUPP); } static int pr_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, struct thread *td) { return (EOPNOTSUPP); } static void pr_init(struct protosw *pr) { KASSERT(pr->pr_attach != NULL, ("%s: protocol doesn't have pr_attach", __func__)); #define DEFAULT(foo, bar) if (pr->foo == NULL) pr->foo = bar DEFAULT(pr_sosend, sosend_generic); DEFAULT(pr_soreceive, soreceive_generic); DEFAULT(pr_sopoll, sopoll_generic); #define NOTSUPP(foo) if (pr->foo == NULL) pr->foo = foo ## _notsupp NOTSUPP(pr_accept); NOTSUPP(pr_aio_queue); NOTSUPP(pr_bind); NOTSUPP(pr_bindat); NOTSUPP(pr_connect); NOTSUPP(pr_connect2); NOTSUPP(pr_connectat); NOTSUPP(pr_control); NOTSUPP(pr_disconnect); NOTSUPP(pr_listen); NOTSUPP(pr_peeraddr); NOTSUPP(pr_rcvd); NOTSUPP(pr_rcvoob); NOTSUPP(pr_send); NOTSUPP(pr_shutdown); NOTSUPP(pr_sockaddr); NOTSUPP(pr_sosend); NOTSUPP(pr_soreceive); NOTSUPP(pr_sopoll); NOTSUPP(pr_ready); } /* * Add a new protocol domain to the list of supported domains * Note: you cant unload it again because a socket may be using it. * XXX can't fail at this time. */ void domain_init(struct domain *dp) { struct protosw *pr; int flags; MPASS(IS_DEFAULT_VNET(curvnet)); flags = atomic_load_acq_int(&dp->dom_flags); if ((flags & DOMF_SUPPORTED) == 0) return; MPASS((flags & DOMF_INITED) == 0); for (int i = 0; i < dp->dom_nprotosw; i++) if ((pr = dp->dom_protosw[i]) != NULL) { pr->pr_domain = dp; pr_init(pr); } - /* - * update global information about maximums - */ - max_hdr = max_linkhdr + max_protohdr; - max_datalen = MHLEN - max_hdr; - if (max_datalen < 1) - panic("%s: max_datalen < 1", __func__); atomic_set_rel_int(&dp->dom_flags, DOMF_INITED); } /* * Add a new protocol domain to the list of supported domains * Note: you cant unload it again because a socket may be using it. * XXX can't fail at this time. */ void domain_add(struct domain *dp) { if (dp->dom_probe != NULL && (*dp->dom_probe)() != 0) return; atomic_set_rel_int(&dp->dom_flags, DOMF_SUPPORTED); mtx_lock(&dom_mtx); dp->dom_next = domains; domains = dp; KASSERT(domain_init_status >= 1, ("attempt to domain_add(%s) before domaininit()", dp->dom_name)); #ifndef INVARIANTS if (domain_init_status < 1) printf("WARNING: attempt to domain_add(%s) before " "domaininit()\n", dp->dom_name); #endif mtx_unlock(&dom_mtx); } void domain_remove(struct domain *dp) { if ((dp->dom_flags & DOMF_UNLOADABLE) == 0) return; mtx_lock(&dom_mtx); if (domains == dp) { domains = dp->dom_next; } else { struct domain *curr; for (curr = domains; curr != NULL; curr = curr->dom_next) { if (curr->dom_next == dp) { curr->dom_next = dp->dom_next; break; } } } mtx_unlock(&dom_mtx); } /* ARGSUSED*/ static void domaininit(void *dummy) { - if (max_linkhdr < 16) /* XXX */ - max_linkhdr = 16; - mtx_lock(&dom_mtx); KASSERT(domain_init_status == 0, ("domaininit called too late!")); domain_init_status = 1; mtx_unlock(&dom_mtx); } /* ARGSUSED*/ static void domainfinalize(void *dummy) { mtx_lock(&dom_mtx); KASSERT(domain_init_status == 1, ("domainfinalize called too late!")); domain_init_status = 2; mtx_unlock(&dom_mtx); } struct domain * pffinddomain(int family) { struct domain *dp; for (dp = domains; dp != NULL; dp = dp->dom_next) if (dp->dom_family == family) return (dp); return (NULL); } struct protosw * pffindtype(int family, int type) { struct domain *dp; struct protosw *pr; dp = pffinddomain(family); if (dp == NULL) return (NULL); for (int i = 0; i < dp->dom_nprotosw; i++) if ((pr = dp->dom_protosw[i]) != NULL && pr->pr_type == type) return (pr); return (NULL); } struct protosw * pffindproto(int family, int protocol, int type) { struct domain *dp; struct protosw *pr; struct protosw *maybe; dp = pffinddomain(family); if (dp == NULL) return (NULL); maybe = NULL; for (int i = 0; i < dp->dom_nprotosw; i++) { if ((pr = dp->dom_protosw[i]) == NULL) continue; if ((pr->pr_protocol == protocol) && (pr->pr_type == type)) return (pr); /* XXX: raw catches all. Why? */ if (type == SOCK_RAW && pr->pr_type == SOCK_RAW && pr->pr_protocol == 0 && maybe == NULL) maybe = pr; } return (maybe); } /* * The caller must make sure that the new protocol is fully set up and ready to * accept requests before it is registered. */ int protosw_register(struct domain *dp, struct protosw *npr) { struct protosw **prp; MPASS(dp); MPASS(npr && npr->pr_type > 0 && npr->pr_protocol > 0); prp = NULL; /* * Protect us against races when two protocol registrations for * the same protocol happen at the same time. */ mtx_lock(&dom_mtx); for (int i = 0; i < dp->dom_nprotosw; i++) { if (dp->dom_protosw[i] == NULL) { /* Remember the first free spacer. */ if (prp == NULL) prp = &dp->dom_protosw[i]; } else { /* * The new protocol must not yet exist. * XXXAO: Check only protocol? * XXXGL: Maybe assert that it doesn't exist? */ if ((dp->dom_protosw[i]->pr_type == npr->pr_type) && (dp->dom_protosw[i]->pr_protocol == npr->pr_protocol)) { mtx_unlock(&dom_mtx); return (EEXIST); } } } /* If no free spacer is found we can't add the new protocol. */ if (prp == NULL) { mtx_unlock(&dom_mtx); return (ENOMEM); } npr->pr_domain = dp; pr_init(npr); *prp = npr; mtx_unlock(&dom_mtx); return (0); } /* * The caller must make sure the protocol and its functions correctly shut down * all sockets and release all locks and memory references. */ int protosw_unregister(struct protosw *pr) { struct domain *dp; struct protosw **prp; dp = pr->pr_domain; prp = NULL; mtx_lock(&dom_mtx); /* The protocol must exist and only once. */ for (int i = 0; i < dp->dom_nprotosw; i++) { if (dp->dom_protosw[i] == pr) { KASSERT(prp == NULL, ("%s: domain %p protocol %p registered twice\n", __func__, dp, pr)); prp = &dp->dom_protosw[i]; } } /* Protocol does not exist. XXXGL: assert that it does? */ if (prp == NULL) { mtx_unlock(&dom_mtx); return (EPROTONOSUPPORT); } /* De-orbit the protocol and make the slot available again. */ *prp = NULL; mtx_unlock(&dom_mtx); return (0); } diff --git a/sys/kern/uipc_mbuf.c b/sys/kern/uipc_mbuf.c index 17fa670c9406..3f400d3563c4 100644 --- a/sys/kern/uipc_mbuf.c +++ b/sys/kern/uipc_mbuf.c @@ -1,2259 +1,2285 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1991, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_param.h" #include "opt_mbuf_stress_test.h" #include "opt_mbuf_profiling.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include SDT_PROBE_DEFINE5_XLATE(sdt, , , m__init, "struct mbuf *", "mbufinfo_t *", "uint32_t", "uint32_t", "uint16_t", "uint16_t", "uint32_t", "uint32_t", "uint32_t", "uint32_t"); SDT_PROBE_DEFINE3_XLATE(sdt, , , m__gethdr_raw, "uint32_t", "uint32_t", "uint16_t", "uint16_t", "struct mbuf *", "mbufinfo_t *"); SDT_PROBE_DEFINE3_XLATE(sdt, , , m__gethdr, "uint32_t", "uint32_t", "uint16_t", "uint16_t", "struct mbuf *", "mbufinfo_t *"); SDT_PROBE_DEFINE3_XLATE(sdt, , , m__get_raw, "uint32_t", "uint32_t", "uint16_t", "uint16_t", "struct mbuf *", "mbufinfo_t *"); SDT_PROBE_DEFINE3_XLATE(sdt, , , m__get, "uint32_t", "uint32_t", "uint16_t", "uint16_t", "struct mbuf *", "mbufinfo_t *"); SDT_PROBE_DEFINE4_XLATE(sdt, , , m__getcl, "uint32_t", "uint32_t", "uint16_t", "uint16_t", "uint32_t", "uint32_t", "struct mbuf *", "mbufinfo_t *"); SDT_PROBE_DEFINE5_XLATE(sdt, , , m__getjcl, "uint32_t", "uint32_t", "uint16_t", "uint16_t", "uint32_t", "uint32_t", "uint32_t", "uint32_t", "struct mbuf *", "mbufinfo_t *"); SDT_PROBE_DEFINE3_XLATE(sdt, , , m__clget, "struct mbuf *", "mbufinfo_t *", "uint32_t", "uint32_t", "uint32_t", "uint32_t"); SDT_PROBE_DEFINE4_XLATE(sdt, , , m__cljget, "struct mbuf *", "mbufinfo_t *", "uint32_t", "uint32_t", "uint32_t", "uint32_t", "void*", "void*"); SDT_PROBE_DEFINE(sdt, , , m__cljset); SDT_PROBE_DEFINE1_XLATE(sdt, , , m__free, "struct mbuf *", "mbufinfo_t *"); SDT_PROBE_DEFINE1_XLATE(sdt, , , m__freem, "struct mbuf *", "mbufinfo_t *"); #include -int max_linkhdr; -int max_protohdr; -int max_hdr; -int max_datalen; +/* + * Provide minimum possible defaults for link and protocol header space, + * assuming IPv4 over Ethernet. Enabling IPv6, IEEE802.11 or some other + * protocol may grow these values. + */ +u_int max_linkhdr = 16; +u_int max_protohdr = 40; +u_int max_hdr = 16 + 40; +SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD, + &max_linkhdr, 16, "Size of largest link layer header"); +SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD, + &max_protohdr, 40, "Size of largest protocol layer header"); +SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD, + &max_hdr, 16 + 40, "Size of largest link plus protocol header"); + +static void +max_hdr_grow(void) +{ + + max_hdr = max_linkhdr + max_protohdr; + MPASS(max_hdr <= MHLEN); +} + +void +max_linkhdr_grow(u_int new) +{ + + if (new > max_linkhdr) { + max_linkhdr = new; + max_hdr_grow(); + } +} + +void +max_protohdr_grow(u_int new) +{ + + if (new > max_protohdr) { + max_protohdr = new; + max_hdr_grow(); + } +} + #ifdef MBUF_STRESS_TEST int m_defragpackets; int m_defragbytes; int m_defraguseless; int m_defragfailure; int m_defragrandomfailures; -#endif -/* - * sysctl(8) exported objects - */ -SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD, - &max_linkhdr, 0, "Size of largest link layer header"); -SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD, - &max_protohdr, 0, "Size of largest protocol layer header"); -SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD, - &max_hdr, 0, "Size of largest link plus protocol header"); -SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD, - &max_datalen, 0, "Minimum space left in mbuf after max_hdr"); -#ifdef MBUF_STRESS_TEST SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, &m_defragpackets, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, &m_defragbytes, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, &m_defraguseless, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, &m_defragfailure, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, &m_defragrandomfailures, 0, ""); #endif /* * Ensure the correct size of various mbuf parameters. It could be off due * to compiler-induced padding and alignment artifacts. */ CTASSERT(MSIZE - offsetof(struct mbuf, m_dat) == MLEN); CTASSERT(MSIZE - offsetof(struct mbuf, m_pktdat) == MHLEN); /* * mbuf data storage should be 64-bit aligned regardless of architectural * pointer size; check this is the case with and without a packet header. */ CTASSERT(offsetof(struct mbuf, m_dat) % 8 == 0); CTASSERT(offsetof(struct mbuf, m_pktdat) % 8 == 0); /* * While the specific values here don't matter too much (i.e., +/- a few * words), we do want to ensure that changes to these values are carefully * reasoned about and properly documented. This is especially the case as * network-protocol and device-driver modules encode these layouts, and must * be recompiled if the structures change. Check these values at compile time * against the ones documented in comments in mbuf.h. * * NB: Possibly they should be documented there via #define's and not just * comments. */ #if defined(__LP64__) CTASSERT(offsetof(struct mbuf, m_dat) == 32); CTASSERT(sizeof(struct pkthdr) == 64); CTASSERT(sizeof(struct m_ext) == 160); #else CTASSERT(offsetof(struct mbuf, m_dat) == 24); CTASSERT(sizeof(struct pkthdr) == 56); #if defined(__powerpc__) && defined(BOOKE) /* PowerPC booke has 64-bit physical pointers. */ CTASSERT(sizeof(struct m_ext) == 176); #else CTASSERT(sizeof(struct m_ext) == 172); #endif #endif /* * Assert that the queue(3) macros produce code of the same size as an old * plain pointer does. */ #ifdef INVARIANTS static struct mbuf __used m_assertbuf; CTASSERT(sizeof(m_assertbuf.m_slist) == sizeof(m_assertbuf.m_next)); CTASSERT(sizeof(m_assertbuf.m_stailq) == sizeof(m_assertbuf.m_next)); CTASSERT(sizeof(m_assertbuf.m_slistpkt) == sizeof(m_assertbuf.m_nextpkt)); CTASSERT(sizeof(m_assertbuf.m_stailqpkt) == sizeof(m_assertbuf.m_nextpkt)); #endif /* * Attach the cluster from *m to *n, set up m_ext in *n * and bump the refcount of the cluster. */ void mb_dupcl(struct mbuf *n, struct mbuf *m) { volatile u_int *refcnt; KASSERT(m->m_flags & (M_EXT|M_EXTPG), ("%s: M_EXT|M_EXTPG not set on %p", __func__, m)); KASSERT(!(n->m_flags & (M_EXT|M_EXTPG)), ("%s: M_EXT|M_EXTPG set on %p", __func__, n)); /* * Cache access optimization. * * o Regular M_EXT storage doesn't need full copy of m_ext, since * the holder of the 'ext_count' is responsible to carry the free * routine and its arguments. * o M_EXTPG data is split between main part of mbuf and m_ext, the * main part is copied in full, the m_ext part is similar to M_EXT. * o EXT_EXTREF, where 'ext_cnt' doesn't point into mbuf at all, is * special - it needs full copy of m_ext into each mbuf, since any * copy could end up as the last to free. */ if (m->m_flags & M_EXTPG) { bcopy(&m->m_epg_startcopy, &n->m_epg_startcopy, __rangeof(struct mbuf, m_epg_startcopy, m_epg_endcopy)); bcopy(&m->m_ext, &n->m_ext, m_epg_ext_copylen); } else if (m->m_ext.ext_type == EXT_EXTREF) bcopy(&m->m_ext, &n->m_ext, sizeof(struct m_ext)); else bcopy(&m->m_ext, &n->m_ext, m_ext_copylen); n->m_flags |= m->m_flags & (M_RDONLY | M_EXT | M_EXTPG); /* See if this is the mbuf that holds the embedded refcount. */ if (m->m_ext.ext_flags & EXT_FLAG_EMBREF) { refcnt = n->m_ext.ext_cnt = &m->m_ext.ext_count; n->m_ext.ext_flags &= ~EXT_FLAG_EMBREF; } else { KASSERT(m->m_ext.ext_cnt != NULL, ("%s: no refcounting pointer on %p", __func__, m)); refcnt = m->m_ext.ext_cnt; } if (*refcnt == 1) *refcnt += 1; else atomic_add_int(refcnt, 1); } void m_demote_pkthdr(struct mbuf *m) { M_ASSERTPKTHDR(m); M_ASSERT_NO_SND_TAG(m); m_tag_delete_chain(m, NULL); m->m_flags &= ~M_PKTHDR; bzero(&m->m_pkthdr, sizeof(struct pkthdr)); } /* * Clean up mbuf (chain) from any tags and packet headers. * If "all" is set then the first mbuf in the chain will be * cleaned too. */ void m_demote(struct mbuf *m0, int all, int flags) { struct mbuf *m; flags |= M_DEMOTEFLAGS; for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) { KASSERT(m->m_nextpkt == NULL, ("%s: m_nextpkt in m %p, m0 %p", __func__, m, m0)); if (m->m_flags & M_PKTHDR) m_demote_pkthdr(m); m->m_flags &= flags; } } /* * Sanity checks on mbuf (chain) for use in KASSERT() and general * debugging. * Returns 0 or panics when bad and 1 on all tests passed. * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they * blow up later. */ int m_sanity(struct mbuf *m0, int sanitize) { struct mbuf *m; caddr_t a, b; int pktlen = 0; #ifdef INVARIANTS #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) #else #define M_SANITY_ACTION(s) printf("mbuf %p: " s, m) #endif for (m = m0; m != NULL; m = m->m_next) { /* * Basic pointer checks. If any of these fails then some * unrelated kernel memory before or after us is trashed. * No way to recover from that. */ a = M_START(m); b = a + M_SIZE(m); if ((caddr_t)m->m_data < a) M_SANITY_ACTION("m_data outside mbuf data range left"); if ((caddr_t)m->m_data > b) M_SANITY_ACTION("m_data outside mbuf data range right"); if ((caddr_t)m->m_data + m->m_len > b) M_SANITY_ACTION("m_data + m_len exeeds mbuf space"); /* m->m_nextpkt may only be set on first mbuf in chain. */ if (m != m0 && m->m_nextpkt != NULL) { if (sanitize) { m_freem(m->m_nextpkt); m->m_nextpkt = (struct mbuf *)0xDEADC0DE; } else M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf"); } /* packet length (not mbuf length!) calculation */ if (m0->m_flags & M_PKTHDR) pktlen += m->m_len; /* m_tags may only be attached to first mbuf in chain. */ if (m != m0 && m->m_flags & M_PKTHDR && !SLIST_EMPTY(&m->m_pkthdr.tags)) { if (sanitize) { m_tag_delete_chain(m, NULL); /* put in 0xDEADC0DE perhaps? */ } else M_SANITY_ACTION("m_tags on in-chain mbuf"); } /* M_PKTHDR may only be set on first mbuf in chain */ if (m != m0 && m->m_flags & M_PKTHDR) { if (sanitize) { bzero(&m->m_pkthdr, sizeof(m->m_pkthdr)); m->m_flags &= ~M_PKTHDR; /* put in 0xDEADCODE and leave hdr flag in */ } else M_SANITY_ACTION("M_PKTHDR on in-chain mbuf"); } } m = m0; if (pktlen && pktlen != m->m_pkthdr.len) { if (sanitize) m->m_pkthdr.len = 0; else M_SANITY_ACTION("m_pkthdr.len != mbuf chain length"); } return 1; #undef M_SANITY_ACTION } /* * Non-inlined part of m_init(). */ int m_pkthdr_init(struct mbuf *m, int how) { #ifdef MAC int error; #endif m->m_data = m->m_pktdat; bzero(&m->m_pkthdr, sizeof(m->m_pkthdr)); #ifdef NUMA m->m_pkthdr.numa_domain = M_NODOM; #endif #ifdef MAC /* If the label init fails, fail the alloc */ error = mac_mbuf_init(m, how); if (error) return (error); #endif return (0); } /* * "Move" mbuf pkthdr from "from" to "to". * "from" must have M_PKTHDR set, and "to" must be empty. */ void m_move_pkthdr(struct mbuf *to, struct mbuf *from) { #if 0 /* see below for why these are not enabled */ M_ASSERTPKTHDR(to); /* Note: with MAC, this may not be a good assertion. */ KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_move_pkthdr: to has tags")); #endif #ifdef MAC /* * XXXMAC: It could be this should also occur for non-MAC? */ if (to->m_flags & M_PKTHDR) m_tag_delete_chain(to, NULL); #endif to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & (M_EXT | M_EXTPG)); if ((to->m_flags & M_EXT) == 0) to->m_data = to->m_pktdat; to->m_pkthdr = from->m_pkthdr; /* especially tags */ SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ from->m_flags &= ~M_PKTHDR; if (from->m_pkthdr.csum_flags & CSUM_SND_TAG) { from->m_pkthdr.csum_flags &= ~CSUM_SND_TAG; from->m_pkthdr.snd_tag = NULL; } } /* * Duplicate "from"'s mbuf pkthdr in "to". * "from" must have M_PKTHDR set, and "to" must be empty. * In particular, this does a deep copy of the packet tags. */ int m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how) { #if 0 /* * The mbuf allocator only initializes the pkthdr * when the mbuf is allocated with m_gethdr(). Many users * (e.g. m_copy*, m_prepend) use m_get() and then * smash the pkthdr as needed causing these * assertions to trip. For now just disable them. */ M_ASSERTPKTHDR(to); /* Note: with MAC, this may not be a good assertion. */ KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags")); #endif MBUF_CHECKSLEEP(how); #ifdef MAC if (to->m_flags & M_PKTHDR) m_tag_delete_chain(to, NULL); #endif to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & (M_EXT | M_EXTPG)); if ((to->m_flags & M_EXT) == 0) to->m_data = to->m_pktdat; to->m_pkthdr = from->m_pkthdr; if (from->m_pkthdr.csum_flags & CSUM_SND_TAG) m_snd_tag_ref(from->m_pkthdr.snd_tag); SLIST_INIT(&to->m_pkthdr.tags); return (m_tag_copy_chain(to, from, how)); } /* * Lesser-used path for M_PREPEND: * allocate new mbuf to prepend to chain, * copy junk along. */ struct mbuf * m_prepend(struct mbuf *m, int len, int how) { struct mbuf *mn; if (m->m_flags & M_PKTHDR) mn = m_gethdr(how, m->m_type); else mn = m_get(how, m->m_type); if (mn == NULL) { m_freem(m); return (NULL); } if (m->m_flags & M_PKTHDR) m_move_pkthdr(mn, m); mn->m_next = m; m = mn; if (len < M_SIZE(m)) M_ALIGN(m, len); m->m_len = len; return (m); } /* * Make a copy of an mbuf chain starting "off0" bytes from the beginning, * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller. * Note that the copy is read-only, because clusters are not copied, * only their reference counts are incremented. */ struct mbuf * m_copym(struct mbuf *m, int off0, int len, int wait) { struct mbuf *n, **np; int off = off0; struct mbuf *top; int copyhdr = 0; KASSERT(off >= 0, ("m_copym, negative off %d", off)); KASSERT(len >= 0, ("m_copym, negative len %d", len)); MBUF_CHECKSLEEP(wait); if (off == 0 && m->m_flags & M_PKTHDR) copyhdr = 1; while (off > 0) { KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); if (off < m->m_len) break; off -= m->m_len; m = m->m_next; } np = ⊤ top = NULL; while (len > 0) { if (m == NULL) { KASSERT(len == M_COPYALL, ("m_copym, length > size of mbuf chain")); break; } if (copyhdr) n = m_gethdr(wait, m->m_type); else n = m_get(wait, m->m_type); *np = n; if (n == NULL) goto nospace; if (copyhdr) { if (!m_dup_pkthdr(n, m, wait)) goto nospace; if (len == M_COPYALL) n->m_pkthdr.len -= off0; else n->m_pkthdr.len = len; copyhdr = 0; } n->m_len = min(len, m->m_len - off); if (m->m_flags & (M_EXT|M_EXTPG)) { n->m_data = m->m_data + off; mb_dupcl(n, m); } else bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), (u_int)n->m_len); if (len != M_COPYALL) len -= n->m_len; off = 0; m = m->m_next; np = &n->m_next; } return (top); nospace: m_freem(top); return (NULL); } /* * Copy an entire packet, including header (which must be present). * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. * Note that the copy is read-only, because clusters are not copied, * only their reference counts are incremented. * Preserve alignment of the first mbuf so if the creator has left * some room at the beginning (e.g. for inserting protocol headers) * the copies still have the room available. */ struct mbuf * m_copypacket(struct mbuf *m, int how) { struct mbuf *top, *n, *o; MBUF_CHECKSLEEP(how); n = m_get(how, m->m_type); top = n; if (n == NULL) goto nospace; if (!m_dup_pkthdr(n, m, how)) goto nospace; n->m_len = m->m_len; if (m->m_flags & (M_EXT|M_EXTPG)) { n->m_data = m->m_data; mb_dupcl(n, m); } else { n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); bcopy(mtod(m, char *), mtod(n, char *), n->m_len); } m = m->m_next; while (m) { o = m_get(how, m->m_type); if (o == NULL) goto nospace; n->m_next = o; n = n->m_next; n->m_len = m->m_len; if (m->m_flags & (M_EXT|M_EXTPG)) { n->m_data = m->m_data; mb_dupcl(n, m); } else { bcopy(mtod(m, char *), mtod(n, char *), n->m_len); } m = m->m_next; } return top; nospace: m_freem(top); return (NULL); } static void m_copyfromunmapped(const struct mbuf *m, int off, int len, caddr_t cp) { struct iovec iov; struct uio uio; int error __diagused; KASSERT(off >= 0, ("m_copyfromunmapped: negative off %d", off)); KASSERT(len >= 0, ("m_copyfromunmapped: negative len %d", len)); KASSERT(off < m->m_len, ("m_copyfromunmapped: len exceeds mbuf length")); iov.iov_base = cp; iov.iov_len = len; uio.uio_resid = len; uio.uio_iov = &iov; uio.uio_segflg = UIO_SYSSPACE; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_rw = UIO_READ; error = m_unmapped_uiomove(m, off, &uio, len); KASSERT(error == 0, ("m_unmapped_uiomove failed: off %d, len %d", off, len)); } /* * Copy data from an mbuf chain starting "off" bytes from the beginning, * continuing for "len" bytes, into the indicated buffer. */ void m_copydata(const struct mbuf *m, int off, int len, caddr_t cp) { u_int count; KASSERT(off >= 0, ("m_copydata, negative off %d", off)); KASSERT(len >= 0, ("m_copydata, negative len %d", len)); while (off > 0) { KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); if (off < m->m_len) break; off -= m->m_len; m = m->m_next; } while (len > 0) { KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); count = min(m->m_len - off, len); if ((m->m_flags & M_EXTPG) != 0) m_copyfromunmapped(m, off, count, cp); else bcopy(mtod(m, caddr_t) + off, cp, count); len -= count; cp += count; off = 0; m = m->m_next; } } /* * Copy a packet header mbuf chain into a completely new chain, including * copying any mbuf clusters. Use this instead of m_copypacket() when * you need a writable copy of an mbuf chain. */ struct mbuf * m_dup(const struct mbuf *m, int how) { struct mbuf **p, *top = NULL; int remain, moff, nsize; MBUF_CHECKSLEEP(how); /* Sanity check */ if (m == NULL) return (NULL); M_ASSERTPKTHDR(m); /* While there's more data, get a new mbuf, tack it on, and fill it */ remain = m->m_pkthdr.len; moff = 0; p = ⊤ while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ struct mbuf *n; /* Get the next new mbuf */ if (remain >= MINCLSIZE) { n = m_getcl(how, m->m_type, 0); nsize = MCLBYTES; } else { n = m_get(how, m->m_type); nsize = MLEN; } if (n == NULL) goto nospace; if (top == NULL) { /* First one, must be PKTHDR */ if (!m_dup_pkthdr(n, m, how)) { m_free(n); goto nospace; } if ((n->m_flags & M_EXT) == 0) nsize = MHLEN; n->m_flags &= ~M_RDONLY; } n->m_len = 0; /* Link it into the new chain */ *p = n; p = &n->m_next; /* Copy data from original mbuf(s) into new mbuf */ while (n->m_len < nsize && m != NULL) { int chunk = min(nsize - n->m_len, m->m_len - moff); m_copydata(m, moff, chunk, n->m_data + n->m_len); moff += chunk; n->m_len += chunk; remain -= chunk; if (moff == m->m_len) { m = m->m_next; moff = 0; } } /* Check correct total mbuf length */ KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), ("%s: bogus m_pkthdr.len", __func__)); } return (top); nospace: m_freem(top); return (NULL); } /* * Concatenate mbuf chain n to m. * Both chains must be of the same type (e.g. MT_DATA). * Any m_pkthdr is not updated. */ void m_cat(struct mbuf *m, struct mbuf *n) { while (m->m_next) m = m->m_next; while (n) { if (!M_WRITABLE(m) || (n->m_flags & M_EXTPG) != 0 || M_TRAILINGSPACE(m) < n->m_len) { /* just join the two chains */ m->m_next = n; return; } /* splat the data from one into the other */ bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, (u_int)n->m_len); m->m_len += n->m_len; n = m_free(n); } } /* * Concatenate two pkthdr mbuf chains. */ void m_catpkt(struct mbuf *m, struct mbuf *n) { M_ASSERTPKTHDR(m); M_ASSERTPKTHDR(n); m->m_pkthdr.len += n->m_pkthdr.len; m_demote(n, 1, 0); m_cat(m, n); } void m_adj(struct mbuf *mp, int req_len) { int len = req_len; struct mbuf *m; int count; if ((m = mp) == NULL) return; if (len >= 0) { /* * Trim from head. */ while (m != NULL && len > 0) { if (m->m_len <= len) { len -= m->m_len; m->m_len = 0; m = m->m_next; } else { m->m_len -= len; m->m_data += len; len = 0; } } if (mp->m_flags & M_PKTHDR) mp->m_pkthdr.len -= (req_len - len); } else { /* * Trim from tail. Scan the mbuf chain, * calculating its length and finding the last mbuf. * If the adjustment only affects this mbuf, then just * adjust and return. Otherwise, rescan and truncate * after the remaining size. */ len = -len; count = 0; for (;;) { count += m->m_len; if (m->m_next == (struct mbuf *)0) break; m = m->m_next; } if (m->m_len >= len) { m->m_len -= len; if (mp->m_flags & M_PKTHDR) mp->m_pkthdr.len -= len; return; } count -= len; if (count < 0) count = 0; /* * Correct length for chain is "count". * Find the mbuf with last data, adjust its length, * and toss data from remaining mbufs on chain. */ m = mp; if (m->m_flags & M_PKTHDR) m->m_pkthdr.len = count; for (; m; m = m->m_next) { if (m->m_len >= count) { m->m_len = count; if (m->m_next != NULL) { m_freem(m->m_next); m->m_next = NULL; } break; } count -= m->m_len; } } } void m_adj_decap(struct mbuf *mp, int len) { uint8_t rsstype; m_adj(mp, len); if ((mp->m_flags & M_PKTHDR) != 0) { /* * If flowid was calculated by card from the inner * headers, move flowid to the decapsulated mbuf * chain, otherwise clear. This depends on the * internals of m_adj, which keeps pkthdr as is, in * particular not changing rsstype and flowid. */ rsstype = mp->m_pkthdr.rsstype; if ((rsstype & M_HASHTYPE_INNER) != 0) { M_HASHTYPE_SET(mp, rsstype & ~M_HASHTYPE_INNER); } else { M_HASHTYPE_CLEAR(mp); } } } /* * Rearange an mbuf chain so that len bytes are contiguous * and in the data area of an mbuf (so that mtod will work * for a structure of size len). Returns the resulting * mbuf chain on success, frees it and returns null on failure. * If there is room, it will add up to max_protohdr-len extra bytes to the * contiguous region in an attempt to avoid being called next time. */ struct mbuf * m_pullup(struct mbuf *n, int len) { struct mbuf *m; int count; int space; KASSERT((n->m_flags & M_EXTPG) == 0, ("%s: unmapped mbuf %p", __func__, n)); /* * If first mbuf has no cluster, and has room for len bytes * without shifting current data, pullup into it, * otherwise allocate a new mbuf to prepend to the chain. */ if ((n->m_flags & M_EXT) == 0 && n->m_data + len < &n->m_dat[MLEN] && n->m_next) { if (n->m_len >= len) return (n); m = n; n = n->m_next; len -= m->m_len; } else { if (len > MHLEN) goto bad; m = m_get(M_NOWAIT, n->m_type); if (m == NULL) goto bad; if (n->m_flags & M_PKTHDR) m_move_pkthdr(m, n); } space = &m->m_dat[MLEN] - (m->m_data + m->m_len); do { count = min(min(max(len, max_protohdr), space), n->m_len); bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, (u_int)count); len -= count; m->m_len += count; n->m_len -= count; space -= count; if (n->m_len) n->m_data += count; else n = m_free(n); } while (len > 0 && n); if (len > 0) { (void) m_free(m); goto bad; } m->m_next = n; return (m); bad: m_freem(n); return (NULL); } /* * Like m_pullup(), except a new mbuf is always allocated, and we allow * the amount of empty space before the data in the new mbuf to be specified * (in the event that the caller expects to prepend later). */ struct mbuf * m_copyup(struct mbuf *n, int len, int dstoff) { struct mbuf *m; int count, space; if (len > (MHLEN - dstoff)) goto bad; m = m_get(M_NOWAIT, n->m_type); if (m == NULL) goto bad; if (n->m_flags & M_PKTHDR) m_move_pkthdr(m, n); m->m_data += dstoff; space = &m->m_dat[MLEN] - (m->m_data + m->m_len); do { count = min(min(max(len, max_protohdr), space), n->m_len); memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t), (unsigned)count); len -= count; m->m_len += count; n->m_len -= count; space -= count; if (n->m_len) n->m_data += count; else n = m_free(n); } while (len > 0 && n); if (len > 0) { (void) m_free(m); goto bad; } m->m_next = n; return (m); bad: m_freem(n); return (NULL); } /* * Partition an mbuf chain in two pieces, returning the tail -- * all but the first len0 bytes. In case of failure, it returns NULL and * attempts to restore the chain to its original state. * * Note that the resulting mbufs might be read-only, because the new * mbuf can end up sharing an mbuf cluster with the original mbuf if * the "breaking point" happens to lie within a cluster mbuf. Use the * M_WRITABLE() macro to check for this case. */ struct mbuf * m_split(struct mbuf *m0, int len0, int wait) { struct mbuf *m, *n; u_int len = len0, remain; MBUF_CHECKSLEEP(wait); for (m = m0; m && len > m->m_len; m = m->m_next) len -= m->m_len; if (m == NULL) return (NULL); remain = m->m_len - len; if (m0->m_flags & M_PKTHDR && remain == 0) { n = m_gethdr(wait, m0->m_type); if (n == NULL) return (NULL); n->m_next = m->m_next; m->m_next = NULL; if (m0->m_pkthdr.csum_flags & CSUM_SND_TAG) { n->m_pkthdr.snd_tag = m_snd_tag_ref(m0->m_pkthdr.snd_tag); n->m_pkthdr.csum_flags |= CSUM_SND_TAG; } else n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; n->m_pkthdr.len = m0->m_pkthdr.len - len0; m0->m_pkthdr.len = len0; return (n); } else if (m0->m_flags & M_PKTHDR) { n = m_gethdr(wait, m0->m_type); if (n == NULL) return (NULL); if (m0->m_pkthdr.csum_flags & CSUM_SND_TAG) { n->m_pkthdr.snd_tag = m_snd_tag_ref(m0->m_pkthdr.snd_tag); n->m_pkthdr.csum_flags |= CSUM_SND_TAG; } else n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; n->m_pkthdr.len = m0->m_pkthdr.len - len0; m0->m_pkthdr.len = len0; if (m->m_flags & (M_EXT|M_EXTPG)) goto extpacket; if (remain > MHLEN) { /* m can't be the lead packet */ M_ALIGN(n, 0); n->m_next = m_split(m, len, wait); if (n->m_next == NULL) { (void) m_free(n); return (NULL); } else { n->m_len = 0; return (n); } } else M_ALIGN(n, remain); } else if (remain == 0) { n = m->m_next; m->m_next = NULL; return (n); } else { n = m_get(wait, m->m_type); if (n == NULL) return (NULL); M_ALIGN(n, remain); } extpacket: if (m->m_flags & (M_EXT|M_EXTPG)) { n->m_data = m->m_data + len; mb_dupcl(n, m); } else { bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); } n->m_len = remain; m->m_len = len; n->m_next = m->m_next; m->m_next = NULL; return (n); } /* * Routine to copy from device local memory into mbufs. * Note that `off' argument is offset into first mbuf of target chain from * which to begin copying the data to. */ struct mbuf * m_devget(char *buf, int totlen, int off, struct ifnet *ifp, void (*copy)(char *from, caddr_t to, u_int len)) { struct mbuf *m; struct mbuf *top = NULL, **mp = ⊤ int len; if (off < 0 || off > MHLEN) return (NULL); while (totlen > 0) { if (top == NULL) { /* First one, must be PKTHDR */ if (totlen + off >= MINCLSIZE) { m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); len = MCLBYTES; } else { m = m_gethdr(M_NOWAIT, MT_DATA); len = MHLEN; /* Place initial small packet/header at end of mbuf */ if (m && totlen + off + max_linkhdr <= MHLEN) { m->m_data += max_linkhdr; len -= max_linkhdr; } } if (m == NULL) return NULL; m->m_pkthdr.rcvif = ifp; m->m_pkthdr.len = totlen; } else { if (totlen + off >= MINCLSIZE) { m = m_getcl(M_NOWAIT, MT_DATA, 0); len = MCLBYTES; } else { m = m_get(M_NOWAIT, MT_DATA); len = MLEN; } if (m == NULL) { m_freem(top); return NULL; } } if (off) { m->m_data += off; len -= off; off = 0; } m->m_len = len = min(totlen, len); if (copy) copy(buf, mtod(m, caddr_t), (u_int)len); else bcopy(buf, mtod(m, caddr_t), (u_int)len); buf += len; *mp = m; mp = &m->m_next; totlen -= len; } return (top); } static void m_copytounmapped(const struct mbuf *m, int off, int len, c_caddr_t cp) { struct iovec iov; struct uio uio; int error __diagused; KASSERT(off >= 0, ("m_copytounmapped: negative off %d", off)); KASSERT(len >= 0, ("m_copytounmapped: negative len %d", len)); KASSERT(off < m->m_len, ("m_copytounmapped: len exceeds mbuf length")); iov.iov_base = __DECONST(caddr_t, cp); iov.iov_len = len; uio.uio_resid = len; uio.uio_iov = &iov; uio.uio_segflg = UIO_SYSSPACE; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_rw = UIO_WRITE; error = m_unmapped_uiomove(m, off, &uio, len); KASSERT(error == 0, ("m_unmapped_uiomove failed: off %d, len %d", off, len)); } /* * Copy data from a buffer back into the indicated mbuf chain, * starting "off" bytes from the beginning, extending the mbuf * chain if necessary. */ void m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp) { int mlen; struct mbuf *m = m0, *n; int totlen = 0; if (m0 == NULL) return; while (off > (mlen = m->m_len)) { off -= mlen; totlen += mlen; if (m->m_next == NULL) { n = m_get(M_NOWAIT, m->m_type); if (n == NULL) goto out; bzero(mtod(n, caddr_t), MLEN); n->m_len = min(MLEN, len + off); m->m_next = n; } m = m->m_next; } while (len > 0) { if (m->m_next == NULL && (len > m->m_len - off)) { m->m_len += min(len - (m->m_len - off), M_TRAILINGSPACE(m)); } mlen = min (m->m_len - off, len); if ((m->m_flags & M_EXTPG) != 0) m_copytounmapped(m, off, mlen, cp); else bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen); cp += mlen; len -= mlen; mlen += off; off = 0; totlen += mlen; if (len == 0) break; if (m->m_next == NULL) { n = m_get(M_NOWAIT, m->m_type); if (n == NULL) break; n->m_len = min(MLEN, len); m->m_next = n; } m = m->m_next; } out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) m->m_pkthdr.len = totlen; } /* * Append the specified data to the indicated mbuf chain, * Extend the mbuf chain if the new data does not fit in * existing space. * * Return 1 if able to complete the job; otherwise 0. */ int m_append(struct mbuf *m0, int len, c_caddr_t cp) { struct mbuf *m, *n; int remainder, space; for (m = m0; m->m_next != NULL; m = m->m_next) ; remainder = len; space = M_TRAILINGSPACE(m); if (space > 0) { /* * Copy into available space. */ if (space > remainder) space = remainder; bcopy(cp, mtod(m, caddr_t) + m->m_len, space); m->m_len += space; cp += space, remainder -= space; } while (remainder > 0) { /* * Allocate a new mbuf; could check space * and allocate a cluster instead. */ n = m_get(M_NOWAIT, m->m_type); if (n == NULL) break; n->m_len = min(MLEN, remainder); bcopy(cp, mtod(n, caddr_t), n->m_len); cp += n->m_len, remainder -= n->m_len; m->m_next = n; m = n; } if (m0->m_flags & M_PKTHDR) m0->m_pkthdr.len += len - remainder; return (remainder == 0); } static int m_apply_extpg_one(struct mbuf *m, int off, int len, int (*f)(void *, void *, u_int), void *arg) { void *p; u_int i, count, pgoff, pglen; int rval; KASSERT(PMAP_HAS_DMAP, ("m_apply_extpg_one does not support unmapped mbufs")); off += mtod(m, vm_offset_t); if (off < m->m_epg_hdrlen) { count = min(m->m_epg_hdrlen - off, len); rval = f(arg, m->m_epg_hdr + off, count); if (rval) return (rval); len -= count; off = 0; } else off -= m->m_epg_hdrlen; pgoff = m->m_epg_1st_off; for (i = 0; i < m->m_epg_npgs && len > 0; i++) { pglen = m_epg_pagelen(m, i, pgoff); if (off < pglen) { count = min(pglen - off, len); p = (void *)PHYS_TO_DMAP(m->m_epg_pa[i] + pgoff + off); rval = f(arg, p, count); if (rval) return (rval); len -= count; off = 0; } else off -= pglen; pgoff = 0; } if (len > 0) { KASSERT(off < m->m_epg_trllen, ("m_apply_extpg_one: offset beyond trailer")); KASSERT(len <= m->m_epg_trllen - off, ("m_apply_extpg_one: length beyond trailer")); return (f(arg, m->m_epg_trail + off, len)); } return (0); } /* Apply function f to the data in a single mbuf. */ static int m_apply_one(struct mbuf *m, int off, int len, int (*f)(void *, void *, u_int), void *arg) { if ((m->m_flags & M_EXTPG) != 0) return (m_apply_extpg_one(m, off, len, f, arg)); else return (f(arg, mtod(m, caddr_t) + off, len)); } /* * Apply function f to the data in an mbuf chain starting "off" bytes from * the beginning, continuing for "len" bytes. */ int m_apply(struct mbuf *m, int off, int len, int (*f)(void *, void *, u_int), void *arg) { u_int count; int rval; KASSERT(off >= 0, ("m_apply, negative off %d", off)); KASSERT(len >= 0, ("m_apply, negative len %d", len)); while (off > 0) { KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); if (off < m->m_len) break; off -= m->m_len; m = m->m_next; } while (len > 0) { KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); count = min(m->m_len - off, len); rval = m_apply_one(m, off, count, f, arg); if (rval) return (rval); len -= count; off = 0; m = m->m_next; } return (0); } /* * Return a pointer to mbuf/offset of location in mbuf chain. */ struct mbuf * m_getptr(struct mbuf *m, int loc, int *off) { while (loc >= 0) { /* Normal end of search. */ if (m->m_len > loc) { *off = loc; return (m); } else { loc -= m->m_len; if (m->m_next == NULL) { if (loc == 0) { /* Point at the end of valid data. */ *off = m->m_len; return (m); } return (NULL); } m = m->m_next; } } return (NULL); } void m_print(const struct mbuf *m, int maxlen) { int len; int pdata; const struct mbuf *m2; if (m == NULL) { printf("mbuf: %p\n", m); return; } if (m->m_flags & M_PKTHDR) len = m->m_pkthdr.len; else len = -1; m2 = m; while (m2 != NULL && (len == -1 || len)) { pdata = m2->m_len; if (maxlen != -1 && pdata > maxlen) pdata = maxlen; printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len, m2->m_next, m2->m_flags, "\20\20freelist\17skipfw" "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly" "\3eor\2pkthdr\1ext", pdata ? "" : "\n"); if (pdata) printf(", %*D\n", pdata, (u_char *)m2->m_data, "-"); if (len != -1) len -= m2->m_len; m2 = m2->m_next; } if (len > 0) printf("%d bytes unaccounted for.\n", len); return; } u_int m_fixhdr(struct mbuf *m0) { u_int len; len = m_length(m0, NULL); m0->m_pkthdr.len = len; return (len); } u_int m_length(struct mbuf *m0, struct mbuf **last) { struct mbuf *m; u_int len; len = 0; for (m = m0; m != NULL; m = m->m_next) { len += m->m_len; if (m->m_next == NULL) break; } if (last != NULL) *last = m; return (len); } /* * Defragment a mbuf chain, returning the shortest possible * chain of mbufs and clusters. If allocation fails and * this cannot be completed, NULL will be returned, but * the passed in chain will be unchanged. Upon success, * the original chain will be freed, and the new chain * will be returned. * * If a non-packet header is passed in, the original * mbuf (chain?) will be returned unharmed. */ struct mbuf * m_defrag(struct mbuf *m0, int how) { struct mbuf *m_new = NULL, *m_final = NULL; int progress = 0, length; MBUF_CHECKSLEEP(how); if (!(m0->m_flags & M_PKTHDR)) return (m0); m_fixhdr(m0); /* Needed sanity check */ #ifdef MBUF_STRESS_TEST if (m_defragrandomfailures) { int temp = arc4random() & 0xff; if (temp == 0xba) goto nospace; } #endif if (m0->m_pkthdr.len > MHLEN) m_final = m_getcl(how, MT_DATA, M_PKTHDR); else m_final = m_gethdr(how, MT_DATA); if (m_final == NULL) goto nospace; if (m_dup_pkthdr(m_final, m0, how) == 0) goto nospace; m_new = m_final; while (progress < m0->m_pkthdr.len) { length = m0->m_pkthdr.len - progress; if (length > MCLBYTES) length = MCLBYTES; if (m_new == NULL) { if (length > MLEN) m_new = m_getcl(how, MT_DATA, 0); else m_new = m_get(how, MT_DATA); if (m_new == NULL) goto nospace; } m_copydata(m0, progress, length, mtod(m_new, caddr_t)); progress += length; m_new->m_len = length; if (m_new != m_final) m_cat(m_final, m_new); m_new = NULL; } #ifdef MBUF_STRESS_TEST if (m0->m_next == NULL) m_defraguseless++; #endif m_freem(m0); m0 = m_final; #ifdef MBUF_STRESS_TEST m_defragpackets++; m_defragbytes += m0->m_pkthdr.len; #endif return (m0); nospace: #ifdef MBUF_STRESS_TEST m_defragfailure++; #endif if (m_final) m_freem(m_final); return (NULL); } /* * Return the number of fragments an mbuf will use. This is usually * used as a proxy for the number of scatter/gather elements needed by * a DMA engine to access an mbuf. In general mapped mbufs are * assumed to be backed by physically contiguous buffers that only * need a single fragment. Unmapped mbufs, on the other hand, can * span disjoint physical pages. */ static int frags_per_mbuf(struct mbuf *m) { int frags; if ((m->m_flags & M_EXTPG) == 0) return (1); /* * The header and trailer are counted as a single fragment * each when present. * * XXX: This overestimates the number of fragments by assuming * all the backing physical pages are disjoint. */ frags = 0; if (m->m_epg_hdrlen != 0) frags++; frags += m->m_epg_npgs; if (m->m_epg_trllen != 0) frags++; return (frags); } /* * Defragment an mbuf chain, returning at most maxfrags separate * mbufs+clusters. If this is not possible NULL is returned and * the original mbuf chain is left in its present (potentially * modified) state. We use two techniques: collapsing consecutive * mbufs and replacing consecutive mbufs by a cluster. * * NB: this should really be named m_defrag but that name is taken */ struct mbuf * m_collapse(struct mbuf *m0, int how, int maxfrags) { struct mbuf *m, *n, *n2, **prev; u_int curfrags; /* * Calculate the current number of frags. */ curfrags = 0; for (m = m0; m != NULL; m = m->m_next) curfrags += frags_per_mbuf(m); /* * First, try to collapse mbufs. Note that we always collapse * towards the front so we don't need to deal with moving the * pkthdr. This may be suboptimal if the first mbuf has much * less data than the following. */ m = m0; again: for (;;) { n = m->m_next; if (n == NULL) break; if (M_WRITABLE(m) && n->m_len < M_TRAILINGSPACE(m)) { m_copydata(n, 0, n->m_len, mtod(m, char *) + m->m_len); m->m_len += n->m_len; m->m_next = n->m_next; curfrags -= frags_per_mbuf(n); m_free(n); if (curfrags <= maxfrags) return m0; } else m = n; } KASSERT(maxfrags > 1, ("maxfrags %u, but normal collapse failed", maxfrags)); /* * Collapse consecutive mbufs to a cluster. */ prev = &m0->m_next; /* NB: not the first mbuf */ while ((n = *prev) != NULL) { if ((n2 = n->m_next) != NULL && n->m_len + n2->m_len < MCLBYTES) { m = m_getcl(how, MT_DATA, 0); if (m == NULL) goto bad; m_copydata(n, 0, n->m_len, mtod(m, char *)); m_copydata(n2, 0, n2->m_len, mtod(m, char *) + n->m_len); m->m_len = n->m_len + n2->m_len; m->m_next = n2->m_next; *prev = m; curfrags += 1; /* For the new cluster */ curfrags -= frags_per_mbuf(n); curfrags -= frags_per_mbuf(n2); m_free(n); m_free(n2); if (curfrags <= maxfrags) return m0; /* * Still not there, try the normal collapse * again before we allocate another cluster. */ goto again; } prev = &n->m_next; } /* * No place where we can collapse to a cluster; punt. * This can occur if, for example, you request 2 frags * but the packet requires that both be clusters (we * never reallocate the first mbuf to avoid moving the * packet header). */ bad: return NULL; } #ifdef MBUF_STRESS_TEST /* * Fragment an mbuf chain. There's no reason you'd ever want to do * this in normal usage, but it's great for stress testing various * mbuf consumers. * * If fragmentation is not possible, the original chain will be * returned. * * Possible length values: * 0 no fragmentation will occur * > 0 each fragment will be of the specified length * -1 each fragment will be the same random value in length * -2 each fragment's length will be entirely random * (Random values range from 1 to 256) */ struct mbuf * m_fragment(struct mbuf *m0, int how, int length) { struct mbuf *m_first, *m_last; int divisor = 255, progress = 0, fraglen; if (!(m0->m_flags & M_PKTHDR)) return (m0); if (length == 0 || length < -2) return (m0); if (length > MCLBYTES) length = MCLBYTES; if (length < 0 && divisor > MCLBYTES) divisor = MCLBYTES; if (length == -1) length = 1 + (arc4random() % divisor); if (length > 0) fraglen = length; m_fixhdr(m0); /* Needed sanity check */ m_first = m_getcl(how, MT_DATA, M_PKTHDR); if (m_first == NULL) goto nospace; if (m_dup_pkthdr(m_first, m0, how) == 0) goto nospace; m_last = m_first; while (progress < m0->m_pkthdr.len) { if (length == -2) fraglen = 1 + (arc4random() % divisor); if (fraglen > m0->m_pkthdr.len - progress) fraglen = m0->m_pkthdr.len - progress; if (progress != 0) { struct mbuf *m_new = m_getcl(how, MT_DATA, 0); if (m_new == NULL) goto nospace; m_last->m_next = m_new; m_last = m_new; } m_copydata(m0, progress, fraglen, mtod(m_last, caddr_t)); progress += fraglen; m_last->m_len = fraglen; } m_freem(m0); m0 = m_first; return (m0); nospace: if (m_first) m_freem(m_first); /* Return the original chain on failure */ return (m0); } #endif /* * Free pages from mbuf_ext_pgs, assuming they were allocated via * vm_page_alloc() and aren't associated with any object. Complement * to allocator from m_uiotombuf_nomap(). */ void mb_free_mext_pgs(struct mbuf *m) { vm_page_t pg; M_ASSERTEXTPG(m); for (int i = 0; i < m->m_epg_npgs; i++) { pg = PHYS_TO_VM_PAGE(m->m_epg_pa[i]); vm_page_unwire_noq(pg); vm_page_free(pg); } } static struct mbuf * m_uiotombuf_nomap(struct uio *uio, int how, int len, int maxseg, int flags) { struct mbuf *m, *mb, *prev; vm_page_t pg_array[MBUF_PEXT_MAX_PGS]; int error, length, i, needed; ssize_t total; int pflags = malloc2vm_flags(how) | VM_ALLOC_NODUMP | VM_ALLOC_WIRED; MPASS((flags & M_PKTHDR) == 0); MPASS((how & M_ZERO) == 0); /* * len can be zero or an arbitrary large value bound by * the total data supplied by the uio. */ if (len > 0) total = MIN(uio->uio_resid, len); else total = uio->uio_resid; if (maxseg == 0) maxseg = MBUF_PEXT_MAX_PGS * PAGE_SIZE; /* * If total is zero, return an empty mbuf. This can occur * for TLS 1.0 connections which send empty fragments as * a countermeasure against the known-IV weakness in CBC * ciphersuites. */ if (__predict_false(total == 0)) { mb = mb_alloc_ext_pgs(how, mb_free_mext_pgs); if (mb == NULL) return (NULL); mb->m_epg_flags = EPG_FLAG_ANON; return (mb); } /* * Allocate the pages */ m = NULL; while (total > 0) { mb = mb_alloc_ext_pgs(how, mb_free_mext_pgs); if (mb == NULL) goto failed; if (m == NULL) m = mb; else prev->m_next = mb; prev = mb; mb->m_epg_flags = EPG_FLAG_ANON; needed = length = MIN(maxseg, total); for (i = 0; needed > 0; i++, needed -= PAGE_SIZE) { retry_page: pg_array[i] = vm_page_alloc_noobj(pflags); if (pg_array[i] == NULL) { if (how & M_NOWAIT) { goto failed; } else { vm_wait(NULL); goto retry_page; } } mb->m_epg_pa[i] = VM_PAGE_TO_PHYS(pg_array[i]); mb->m_epg_npgs++; } mb->m_epg_last_len = length - PAGE_SIZE * (mb->m_epg_npgs - 1); MBUF_EXT_PGS_ASSERT_SANITY(mb); total -= length; error = uiomove_fromphys(pg_array, 0, length, uio); if (error != 0) goto failed; mb->m_len = length; mb->m_ext.ext_size += PAGE_SIZE * mb->m_epg_npgs; if (flags & M_PKTHDR) m->m_pkthdr.len += length; } return (m); failed: m_freem(m); return (NULL); } /* * Copy the contents of uio into a properly sized mbuf chain. */ struct mbuf * m_uiotombuf(struct uio *uio, int how, int len, int align, int flags) { struct mbuf *m, *mb; int error, length; ssize_t total; int progress = 0; if (flags & M_EXTPG) return (m_uiotombuf_nomap(uio, how, len, align, flags)); /* * len can be zero or an arbitrary large value bound by * the total data supplied by the uio. */ if (len > 0) total = (uio->uio_resid < len) ? uio->uio_resid : len; else total = uio->uio_resid; /* * The smallest unit returned by m_getm2() is a single mbuf * with pkthdr. We can't align past it. */ if (align >= MHLEN) return (NULL); /* * Give us the full allocation or nothing. * If len is zero return the smallest empty mbuf. */ m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags); if (m == NULL) return (NULL); m->m_data += align; /* Fill all mbufs with uio data and update header information. */ for (mb = m; mb != NULL; mb = mb->m_next) { length = min(M_TRAILINGSPACE(mb), total - progress); error = uiomove(mtod(mb, void *), length, uio); if (error) { m_freem(m); return (NULL); } mb->m_len = length; progress += length; if (flags & M_PKTHDR) { m->m_pkthdr.len += length; m->m_pkthdr.memlen += MSIZE; if (mb->m_flags & M_EXT) m->m_pkthdr.memlen += mb->m_ext.ext_size; } } KASSERT(progress == total, ("%s: progress != total", __func__)); return (m); } /* * Copy data to/from an unmapped mbuf into a uio limited by len if set. */ int m_unmapped_uiomove(const struct mbuf *m, int m_off, struct uio *uio, int len) { vm_page_t pg; int error, i, off, pglen, pgoff, seglen, segoff; M_ASSERTEXTPG(m); error = 0; /* Skip over any data removed from the front. */ off = mtod(m, vm_offset_t); off += m_off; if (m->m_epg_hdrlen != 0) { if (off >= m->m_epg_hdrlen) { off -= m->m_epg_hdrlen; } else { seglen = m->m_epg_hdrlen - off; segoff = off; seglen = min(seglen, len); off = 0; len -= seglen; error = uiomove(__DECONST(void *, &m->m_epg_hdr[segoff]), seglen, uio); } } pgoff = m->m_epg_1st_off; for (i = 0; i < m->m_epg_npgs && error == 0 && len > 0; i++) { pglen = m_epg_pagelen(m, i, pgoff); if (off >= pglen) { off -= pglen; pgoff = 0; continue; } seglen = pglen - off; segoff = pgoff + off; off = 0; seglen = min(seglen, len); len -= seglen; pg = PHYS_TO_VM_PAGE(m->m_epg_pa[i]); error = uiomove_fromphys(&pg, segoff, seglen, uio); pgoff = 0; }; if (len != 0 && error == 0) { KASSERT((off + len) <= m->m_epg_trllen, ("off + len > trail (%d + %d > %d, m_off = %d)", off, len, m->m_epg_trllen, m_off)); error = uiomove(__DECONST(void *, &m->m_epg_trail[off]), len, uio); } return (error); } /* * Copy an mbuf chain into a uio limited by len if set. */ int m_mbuftouio(struct uio *uio, const struct mbuf *m, int len) { int error, length, total; int progress = 0; if (len > 0) total = min(uio->uio_resid, len); else total = uio->uio_resid; /* Fill the uio with data from the mbufs. */ for (; m != NULL; m = m->m_next) { length = min(m->m_len, total - progress); if ((m->m_flags & M_EXTPG) != 0) error = m_unmapped_uiomove(m, 0, uio, length); else error = uiomove(mtod(m, void *), length, uio); if (error) return (error); progress += length; } return (0); } /* * Create a writable copy of the mbuf chain. While doing this * we compact the chain with a goal of producing a chain with * at most two mbufs. The second mbuf in this chain is likely * to be a cluster. The primary purpose of this work is to create * a writable packet for encryption, compression, etc. The * secondary goal is to linearize the data so the data can be * passed to crypto hardware in the most efficient manner possible. */ struct mbuf * m_unshare(struct mbuf *m0, int how) { struct mbuf *m, *mprev; struct mbuf *n, *mfirst, *mlast; int len, off; mprev = NULL; for (m = m0; m != NULL; m = mprev->m_next) { /* * Regular mbufs are ignored unless there's a cluster * in front of it that we can use to coalesce. We do * the latter mainly so later clusters can be coalesced * also w/o having to handle them specially (i.e. convert * mbuf+cluster -> cluster). This optimization is heavily * influenced by the assumption that we're running over * Ethernet where MCLBYTES is large enough that the max * packet size will permit lots of coalescing into a * single cluster. This in turn permits efficient * crypto operations, especially when using hardware. */ if ((m->m_flags & M_EXT) == 0) { if (mprev && (mprev->m_flags & M_EXT) && m->m_len <= M_TRAILINGSPACE(mprev)) { /* XXX: this ignores mbuf types */ memcpy(mtod(mprev, caddr_t) + mprev->m_len, mtod(m, caddr_t), m->m_len); mprev->m_len += m->m_len; mprev->m_next = m->m_next; /* unlink from chain */ m_free(m); /* reclaim mbuf */ } else { mprev = m; } continue; } /* * Writable mbufs are left alone (for now). */ if (M_WRITABLE(m)) { mprev = m; continue; } /* * Not writable, replace with a copy or coalesce with * the previous mbuf if possible (since we have to copy * it anyway, we try to reduce the number of mbufs and * clusters so that future work is easier). */ KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags)); /* NB: we only coalesce into a cluster or larger */ if (mprev != NULL && (mprev->m_flags & M_EXT) && m->m_len <= M_TRAILINGSPACE(mprev)) { /* XXX: this ignores mbuf types */ memcpy(mtod(mprev, caddr_t) + mprev->m_len, mtod(m, caddr_t), m->m_len); mprev->m_len += m->m_len; mprev->m_next = m->m_next; /* unlink from chain */ m_free(m); /* reclaim mbuf */ continue; } /* * Allocate new space to hold the copy and copy the data. * We deal with jumbo mbufs (i.e. m_len > MCLBYTES) by * splitting them into clusters. We could just malloc a * buffer and make it external but too many device drivers * don't know how to break up the non-contiguous memory when * doing DMA. */ n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS); if (n == NULL) { m_freem(m0); return (NULL); } if (m->m_flags & M_PKTHDR) { KASSERT(mprev == NULL, ("%s: m0 %p, m %p has M_PKTHDR", __func__, m0, m)); m_move_pkthdr(n, m); } len = m->m_len; off = 0; mfirst = n; mlast = NULL; for (;;) { int cc = min(len, MCLBYTES); memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc); n->m_len = cc; if (mlast != NULL) mlast->m_next = n; mlast = n; #if 0 newipsecstat.ips_clcopied++; #endif len -= cc; if (len <= 0) break; off += cc; n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS); if (n == NULL) { m_freem(mfirst); m_freem(m0); return (NULL); } } n->m_next = m->m_next; if (mprev == NULL) m0 = mfirst; /* new head of chain */ else mprev->m_next = mfirst; /* replace old mbuf */ m_free(m); /* release old mbuf */ mprev = mfirst; } return (m0); } #ifdef MBUF_PROFILING #define MP_BUCKETS 32 /* don't just change this as things may overflow.*/ struct mbufprofile { uintmax_t wasted[MP_BUCKETS]; uintmax_t used[MP_BUCKETS]; uintmax_t segments[MP_BUCKETS]; } mbprof; void m_profile(struct mbuf *m) { int segments = 0; int used = 0; int wasted = 0; while (m) { segments++; used += m->m_len; if (m->m_flags & M_EXT) { wasted += MHLEN - sizeof(m->m_ext) + m->m_ext.ext_size - m->m_len; } else { if (m->m_flags & M_PKTHDR) wasted += MHLEN - m->m_len; else wasted += MLEN - m->m_len; } m = m->m_next; } /* be paranoid.. it helps */ if (segments > MP_BUCKETS - 1) segments = MP_BUCKETS - 1; if (used > 100000) used = 100000; if (wasted > 100000) wasted = 100000; /* store in the appropriate bucket */ /* don't bother locking. if it's slightly off, so what? */ mbprof.segments[segments]++; mbprof.used[fls(used)]++; mbprof.wasted[fls(wasted)]++; } static int mbprof_handler(SYSCTL_HANDLER_ARGS) { char buf[256]; struct sbuf sb; int error; uint64_t *p; sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req); p = &mbprof.wasted[0]; sbuf_printf(&sb, "wasted:\n" "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %ju\n", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #ifdef BIG_ARRAY p = &mbprof.wasted[16]; sbuf_printf(&sb, "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %ju\n", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #endif p = &mbprof.used[0]; sbuf_printf(&sb, "used:\n" "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %ju\n", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #ifdef BIG_ARRAY p = &mbprof.used[16]; sbuf_printf(&sb, "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %ju\n", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #endif p = &mbprof.segments[0]; sbuf_printf(&sb, "segments:\n" "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %ju\n", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #ifdef BIG_ARRAY p = &mbprof.segments[16]; sbuf_printf(&sb, "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %jju", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #endif error = sbuf_finish(&sb); sbuf_delete(&sb); return (error); } static int mbprof_clr_handler(SYSCTL_HANDLER_ARGS) { int clear, error; clear = 0; error = sysctl_handle_int(oidp, &clear, 0, req); if (error || !req->newptr) return (error); if (clear) { bzero(&mbprof, sizeof(mbprof)); } return (error); } SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, mbprof_handler, "A", "mbuf profiling statistics"); SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0, mbprof_clr_handler, "I", "clear mbuf profiling statistics"); #endif diff --git a/sys/net80211/ieee80211_proto.c b/sys/net80211/ieee80211_proto.c index 01da0c2a0768..87fed05a281c 100644 --- a/sys/net80211/ieee80211_proto.c +++ b/sys/net80211/ieee80211_proto.c @@ -1,2830 +1,2825 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2001 Atsushi Onoe * Copyright (c) 2002-2008 Sam Leffler, Errno Consulting * Copyright (c) 2012 IEEE * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); /* * IEEE 802.11 protocol support. */ #include "opt_inet.h" #include "opt_wlan.h" #include #include #include #include #include #include #include #include #include #include /* XXX for ether_sprintf */ #include #include #include #include #include #ifdef IEEE80211_SUPPORT_MESH #include #endif #include #include /* XXX tunables */ #define AGGRESSIVE_MODE_SWITCH_HYSTERESIS 3 /* pkts / 100ms */ #define HIGH_PRI_SWITCH_THRESH 10 /* pkts / 100ms */ const char *mgt_subtype_name[] = { "assoc_req", "assoc_resp", "reassoc_req", "reassoc_resp", "probe_req", "probe_resp", "timing_adv", "reserved#7", "beacon", "atim", "disassoc", "auth", "deauth", "action", "action_noack", "reserved#15" }; const char *ctl_subtype_name[] = { "reserved#0", "reserved#1", "reserved#2", "reserved#3", "reserved#4", "reserved#5", "reserved#6", "control_wrap", "bar", "ba", "ps_poll", "rts", "cts", "ack", "cf_end", "cf_end_ack" }; const char *ieee80211_opmode_name[IEEE80211_OPMODE_MAX] = { "IBSS", /* IEEE80211_M_IBSS */ "STA", /* IEEE80211_M_STA */ "WDS", /* IEEE80211_M_WDS */ "AHDEMO", /* IEEE80211_M_AHDEMO */ "HOSTAP", /* IEEE80211_M_HOSTAP */ "MONITOR", /* IEEE80211_M_MONITOR */ "MBSS" /* IEEE80211_M_MBSS */ }; const char *ieee80211_state_name[IEEE80211_S_MAX] = { "INIT", /* IEEE80211_S_INIT */ "SCAN", /* IEEE80211_S_SCAN */ "AUTH", /* IEEE80211_S_AUTH */ "ASSOC", /* IEEE80211_S_ASSOC */ "CAC", /* IEEE80211_S_CAC */ "RUN", /* IEEE80211_S_RUN */ "CSA", /* IEEE80211_S_CSA */ "SLEEP", /* IEEE80211_S_SLEEP */ }; const char *ieee80211_wme_acnames[] = { "WME_AC_BE", "WME_AC_BK", "WME_AC_VI", "WME_AC_VO", "WME_UPSD", }; /* * Reason code descriptions were (mostly) obtained from * IEEE Std 802.11-2012, pp. 442-445 Table 8-36. */ const char * ieee80211_reason_to_string(uint16_t reason) { switch (reason) { case IEEE80211_REASON_UNSPECIFIED: return ("unspecified"); case IEEE80211_REASON_AUTH_EXPIRE: return ("previous authentication is expired"); case IEEE80211_REASON_AUTH_LEAVE: return ("sending STA is leaving/has left IBSS or ESS"); case IEEE80211_REASON_ASSOC_EXPIRE: return ("disassociated due to inactivity"); case IEEE80211_REASON_ASSOC_TOOMANY: return ("too many associated STAs"); case IEEE80211_REASON_NOT_AUTHED: return ("class 2 frame received from nonauthenticated STA"); case IEEE80211_REASON_NOT_ASSOCED: return ("class 3 frame received from nonassociated STA"); case IEEE80211_REASON_ASSOC_LEAVE: return ("sending STA is leaving/has left BSS"); case IEEE80211_REASON_ASSOC_NOT_AUTHED: return ("STA requesting (re)association is not authenticated"); case IEEE80211_REASON_DISASSOC_PWRCAP_BAD: return ("information in the Power Capability element is " "unacceptable"); case IEEE80211_REASON_DISASSOC_SUPCHAN_BAD: return ("information in the Supported Channels element is " "unacceptable"); case IEEE80211_REASON_IE_INVALID: return ("invalid element"); case IEEE80211_REASON_MIC_FAILURE: return ("MIC failure"); case IEEE80211_REASON_4WAY_HANDSHAKE_TIMEOUT: return ("4-Way handshake timeout"); case IEEE80211_REASON_GROUP_KEY_UPDATE_TIMEOUT: return ("group key update timeout"); case IEEE80211_REASON_IE_IN_4WAY_DIFFERS: return ("element in 4-Way handshake different from " "(re)association request/probe response/beacon frame"); case IEEE80211_REASON_GROUP_CIPHER_INVALID: return ("invalid group cipher"); case IEEE80211_REASON_PAIRWISE_CIPHER_INVALID: return ("invalid pairwise cipher"); case IEEE80211_REASON_AKMP_INVALID: return ("invalid AKMP"); case IEEE80211_REASON_UNSUPP_RSN_IE_VERSION: return ("unsupported version in RSN IE"); case IEEE80211_REASON_INVALID_RSN_IE_CAP: return ("invalid capabilities in RSN IE"); case IEEE80211_REASON_802_1X_AUTH_FAILED: return ("IEEE 802.1X authentication failed"); case IEEE80211_REASON_CIPHER_SUITE_REJECTED: return ("cipher suite rejected because of the security " "policy"); case IEEE80211_REASON_UNSPECIFIED_QOS: return ("unspecified (QoS-related)"); case IEEE80211_REASON_INSUFFICIENT_BW: return ("QoS AP lacks sufficient bandwidth for this QoS STA"); case IEEE80211_REASON_TOOMANY_FRAMES: return ("too many frames need to be acknowledged"); case IEEE80211_REASON_OUTSIDE_TXOP: return ("STA is transmitting outside the limits of its TXOPs"); case IEEE80211_REASON_LEAVING_QBSS: return ("requested from peer STA (the STA is " "resetting/leaving the BSS)"); case IEEE80211_REASON_BAD_MECHANISM: return ("requested from peer STA (it does not want to use " "the mechanism)"); case IEEE80211_REASON_SETUP_NEEDED: return ("requested from peer STA (setup is required for the " "used mechanism)"); case IEEE80211_REASON_TIMEOUT: return ("requested from peer STA (timeout)"); case IEEE80211_REASON_PEER_LINK_CANCELED: return ("SME cancels the mesh peering instance (not related " "to the maximum number of peer mesh STAs)"); case IEEE80211_REASON_MESH_MAX_PEERS: return ("maximum number of peer mesh STAs was reached"); case IEEE80211_REASON_MESH_CPVIOLATION: return ("the received information violates the Mesh " "Configuration policy configured in the mesh STA " "profile"); case IEEE80211_REASON_MESH_CLOSE_RCVD: return ("the mesh STA has received a Mesh Peering Close " "message requesting to close the mesh peering"); case IEEE80211_REASON_MESH_MAX_RETRIES: return ("the mesh STA has resent dot11MeshMaxRetries Mesh " "Peering Open messages, without receiving a Mesh " "Peering Confirm message"); case IEEE80211_REASON_MESH_CONFIRM_TIMEOUT: return ("the confirmTimer for the mesh peering instance times " "out"); case IEEE80211_REASON_MESH_INVALID_GTK: return ("the mesh STA fails to unwrap the GTK or the values " "in the wrapped contents do not match"); case IEEE80211_REASON_MESH_INCONS_PARAMS: return ("the mesh STA receives inconsistent information about " "the mesh parameters between Mesh Peering Management " "frames"); case IEEE80211_REASON_MESH_INVALID_SECURITY: return ("the mesh STA fails the authenticated mesh peering " "exchange because due to failure in selecting " "pairwise/group ciphersuite"); case IEEE80211_REASON_MESH_PERR_NO_PROXY: return ("the mesh STA does not have proxy information for " "this external destination"); case IEEE80211_REASON_MESH_PERR_NO_FI: return ("the mesh STA does not have forwarding information " "for this destination"); case IEEE80211_REASON_MESH_PERR_DEST_UNREACH: return ("the mesh STA determines that the link to the next " "hop of an active path in its forwarding information " "is no longer usable"); case IEEE80211_REASON_MESH_MAC_ALRDY_EXISTS_MBSS: return ("the MAC address of the STA already exists in the " "mesh BSS"); case IEEE80211_REASON_MESH_CHAN_SWITCH_REG: return ("the mesh STA performs channel switch to meet " "regulatory requirements"); case IEEE80211_REASON_MESH_CHAN_SWITCH_UNSPEC: return ("the mesh STA performs channel switch with " "unspecified reason"); default: return ("reserved/unknown"); } } static void beacon_miss(void *, int); static void beacon_swmiss(void *, int); static void parent_updown(void *, int); static void update_mcast(void *, int); static void update_promisc(void *, int); static void update_channel(void *, int); static void update_chw(void *, int); static void vap_update_wme(void *, int); static void vap_update_slot(void *, int); static void restart_vaps(void *, int); static void vap_update_erp_protmode(void *, int); static void vap_update_preamble(void *, int); static void vap_update_ht_protmode(void *, int); static void ieee80211_newstate_cb(void *, int); static struct ieee80211_node *vap_update_bss(struct ieee80211vap *, struct ieee80211_node *); static int null_raw_xmit(struct ieee80211_node *ni, struct mbuf *m, const struct ieee80211_bpf_params *params) { ic_printf(ni->ni_ic, "missing ic_raw_xmit callback, drop frame\n"); m_freem(m); return ENETDOWN; } void ieee80211_proto_attach(struct ieee80211com *ic) { uint8_t hdrlen; /* override the 802.3 setting */ hdrlen = ic->ic_headroom + sizeof(struct ieee80211_qosframe_addr4) + IEEE80211_WEP_IVLEN + IEEE80211_WEP_KIDLEN + IEEE80211_WEP_EXTIVLEN; /* XXX no way to recalculate on ifdetach */ - if (ALIGN(hdrlen) > max_linkhdr) { - /* XXX sanity check... */ - max_linkhdr = ALIGN(hdrlen); - max_hdr = max_linkhdr + max_protohdr; - max_datalen = MHLEN - max_hdr; - } + max_linkhdr_grow(ALIGN(hdrlen)); //ic->ic_protmode = IEEE80211_PROT_CTSONLY; TASK_INIT(&ic->ic_parent_task, 0, parent_updown, ic); TASK_INIT(&ic->ic_mcast_task, 0, update_mcast, ic); TASK_INIT(&ic->ic_promisc_task, 0, update_promisc, ic); TASK_INIT(&ic->ic_chan_task, 0, update_channel, ic); TASK_INIT(&ic->ic_bmiss_task, 0, beacon_miss, ic); TASK_INIT(&ic->ic_chw_task, 0, update_chw, ic); TASK_INIT(&ic->ic_restart_task, 0, restart_vaps, ic); ic->ic_wme.wme_hipri_switch_hysteresis = AGGRESSIVE_MODE_SWITCH_HYSTERESIS; /* initialize management frame handlers */ ic->ic_send_mgmt = ieee80211_send_mgmt; ic->ic_raw_xmit = null_raw_xmit; ieee80211_adhoc_attach(ic); ieee80211_sta_attach(ic); ieee80211_wds_attach(ic); ieee80211_hostap_attach(ic); #ifdef IEEE80211_SUPPORT_MESH ieee80211_mesh_attach(ic); #endif ieee80211_monitor_attach(ic); } void ieee80211_proto_detach(struct ieee80211com *ic) { ieee80211_monitor_detach(ic); #ifdef IEEE80211_SUPPORT_MESH ieee80211_mesh_detach(ic); #endif ieee80211_hostap_detach(ic); ieee80211_wds_detach(ic); ieee80211_adhoc_detach(ic); ieee80211_sta_detach(ic); } static void null_update_beacon(struct ieee80211vap *vap, int item) { } void ieee80211_proto_vattach(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; struct ifnet *ifp = vap->iv_ifp; int i; /* override the 802.3 setting */ ifp->if_hdrlen = ic->ic_headroom + sizeof(struct ieee80211_qosframe_addr4) + IEEE80211_WEP_IVLEN + IEEE80211_WEP_KIDLEN + IEEE80211_WEP_EXTIVLEN; vap->iv_rtsthreshold = IEEE80211_RTS_DEFAULT; vap->iv_fragthreshold = IEEE80211_FRAG_DEFAULT; vap->iv_bmiss_max = IEEE80211_BMISS_MAX; callout_init_mtx(&vap->iv_swbmiss, IEEE80211_LOCK_OBJ(ic), 0); callout_init(&vap->iv_mgtsend, 1); TASK_INIT(&vap->iv_nstate_task, 0, ieee80211_newstate_cb, vap); TASK_INIT(&vap->iv_swbmiss_task, 0, beacon_swmiss, vap); TASK_INIT(&vap->iv_wme_task, 0, vap_update_wme, vap); TASK_INIT(&vap->iv_slot_task, 0, vap_update_slot, vap); TASK_INIT(&vap->iv_erp_protmode_task, 0, vap_update_erp_protmode, vap); TASK_INIT(&vap->iv_ht_protmode_task, 0, vap_update_ht_protmode, vap); TASK_INIT(&vap->iv_preamble_task, 0, vap_update_preamble, vap); /* * Install default tx rate handling: no fixed rate, lowest * supported rate for mgmt and multicast frames. Default * max retry count. These settings can be changed by the * driver and/or user applications. */ for (i = IEEE80211_MODE_11A; i < IEEE80211_MODE_MAX; i++) { if (isclr(ic->ic_modecaps, i)) continue; const struct ieee80211_rateset *rs = &ic->ic_sup_rates[i]; vap->iv_txparms[i].ucastrate = IEEE80211_FIXED_RATE_NONE; /* * Setting the management rate to MCS 0 assumes that the * BSS Basic rate set is empty and the BSS Basic MCS set * is not. * * Since we're not checking this, default to the lowest * defined rate for this mode. * * At least one 11n AP (DLINK DIR-825) is reported to drop * some MCS management traffic (eg BA response frames.) * * See also: 9.6.0 of the 802.11n-2009 specification. */ #ifdef NOTYET if (i == IEEE80211_MODE_11NA || i == IEEE80211_MODE_11NG) { vap->iv_txparms[i].mgmtrate = 0 | IEEE80211_RATE_MCS; vap->iv_txparms[i].mcastrate = 0 | IEEE80211_RATE_MCS; } else { vap->iv_txparms[i].mgmtrate = rs->rs_rates[0] & IEEE80211_RATE_VAL; vap->iv_txparms[i].mcastrate = rs->rs_rates[0] & IEEE80211_RATE_VAL; } #endif vap->iv_txparms[i].mgmtrate = rs->rs_rates[0] & IEEE80211_RATE_VAL; vap->iv_txparms[i].mcastrate = rs->rs_rates[0] & IEEE80211_RATE_VAL; vap->iv_txparms[i].maxretry = IEEE80211_TXMAX_DEFAULT; } vap->iv_roaming = IEEE80211_ROAMING_AUTO; vap->iv_update_beacon = null_update_beacon; vap->iv_deliver_data = ieee80211_deliver_data; vap->iv_protmode = IEEE80211_PROT_CTSONLY; vap->iv_update_bss = vap_update_bss; /* attach support for operating mode */ ic->ic_vattach[vap->iv_opmode](vap); } void ieee80211_proto_vdetach(struct ieee80211vap *vap) { #define FREEAPPIE(ie) do { \ if (ie != NULL) \ IEEE80211_FREE(ie, M_80211_NODE_IE); \ } while (0) /* * Detach operating mode module. */ if (vap->iv_opdetach != NULL) vap->iv_opdetach(vap); /* * This should not be needed as we detach when reseting * the state but be conservative here since the * authenticator may do things like spawn kernel threads. */ if (vap->iv_auth->ia_detach != NULL) vap->iv_auth->ia_detach(vap); /* * Detach any ACL'ator. */ if (vap->iv_acl != NULL) vap->iv_acl->iac_detach(vap); FREEAPPIE(vap->iv_appie_beacon); FREEAPPIE(vap->iv_appie_probereq); FREEAPPIE(vap->iv_appie_proberesp); FREEAPPIE(vap->iv_appie_assocreq); FREEAPPIE(vap->iv_appie_assocresp); FREEAPPIE(vap->iv_appie_wpa); #undef FREEAPPIE } /* * Simple-minded authenticator module support. */ #define IEEE80211_AUTH_MAX (IEEE80211_AUTH_WPA+1) /* XXX well-known names */ static const char *auth_modnames[IEEE80211_AUTH_MAX] = { "wlan_internal", /* IEEE80211_AUTH_NONE */ "wlan_internal", /* IEEE80211_AUTH_OPEN */ "wlan_internal", /* IEEE80211_AUTH_SHARED */ "wlan_xauth", /* IEEE80211_AUTH_8021X */ "wlan_internal", /* IEEE80211_AUTH_AUTO */ "wlan_xauth", /* IEEE80211_AUTH_WPA */ }; static const struct ieee80211_authenticator *authenticators[IEEE80211_AUTH_MAX]; static const struct ieee80211_authenticator auth_internal = { .ia_name = "wlan_internal", .ia_attach = NULL, .ia_detach = NULL, .ia_node_join = NULL, .ia_node_leave = NULL, }; /* * Setup internal authenticators once; they are never unregistered. */ static void ieee80211_auth_setup(void) { ieee80211_authenticator_register(IEEE80211_AUTH_OPEN, &auth_internal); ieee80211_authenticator_register(IEEE80211_AUTH_SHARED, &auth_internal); ieee80211_authenticator_register(IEEE80211_AUTH_AUTO, &auth_internal); } SYSINIT(wlan_auth, SI_SUB_DRIVERS, SI_ORDER_FIRST, ieee80211_auth_setup, NULL); const struct ieee80211_authenticator * ieee80211_authenticator_get(int auth) { if (auth >= IEEE80211_AUTH_MAX) return NULL; if (authenticators[auth] == NULL) ieee80211_load_module(auth_modnames[auth]); return authenticators[auth]; } void ieee80211_authenticator_register(int type, const struct ieee80211_authenticator *auth) { if (type >= IEEE80211_AUTH_MAX) return; authenticators[type] = auth; } void ieee80211_authenticator_unregister(int type) { if (type >= IEEE80211_AUTH_MAX) return; authenticators[type] = NULL; } /* * Very simple-minded ACL module support. */ /* XXX just one for now */ static const struct ieee80211_aclator *acl = NULL; void ieee80211_aclator_register(const struct ieee80211_aclator *iac) { printf("wlan: %s acl policy registered\n", iac->iac_name); acl = iac; } void ieee80211_aclator_unregister(const struct ieee80211_aclator *iac) { if (acl == iac) acl = NULL; printf("wlan: %s acl policy unregistered\n", iac->iac_name); } const struct ieee80211_aclator * ieee80211_aclator_get(const char *name) { if (acl == NULL) ieee80211_load_module("wlan_acl"); return acl != NULL && strcmp(acl->iac_name, name) == 0 ? acl : NULL; } void ieee80211_print_essid(const uint8_t *essid, int len) { const uint8_t *p; int i; if (len > IEEE80211_NWID_LEN) len = IEEE80211_NWID_LEN; /* determine printable or not */ for (i = 0, p = essid; i < len; i++, p++) { if (*p < ' ' || *p > 0x7e) break; } if (i == len) { printf("\""); for (i = 0, p = essid; i < len; i++, p++) printf("%c", *p); printf("\""); } else { printf("0x"); for (i = 0, p = essid; i < len; i++, p++) printf("%02x", *p); } } void ieee80211_dump_pkt(struct ieee80211com *ic, const uint8_t *buf, int len, int rate, int rssi) { const struct ieee80211_frame *wh; int i; wh = (const struct ieee80211_frame *)buf; switch (wh->i_fc[1] & IEEE80211_FC1_DIR_MASK) { case IEEE80211_FC1_DIR_NODS: printf("NODS %s", ether_sprintf(wh->i_addr2)); printf("->%s", ether_sprintf(wh->i_addr1)); printf("(%s)", ether_sprintf(wh->i_addr3)); break; case IEEE80211_FC1_DIR_TODS: printf("TODS %s", ether_sprintf(wh->i_addr2)); printf("->%s", ether_sprintf(wh->i_addr3)); printf("(%s)", ether_sprintf(wh->i_addr1)); break; case IEEE80211_FC1_DIR_FROMDS: printf("FRDS %s", ether_sprintf(wh->i_addr3)); printf("->%s", ether_sprintf(wh->i_addr1)); printf("(%s)", ether_sprintf(wh->i_addr2)); break; case IEEE80211_FC1_DIR_DSTODS: printf("DSDS %s", ether_sprintf((const uint8_t *)&wh[1])); printf("->%s", ether_sprintf(wh->i_addr3)); printf("(%s", ether_sprintf(wh->i_addr2)); printf("->%s)", ether_sprintf(wh->i_addr1)); break; } switch (wh->i_fc[0] & IEEE80211_FC0_TYPE_MASK) { case IEEE80211_FC0_TYPE_DATA: printf(" data"); break; case IEEE80211_FC0_TYPE_MGT: printf(" %s", ieee80211_mgt_subtype_name(wh->i_fc[0])); break; default: printf(" type#%d", wh->i_fc[0] & IEEE80211_FC0_TYPE_MASK); break; } if (IEEE80211_QOS_HAS_SEQ(wh)) { const struct ieee80211_qosframe *qwh = (const struct ieee80211_qosframe *)buf; printf(" QoS [TID %u%s]", qwh->i_qos[0] & IEEE80211_QOS_TID, qwh->i_qos[0] & IEEE80211_QOS_ACKPOLICY ? " ACM" : ""); } if (IEEE80211_IS_PROTECTED(wh)) { int off; off = ieee80211_anyhdrspace(ic, wh); printf(" WEP [IV %.02x %.02x %.02x", buf[off+0], buf[off+1], buf[off+2]); if (buf[off+IEEE80211_WEP_IVLEN] & IEEE80211_WEP_EXTIV) printf(" %.02x %.02x %.02x", buf[off+4], buf[off+5], buf[off+6]); printf(" KID %u]", buf[off+IEEE80211_WEP_IVLEN] >> 6); } if (rate >= 0) printf(" %dM", rate / 2); if (rssi >= 0) printf(" +%d", rssi); printf("\n"); if (len > 0) { for (i = 0; i < len; i++) { if ((i & 1) == 0) printf(" "); printf("%02x", buf[i]); } printf("\n"); } } static __inline int findrix(const struct ieee80211_rateset *rs, int r) { int i; for (i = 0; i < rs->rs_nrates; i++) if ((rs->rs_rates[i] & IEEE80211_RATE_VAL) == r) return i; return -1; } int ieee80211_fix_rate(struct ieee80211_node *ni, struct ieee80211_rateset *nrs, int flags) { struct ieee80211vap *vap = ni->ni_vap; struct ieee80211com *ic = ni->ni_ic; int i, j, rix, error; int okrate, badrate, fixedrate, ucastrate; const struct ieee80211_rateset *srs; uint8_t r; error = 0; okrate = badrate = 0; ucastrate = vap->iv_txparms[ieee80211_chan2mode(ni->ni_chan)].ucastrate; if (ucastrate != IEEE80211_FIXED_RATE_NONE) { /* * Workaround awkwardness with fixed rate. We are called * to check both the legacy rate set and the HT rate set * but we must apply any legacy fixed rate check only to the * legacy rate set and vice versa. We cannot tell what type * of rate set we've been given (legacy or HT) but we can * distinguish the fixed rate type (MCS have 0x80 set). * So to deal with this the caller communicates whether to * check MCS or legacy rate using the flags and we use the * type of any fixed rate to avoid applying an MCS to a * legacy rate and vice versa. */ if (ucastrate & 0x80) { if (flags & IEEE80211_F_DOFRATE) flags &= ~IEEE80211_F_DOFRATE; } else if ((ucastrate & 0x80) == 0) { if (flags & IEEE80211_F_DOFMCS) flags &= ~IEEE80211_F_DOFMCS; } /* NB: required to make MCS match below work */ ucastrate &= IEEE80211_RATE_VAL; } fixedrate = IEEE80211_FIXED_RATE_NONE; /* * XXX we are called to process both MCS and legacy rates; * we must use the appropriate basic rate set or chaos will * ensue; for now callers that want MCS must supply * IEEE80211_F_DOBRS; at some point we'll need to split this * function so there are two variants, one for MCS and one * for legacy rates. */ if (flags & IEEE80211_F_DOBRS) srs = (const struct ieee80211_rateset *) ieee80211_get_suphtrates(ic, ni->ni_chan); else srs = ieee80211_get_suprates(ic, ni->ni_chan); for (i = 0; i < nrs->rs_nrates; ) { if (flags & IEEE80211_F_DOSORT) { /* * Sort rates. */ for (j = i + 1; j < nrs->rs_nrates; j++) { if (IEEE80211_RV(nrs->rs_rates[i]) > IEEE80211_RV(nrs->rs_rates[j])) { r = nrs->rs_rates[i]; nrs->rs_rates[i] = nrs->rs_rates[j]; nrs->rs_rates[j] = r; } } } r = nrs->rs_rates[i] & IEEE80211_RATE_VAL; badrate = r; /* * Check for fixed rate. */ if (r == ucastrate) fixedrate = r; /* * Check against supported rates. */ rix = findrix(srs, r); if (flags & IEEE80211_F_DONEGO) { if (rix < 0) { /* * A rate in the node's rate set is not * supported. If this is a basic rate and we * are operating as a STA then this is an error. * Otherwise we just discard/ignore the rate. */ if ((flags & IEEE80211_F_JOIN) && (nrs->rs_rates[i] & IEEE80211_RATE_BASIC)) error++; } else if ((flags & IEEE80211_F_JOIN) == 0) { /* * Overwrite with the supported rate * value so any basic rate bit is set. */ nrs->rs_rates[i] = srs->rs_rates[rix]; } } if ((flags & IEEE80211_F_DODEL) && rix < 0) { /* * Delete unacceptable rates. */ nrs->rs_nrates--; for (j = i; j < nrs->rs_nrates; j++) nrs->rs_rates[j] = nrs->rs_rates[j + 1]; nrs->rs_rates[j] = 0; continue; } if (rix >= 0) okrate = nrs->rs_rates[i]; i++; } if (okrate == 0 || error != 0 || ((flags & (IEEE80211_F_DOFRATE|IEEE80211_F_DOFMCS)) && fixedrate != ucastrate)) { IEEE80211_NOTE(vap, IEEE80211_MSG_XRATE | IEEE80211_MSG_11N, ni, "%s: flags 0x%x okrate %d error %d fixedrate 0x%x " "ucastrate %x\n", __func__, fixedrate, ucastrate, flags); return badrate | IEEE80211_RATE_BASIC; } else return IEEE80211_RV(okrate); } /* * Reset 11g-related state. * * This is for per-VAP ERP/11g state. * * Eventually everything in ieee80211_reset_erp() will be * per-VAP and in here. */ void ieee80211_vap_reset_erp(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; vap->iv_nonerpsta = 0; vap->iv_longslotsta = 0; vap->iv_flags &= ~IEEE80211_F_USEPROT; /* * Set short preamble and ERP barker-preamble flags. */ if (IEEE80211_IS_CHAN_A(ic->ic_curchan) || (vap->iv_caps & IEEE80211_C_SHPREAMBLE)) { vap->iv_flags |= IEEE80211_F_SHPREAMBLE; vap->iv_flags &= ~IEEE80211_F_USEBARKER; } else { vap->iv_flags &= ~IEEE80211_F_SHPREAMBLE; vap->iv_flags |= IEEE80211_F_USEBARKER; } /* * Short slot time is enabled only when operating in 11g * and not in an IBSS. We must also honor whether or not * the driver is capable of doing it. */ ieee80211_vap_set_shortslottime(vap, IEEE80211_IS_CHAN_A(ic->ic_curchan) || IEEE80211_IS_CHAN_HT(ic->ic_curchan) || (IEEE80211_IS_CHAN_ANYG(ic->ic_curchan) && vap->iv_opmode == IEEE80211_M_HOSTAP && (ic->ic_caps & IEEE80211_C_SHSLOT))); } /* * Reset 11g-related state. * * Note this resets the global state and a caller should schedule * a re-check of all the VAPs after setup to update said state. */ void ieee80211_reset_erp(struct ieee80211com *ic) { #if 0 ic->ic_flags &= ~IEEE80211_F_USEPROT; /* * Set short preamble and ERP barker-preamble flags. */ if (IEEE80211_IS_CHAN_A(ic->ic_curchan) || (ic->ic_caps & IEEE80211_C_SHPREAMBLE)) { ic->ic_flags |= IEEE80211_F_SHPREAMBLE; ic->ic_flags &= ~IEEE80211_F_USEBARKER; } else { ic->ic_flags &= ~IEEE80211_F_SHPREAMBLE; ic->ic_flags |= IEEE80211_F_USEBARKER; } #endif /* XXX TODO: schedule a new per-VAP ERP calculation */ } static struct ieee80211_node * vap_update_bss(struct ieee80211vap *vap, struct ieee80211_node *ni) { struct ieee80211_node *obss; obss = vap->iv_bss; vap->iv_bss = ni; return (obss); } /* * Deferred slot time update. * * For per-VAP slot time configuration, call the VAP * method if the VAP requires it. Otherwise, just call the * older global method. * * If the per-VAP method is called then it's expected that * the driver/firmware will take care of turning the per-VAP * flags into slot time configuration. * * If the per-VAP method is not called then the global flags will be * flipped into sync with the VAPs; ic_flags IEEE80211_F_SHSLOT will * be set only if all of the vaps will have it set. * * Look at the comments for vap_update_erp_protmode() for more * background; this assumes all VAPs are on the same channel. */ static void vap_update_slot(void *arg, int npending) { struct ieee80211vap *vap = arg; struct ieee80211com *ic = vap->iv_ic; struct ieee80211vap *iv; int num_shslot = 0, num_lgslot = 0; /* * Per-VAP path - we've already had the flags updated; * so just notify the driver and move on. */ if (vap->iv_updateslot != NULL) { vap->iv_updateslot(vap); return; } /* * Iterate over all of the VAP flags to update the * global flag. * * If all vaps have short slot enabled then flip on * short slot. If any vap has it disabled then * we leave it globally disabled. This should provide * correct behaviour in a multi-BSS scenario where * at least one VAP has short slot disabled for some * reason. */ IEEE80211_LOCK(ic); TAILQ_FOREACH(iv, &ic->ic_vaps, iv_next) { if (iv->iv_flags & IEEE80211_F_SHSLOT) num_shslot++; else num_lgslot++; } /* * It looks backwards but - if the number of short slot VAPs * is zero then we're not short slot. Else, we have one * or more short slot VAPs and we're checking to see if ANY * of them have short slot disabled. */ if (num_shslot == 0) ic->ic_flags &= ~IEEE80211_F_SHSLOT; else if (num_lgslot == 0) ic->ic_flags |= IEEE80211_F_SHSLOT; IEEE80211_UNLOCK(ic); /* * Call the driver with our new global slot time flags. */ if (ic->ic_updateslot != NULL) ic->ic_updateslot(ic); } /* * Deferred ERP protmode update. * * This currently calculates the global ERP protection mode flag * based on each of the VAPs. Any VAP with it enabled is enough * for the global flag to be enabled. All VAPs with it disabled * is enough for it to be disabled. * * This may make sense right now for the supported hardware where * net80211 is controlling the single channel configuration, but * offload firmware that's doing channel changes (eg off-channel * TDLS, off-channel STA, off-channel P2P STA/AP) may get some * silly looking flag updates. * * Ideally the protection mode calculation is done based on the * channel, and all VAPs using that channel will inherit it. * But until that's what net80211 does, this wil have to do. */ static void vap_update_erp_protmode(void *arg, int npending) { struct ieee80211vap *vap = arg; struct ieee80211com *ic = vap->iv_ic; struct ieee80211vap *iv; int enable_protmode = 0; int non_erp_present = 0; /* * Iterate over all of the VAPs to calculate the overlapping * ERP protection mode configuration and ERP present math. * * For now we assume that if a driver can handle this per-VAP * then it'll ignore the ic->ic_protmode variant and instead * will look at the vap related flags. */ IEEE80211_LOCK(ic); TAILQ_FOREACH(iv, &ic->ic_vaps, iv_next) { if (iv->iv_flags & IEEE80211_F_USEPROT) enable_protmode = 1; if (iv->iv_flags_ext & IEEE80211_FEXT_NONERP_PR) non_erp_present = 1; } if (enable_protmode) ic->ic_flags |= IEEE80211_F_USEPROT; else ic->ic_flags &= ~IEEE80211_F_USEPROT; if (non_erp_present) ic->ic_flags_ext |= IEEE80211_FEXT_NONERP_PR; else ic->ic_flags_ext &= ~IEEE80211_FEXT_NONERP_PR; /* Beacon update on all VAPs */ ieee80211_notify_erp_locked(ic); IEEE80211_UNLOCK(ic); IEEE80211_DPRINTF(vap, IEEE80211_MSG_DEBUG, "%s: called; enable_protmode=%d, non_erp_present=%d\n", __func__, enable_protmode, non_erp_present); /* * Now that the global configuration flags are calculated, * notify the VAP about its configuration. * * The global flags will be used when assembling ERP IEs * for multi-VAP operation, even if it's on a different * channel. Yes, that's going to need fixing in the * future. */ if (vap->iv_erp_protmode_update != NULL) vap->iv_erp_protmode_update(vap); } /* * Deferred ERP short preamble/barker update. * * All VAPs need to use short preamble for it to be globally * enabled or not. * * Look at the comments for vap_update_erp_protmode() for more * background; this assumes all VAPs are on the same channel. */ static void vap_update_preamble(void *arg, int npending) { struct ieee80211vap *vap = arg; struct ieee80211com *ic = vap->iv_ic; struct ieee80211vap *iv; int barker_count = 0, short_preamble_count = 0, count = 0; /* * Iterate over all of the VAPs to calculate the overlapping * short or long preamble configuration. * * For now we assume that if a driver can handle this per-VAP * then it'll ignore the ic->ic_flags variant and instead * will look at the vap related flags. */ IEEE80211_LOCK(ic); TAILQ_FOREACH(iv, &ic->ic_vaps, iv_next) { if (iv->iv_flags & IEEE80211_F_USEBARKER) barker_count++; if (iv->iv_flags & IEEE80211_F_SHPREAMBLE) short_preamble_count++; count++; } /* * As with vap_update_erp_protmode(), the global flags are * currently used for beacon IEs. */ IEEE80211_DPRINTF(vap, IEEE80211_MSG_DEBUG, "%s: called; barker_count=%d, short_preamble_count=%d\n", __func__, barker_count, short_preamble_count); /* * Only flip on short preamble if all of the VAPs support * it. */ if (barker_count == 0 && short_preamble_count == count) { ic->ic_flags |= IEEE80211_F_SHPREAMBLE; ic->ic_flags &= ~IEEE80211_F_USEBARKER; } else { ic->ic_flags &= ~IEEE80211_F_SHPREAMBLE; ic->ic_flags |= IEEE80211_F_USEBARKER; } IEEE80211_DPRINTF(vap, IEEE80211_MSG_DEBUG, "%s: global barker=%d preamble=%d\n", __func__, !! (ic->ic_flags & IEEE80211_F_USEBARKER), !! (ic->ic_flags & IEEE80211_F_SHPREAMBLE)); /* Beacon update on all VAPs */ ieee80211_notify_erp_locked(ic); IEEE80211_UNLOCK(ic); /* Driver notification */ if (vap->iv_erp_protmode_update != NULL) vap->iv_preamble_update(vap); } /* * Deferred HT protmode update and beacon update. * * Look at the comments for vap_update_erp_protmode() for more * background; this assumes all VAPs are on the same channel. */ static void vap_update_ht_protmode(void *arg, int npending) { struct ieee80211vap *vap = arg; struct ieee80211vap *iv; struct ieee80211com *ic = vap->iv_ic; int num_vaps = 0, num_pure = 0; int num_optional = 0, num_ht2040 = 0, num_nonht = 0; int num_ht_sta = 0, num_ht40_sta = 0, num_sta = 0; int num_nonhtpr = 0; /* * Iterate over all of the VAPs to calculate everything. * * There are a few different flags to calculate: * * + whether there's HT only or HT+legacy stations; * + whether there's HT20, HT40, or HT20+HT40 stations; * + whether the desired protection mode is mixed, pure or * one of the two above. * * For now we assume that if a driver can handle this per-VAP * then it'll ignore the ic->ic_htprotmode / ic->ic_curhtprotmode * variant and instead will look at the vap related variables. * * XXX TODO: non-greenfield STAs present (IEEE80211_HTINFO_NONGF_PRESENT) ! */ IEEE80211_LOCK(ic); TAILQ_FOREACH(iv, &ic->ic_vaps, iv_next) { num_vaps++; /* overlapping BSSes advertising non-HT status present */ if (iv->iv_flags_ht & IEEE80211_FHT_NONHT_PR) num_nonht++; /* Operating mode flags */ if (iv->iv_curhtprotmode & IEEE80211_HTINFO_NONHT_PRESENT) num_nonhtpr++; switch (iv->iv_curhtprotmode & IEEE80211_HTINFO_OPMODE) { case IEEE80211_HTINFO_OPMODE_PURE: num_pure++; break; case IEEE80211_HTINFO_OPMODE_PROTOPT: num_optional++; break; case IEEE80211_HTINFO_OPMODE_HT20PR: num_ht2040++; break; } IEEE80211_DPRINTF(vap, IEEE80211_MSG_11N, "%s: vap %s: nonht_pr=%d, curhtprotmode=0x%02x\n", __func__, ieee80211_get_vap_ifname(iv), !! (iv->iv_flags_ht & IEEE80211_FHT_NONHT_PR), iv->iv_curhtprotmode); num_ht_sta += iv->iv_ht_sta_assoc; num_ht40_sta += iv->iv_ht40_sta_assoc; num_sta += iv->iv_sta_assoc; } /* * Step 1 - if any VAPs indicate NONHT_PR set (overlapping BSS * non-HT present), set it here. This shouldn't be used by * anything but the old overlapping BSS logic so if any drivers * consume it, it's up to date. */ if (num_nonht > 0) ic->ic_flags_ht |= IEEE80211_FHT_NONHT_PR; else ic->ic_flags_ht &= ~IEEE80211_FHT_NONHT_PR; /* * Step 2 - default HT protection mode to MIXED (802.11-2016 10.26.3.1.) * * + If all VAPs are PURE, we can stay PURE. * + If all VAPs are PROTOPT, we can go to PROTOPT. * + If any VAP has HT20PR then it sees at least a HT40+HT20 station. * Note that we may have a VAP with one HT20 and a VAP with one HT40; * So we look at the sum ht and sum ht40 sta counts; if we have a * HT station and the HT20 != HT40 count, we have to do HT20PR here. * Note all stations need to be HT for this to be an option. * + The fall-through is MIXED, because it means we have some odd * non HT40-involved combination of opmode and this is the most * sensible default. */ ic->ic_curhtprotmode = IEEE80211_HTINFO_OPMODE_MIXED; if (num_pure == num_vaps) ic->ic_curhtprotmode = IEEE80211_HTINFO_OPMODE_PURE; if (num_optional == num_vaps) ic->ic_curhtprotmode = IEEE80211_HTINFO_OPMODE_PROTOPT; /* * Note: we need /a/ HT40 station somewhere for this to * be a possibility. */ if ((num_ht2040 > 0) || ((num_ht_sta > 0) && (num_ht40_sta > 0) && (num_ht_sta != num_ht40_sta))) ic->ic_curhtprotmode = IEEE80211_HTINFO_OPMODE_HT20PR; /* * Step 3 - if any of the stations across the VAPs are * non-HT then this needs to be flipped back to MIXED. */ if (num_ht_sta != num_sta) ic->ic_curhtprotmode = IEEE80211_HTINFO_OPMODE_MIXED; /* * Step 4 - If we see any overlapping BSS non-HT stations * via beacons then flip on NONHT_PRESENT. */ if (num_nonhtpr > 0) ic->ic_curhtprotmode |= IEEE80211_HTINFO_NONHT_PRESENT; /* Notify all VAPs to potentially update their beacons */ TAILQ_FOREACH(iv, &ic->ic_vaps, iv_next) ieee80211_htinfo_notify(iv); IEEE80211_UNLOCK(ic); IEEE80211_DPRINTF(vap, IEEE80211_MSG_11N, "%s: global: nonht_pr=%d ht_opmode=0x%02x\n", __func__, !! (ic->ic_flags_ht & IEEE80211_FHT_NONHT_PR), ic->ic_curhtprotmode); /* Driver update */ if (vap->iv_erp_protmode_update != NULL) vap->iv_ht_protmode_update(vap); } /* * Set the short slot time state and notify the driver. * * This is the per-VAP slot time state. */ void ieee80211_vap_set_shortslottime(struct ieee80211vap *vap, int onoff) { struct ieee80211com *ic = vap->iv_ic; /* XXX lock? */ /* * Only modify the per-VAP slot time. */ if (onoff) vap->iv_flags |= IEEE80211_F_SHSLOT; else vap->iv_flags &= ~IEEE80211_F_SHSLOT; IEEE80211_DPRINTF(vap, IEEE80211_MSG_DEBUG, "%s: called; onoff=%d\n", __func__, onoff); /* schedule the deferred slot flag update and update */ ieee80211_runtask(ic, &vap->iv_slot_task); } /* * Update the VAP short /long / barker preamble state and * update beacon state if needed. * * For now it simply copies the global flags into the per-vap * flags and schedules the callback. Later this will support * both global and per-VAP flags, especially useful for * and STA+STA multi-channel operation (eg p2p). */ void ieee80211_vap_update_preamble(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; /* XXX lock? */ IEEE80211_DPRINTF(vap, IEEE80211_MSG_DEBUG, "%s: called\n", __func__); /* schedule the deferred slot flag update and update */ ieee80211_runtask(ic, &vap->iv_preamble_task); } /* * Update the VAP 11g protection mode and update beacon state * if needed. */ void ieee80211_vap_update_erp_protmode(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; /* XXX lock? */ IEEE80211_DPRINTF(vap, IEEE80211_MSG_DEBUG, "%s: called\n", __func__); /* schedule the deferred slot flag update and update */ ieee80211_runtask(ic, &vap->iv_erp_protmode_task); } /* * Update the VAP 11n protection mode and update beacon state * if needed. */ void ieee80211_vap_update_ht_protmode(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; /* XXX lock? */ IEEE80211_DPRINTF(vap, IEEE80211_MSG_DEBUG, "%s: called\n", __func__); /* schedule the deferred protmode update */ ieee80211_runtask(ic, &vap->iv_ht_protmode_task); } /* * Check if the specified rate set supports ERP. * NB: the rate set is assumed to be sorted. */ int ieee80211_iserp_rateset(const struct ieee80211_rateset *rs) { static const int rates[] = { 2, 4, 11, 22, 12, 24, 48 }; int i, j; if (rs->rs_nrates < nitems(rates)) return 0; for (i = 0; i < nitems(rates); i++) { for (j = 0; j < rs->rs_nrates; j++) { int r = rs->rs_rates[j] & IEEE80211_RATE_VAL; if (rates[i] == r) goto next; if (r > rates[i]) return 0; } return 0; next: ; } return 1; } /* * Mark the basic rates for the rate table based on the * operating mode. For real 11g we mark all the 11b rates * and 6, 12, and 24 OFDM. For 11b compatibility we mark only * 11b rates. There's also a pseudo 11a-mode used to mark only * the basic OFDM rates. */ static void setbasicrates(struct ieee80211_rateset *rs, enum ieee80211_phymode mode, int add) { static const struct ieee80211_rateset basic[IEEE80211_MODE_MAX] = { [IEEE80211_MODE_11A] = { 3, { 12, 24, 48 } }, [IEEE80211_MODE_11B] = { 2, { 2, 4 } }, /* NB: mixed b/g */ [IEEE80211_MODE_11G] = { 4, { 2, 4, 11, 22 } }, [IEEE80211_MODE_TURBO_A] = { 3, { 12, 24, 48 } }, [IEEE80211_MODE_TURBO_G] = { 4, { 2, 4, 11, 22 } }, [IEEE80211_MODE_STURBO_A] = { 3, { 12, 24, 48 } }, [IEEE80211_MODE_HALF] = { 3, { 6, 12, 24 } }, [IEEE80211_MODE_QUARTER] = { 3, { 3, 6, 12 } }, [IEEE80211_MODE_11NA] = { 3, { 12, 24, 48 } }, /* NB: mixed b/g */ [IEEE80211_MODE_11NG] = { 4, { 2, 4, 11, 22 } }, /* NB: mixed b/g */ [IEEE80211_MODE_VHT_2GHZ] = { 4, { 2, 4, 11, 22 } }, [IEEE80211_MODE_VHT_5GHZ] = { 3, { 12, 24, 48 } }, }; int i, j; for (i = 0; i < rs->rs_nrates; i++) { if (!add) rs->rs_rates[i] &= IEEE80211_RATE_VAL; for (j = 0; j < basic[mode].rs_nrates; j++) if (basic[mode].rs_rates[j] == rs->rs_rates[i]) { rs->rs_rates[i] |= IEEE80211_RATE_BASIC; break; } } } /* * Set the basic rates in a rate set. */ void ieee80211_setbasicrates(struct ieee80211_rateset *rs, enum ieee80211_phymode mode) { setbasicrates(rs, mode, 0); } /* * Add basic rates to a rate set. */ void ieee80211_addbasicrates(struct ieee80211_rateset *rs, enum ieee80211_phymode mode) { setbasicrates(rs, mode, 1); } /* * WME protocol support. * * The default 11a/b/g/n parameters come from the WiFi Alliance WMM * System Interopability Test Plan (v1.4, Appendix F) and the 802.11n * Draft 2.0 Test Plan (Appendix D). * * Static/Dynamic Turbo mode settings come from Atheros. */ typedef struct phyParamType { uint8_t aifsn; uint8_t logcwmin; uint8_t logcwmax; uint16_t txopLimit; uint8_t acm; } paramType; static const struct phyParamType phyParamForAC_BE[IEEE80211_MODE_MAX] = { [IEEE80211_MODE_AUTO] = { 3, 4, 6, 0, 0 }, [IEEE80211_MODE_11A] = { 3, 4, 6, 0, 0 }, [IEEE80211_MODE_11B] = { 3, 4, 6, 0, 0 }, [IEEE80211_MODE_11G] = { 3, 4, 6, 0, 0 }, [IEEE80211_MODE_FH] = { 3, 4, 6, 0, 0 }, [IEEE80211_MODE_TURBO_A]= { 2, 3, 5, 0, 0 }, [IEEE80211_MODE_TURBO_G]= { 2, 3, 5, 0, 0 }, [IEEE80211_MODE_STURBO_A]={ 2, 3, 5, 0, 0 }, [IEEE80211_MODE_HALF] = { 3, 4, 6, 0, 0 }, [IEEE80211_MODE_QUARTER]= { 3, 4, 6, 0, 0 }, [IEEE80211_MODE_11NA] = { 3, 4, 6, 0, 0 }, [IEEE80211_MODE_11NG] = { 3, 4, 6, 0, 0 }, [IEEE80211_MODE_VHT_2GHZ] = { 3, 4, 6, 0, 0 }, [IEEE80211_MODE_VHT_5GHZ] = { 3, 4, 6, 0, 0 }, }; static const struct phyParamType phyParamForAC_BK[IEEE80211_MODE_MAX] = { [IEEE80211_MODE_AUTO] = { 7, 4, 10, 0, 0 }, [IEEE80211_MODE_11A] = { 7, 4, 10, 0, 0 }, [IEEE80211_MODE_11B] = { 7, 4, 10, 0, 0 }, [IEEE80211_MODE_11G] = { 7, 4, 10, 0, 0 }, [IEEE80211_MODE_FH] = { 7, 4, 10, 0, 0 }, [IEEE80211_MODE_TURBO_A]= { 7, 3, 10, 0, 0 }, [IEEE80211_MODE_TURBO_G]= { 7, 3, 10, 0, 0 }, [IEEE80211_MODE_STURBO_A]={ 7, 3, 10, 0, 0 }, [IEEE80211_MODE_HALF] = { 7, 4, 10, 0, 0 }, [IEEE80211_MODE_QUARTER]= { 7, 4, 10, 0, 0 }, [IEEE80211_MODE_11NA] = { 7, 4, 10, 0, 0 }, [IEEE80211_MODE_11NG] = { 7, 4, 10, 0, 0 }, [IEEE80211_MODE_VHT_2GHZ] = { 7, 4, 10, 0, 0 }, [IEEE80211_MODE_VHT_5GHZ] = { 7, 4, 10, 0, 0 }, }; static const struct phyParamType phyParamForAC_VI[IEEE80211_MODE_MAX] = { [IEEE80211_MODE_AUTO] = { 1, 3, 4, 94, 0 }, [IEEE80211_MODE_11A] = { 1, 3, 4, 94, 0 }, [IEEE80211_MODE_11B] = { 1, 3, 4, 188, 0 }, [IEEE80211_MODE_11G] = { 1, 3, 4, 94, 0 }, [IEEE80211_MODE_FH] = { 1, 3, 4, 188, 0 }, [IEEE80211_MODE_TURBO_A]= { 1, 2, 3, 94, 0 }, [IEEE80211_MODE_TURBO_G]= { 1, 2, 3, 94, 0 }, [IEEE80211_MODE_STURBO_A]={ 1, 2, 3, 94, 0 }, [IEEE80211_MODE_HALF] = { 1, 3, 4, 94, 0 }, [IEEE80211_MODE_QUARTER]= { 1, 3, 4, 94, 0 }, [IEEE80211_MODE_11NA] = { 1, 3, 4, 94, 0 }, [IEEE80211_MODE_11NG] = { 1, 3, 4, 94, 0 }, [IEEE80211_MODE_VHT_2GHZ] = { 1, 3, 4, 94, 0 }, [IEEE80211_MODE_VHT_5GHZ] = { 1, 3, 4, 94, 0 }, }; static const struct phyParamType phyParamForAC_VO[IEEE80211_MODE_MAX] = { [IEEE80211_MODE_AUTO] = { 1, 2, 3, 47, 0 }, [IEEE80211_MODE_11A] = { 1, 2, 3, 47, 0 }, [IEEE80211_MODE_11B] = { 1, 2, 3, 102, 0 }, [IEEE80211_MODE_11G] = { 1, 2, 3, 47, 0 }, [IEEE80211_MODE_FH] = { 1, 2, 3, 102, 0 }, [IEEE80211_MODE_TURBO_A]= { 1, 2, 2, 47, 0 }, [IEEE80211_MODE_TURBO_G]= { 1, 2, 2, 47, 0 }, [IEEE80211_MODE_STURBO_A]={ 1, 2, 2, 47, 0 }, [IEEE80211_MODE_HALF] = { 1, 2, 3, 47, 0 }, [IEEE80211_MODE_QUARTER]= { 1, 2, 3, 47, 0 }, [IEEE80211_MODE_11NA] = { 1, 2, 3, 47, 0 }, [IEEE80211_MODE_11NG] = { 1, 2, 3, 47, 0 }, [IEEE80211_MODE_VHT_2GHZ] = { 1, 2, 3, 47, 0 }, [IEEE80211_MODE_VHT_5GHZ] = { 1, 2, 3, 47, 0 }, }; static const struct phyParamType bssPhyParamForAC_BE[IEEE80211_MODE_MAX] = { [IEEE80211_MODE_AUTO] = { 3, 4, 10, 0, 0 }, [IEEE80211_MODE_11A] = { 3, 4, 10, 0, 0 }, [IEEE80211_MODE_11B] = { 3, 4, 10, 0, 0 }, [IEEE80211_MODE_11G] = { 3, 4, 10, 0, 0 }, [IEEE80211_MODE_FH] = { 3, 4, 10, 0, 0 }, [IEEE80211_MODE_TURBO_A]= { 2, 3, 10, 0, 0 }, [IEEE80211_MODE_TURBO_G]= { 2, 3, 10, 0, 0 }, [IEEE80211_MODE_STURBO_A]={ 2, 3, 10, 0, 0 }, [IEEE80211_MODE_HALF] = { 3, 4, 10, 0, 0 }, [IEEE80211_MODE_QUARTER]= { 3, 4, 10, 0, 0 }, [IEEE80211_MODE_11NA] = { 3, 4, 10, 0, 0 }, [IEEE80211_MODE_11NG] = { 3, 4, 10, 0, 0 }, }; static const struct phyParamType bssPhyParamForAC_VI[IEEE80211_MODE_MAX] = { [IEEE80211_MODE_AUTO] = { 2, 3, 4, 94, 0 }, [IEEE80211_MODE_11A] = { 2, 3, 4, 94, 0 }, [IEEE80211_MODE_11B] = { 2, 3, 4, 188, 0 }, [IEEE80211_MODE_11G] = { 2, 3, 4, 94, 0 }, [IEEE80211_MODE_FH] = { 2, 3, 4, 188, 0 }, [IEEE80211_MODE_TURBO_A]= { 2, 2, 3, 94, 0 }, [IEEE80211_MODE_TURBO_G]= { 2, 2, 3, 94, 0 }, [IEEE80211_MODE_STURBO_A]={ 2, 2, 3, 94, 0 }, [IEEE80211_MODE_HALF] = { 2, 3, 4, 94, 0 }, [IEEE80211_MODE_QUARTER]= { 2, 3, 4, 94, 0 }, [IEEE80211_MODE_11NA] = { 2, 3, 4, 94, 0 }, [IEEE80211_MODE_11NG] = { 2, 3, 4, 94, 0 }, }; static const struct phyParamType bssPhyParamForAC_VO[IEEE80211_MODE_MAX] = { [IEEE80211_MODE_AUTO] = { 2, 2, 3, 47, 0 }, [IEEE80211_MODE_11A] = { 2, 2, 3, 47, 0 }, [IEEE80211_MODE_11B] = { 2, 2, 3, 102, 0 }, [IEEE80211_MODE_11G] = { 2, 2, 3, 47, 0 }, [IEEE80211_MODE_FH] = { 2, 2, 3, 102, 0 }, [IEEE80211_MODE_TURBO_A]= { 1, 2, 2, 47, 0 }, [IEEE80211_MODE_TURBO_G]= { 1, 2, 2, 47, 0 }, [IEEE80211_MODE_STURBO_A]={ 1, 2, 2, 47, 0 }, [IEEE80211_MODE_HALF] = { 2, 2, 3, 47, 0 }, [IEEE80211_MODE_QUARTER]= { 2, 2, 3, 47, 0 }, [IEEE80211_MODE_11NA] = { 2, 2, 3, 47, 0 }, [IEEE80211_MODE_11NG] = { 2, 2, 3, 47, 0 }, }; static void _setifsparams(struct wmeParams *wmep, const paramType *phy) { wmep->wmep_aifsn = phy->aifsn; wmep->wmep_logcwmin = phy->logcwmin; wmep->wmep_logcwmax = phy->logcwmax; wmep->wmep_txopLimit = phy->txopLimit; } static void setwmeparams(struct ieee80211vap *vap, const char *type, int ac, struct wmeParams *wmep, const paramType *phy) { wmep->wmep_acm = phy->acm; _setifsparams(wmep, phy); IEEE80211_DPRINTF(vap, IEEE80211_MSG_WME, "set %s (%s) [acm %u aifsn %u logcwmin %u logcwmax %u txop %u]\n", ieee80211_wme_acnames[ac], type, wmep->wmep_acm, wmep->wmep_aifsn, wmep->wmep_logcwmin, wmep->wmep_logcwmax, wmep->wmep_txopLimit); } static void ieee80211_wme_initparams_locked(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; struct ieee80211_wme_state *wme = &ic->ic_wme; const paramType *pPhyParam, *pBssPhyParam; struct wmeParams *wmep; enum ieee80211_phymode mode; int i; IEEE80211_LOCK_ASSERT(ic); if ((ic->ic_caps & IEEE80211_C_WME) == 0 || ic->ic_nrunning > 1) return; /* * Clear the wme cap_info field so a qoscount from a previous * vap doesn't confuse later code which only parses the beacon * field and updates hardware when said field changes. * Otherwise the hardware is programmed with defaults, not what * the beacon actually announces. * * Note that we can't ever have 0xff as an actual value; * the only valid values are 0..15. */ wme->wme_wmeChanParams.cap_info = 0xfe; /* * Select mode; we can be called early in which case we * always use auto mode. We know we'll be called when * entering the RUN state with bsschan setup properly * so state will eventually get set correctly */ if (ic->ic_bsschan != IEEE80211_CHAN_ANYC) mode = ieee80211_chan2mode(ic->ic_bsschan); else mode = IEEE80211_MODE_AUTO; for (i = 0; i < WME_NUM_AC; i++) { switch (i) { case WME_AC_BK: pPhyParam = &phyParamForAC_BK[mode]; pBssPhyParam = &phyParamForAC_BK[mode]; break; case WME_AC_VI: pPhyParam = &phyParamForAC_VI[mode]; pBssPhyParam = &bssPhyParamForAC_VI[mode]; break; case WME_AC_VO: pPhyParam = &phyParamForAC_VO[mode]; pBssPhyParam = &bssPhyParamForAC_VO[mode]; break; case WME_AC_BE: default: pPhyParam = &phyParamForAC_BE[mode]; pBssPhyParam = &bssPhyParamForAC_BE[mode]; break; } wmep = &wme->wme_wmeChanParams.cap_wmeParams[i]; if (ic->ic_opmode == IEEE80211_M_HOSTAP) { setwmeparams(vap, "chan", i, wmep, pPhyParam); } else { setwmeparams(vap, "chan", i, wmep, pBssPhyParam); } wmep = &wme->wme_wmeBssChanParams.cap_wmeParams[i]; setwmeparams(vap, "bss ", i, wmep, pBssPhyParam); } /* NB: check ic_bss to avoid NULL deref on initial attach */ if (vap->iv_bss != NULL) { /* * Calculate aggressive mode switching threshold based * on beacon interval. This doesn't need locking since * we're only called before entering the RUN state at * which point we start sending beacon frames. */ wme->wme_hipri_switch_thresh = (HIGH_PRI_SWITCH_THRESH * vap->iv_bss->ni_intval) / 100; wme->wme_flags &= ~WME_F_AGGRMODE; ieee80211_wme_updateparams(vap); } } void ieee80211_wme_initparams(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; IEEE80211_LOCK(ic); ieee80211_wme_initparams_locked(vap); IEEE80211_UNLOCK(ic); } /* * Update WME parameters for ourself and the BSS. */ void ieee80211_wme_updateparams_locked(struct ieee80211vap *vap) { static const paramType aggrParam[IEEE80211_MODE_MAX] = { [IEEE80211_MODE_AUTO] = { 2, 4, 10, 64, 0 }, [IEEE80211_MODE_11A] = { 2, 4, 10, 64, 0 }, [IEEE80211_MODE_11B] = { 2, 5, 10, 64, 0 }, [IEEE80211_MODE_11G] = { 2, 4, 10, 64, 0 }, [IEEE80211_MODE_FH] = { 2, 5, 10, 64, 0 }, [IEEE80211_MODE_TURBO_A] = { 1, 3, 10, 64, 0 }, [IEEE80211_MODE_TURBO_G] = { 1, 3, 10, 64, 0 }, [IEEE80211_MODE_STURBO_A] = { 1, 3, 10, 64, 0 }, [IEEE80211_MODE_HALF] = { 2, 4, 10, 64, 0 }, [IEEE80211_MODE_QUARTER] = { 2, 4, 10, 64, 0 }, [IEEE80211_MODE_11NA] = { 2, 4, 10, 64, 0 }, /* XXXcheck*/ [IEEE80211_MODE_11NG] = { 2, 4, 10, 64, 0 }, /* XXXcheck*/ [IEEE80211_MODE_VHT_2GHZ] = { 2, 4, 10, 64, 0 }, /* XXXcheck*/ [IEEE80211_MODE_VHT_5GHZ] = { 2, 4, 10, 64, 0 }, /* XXXcheck*/ }; struct ieee80211com *ic = vap->iv_ic; struct ieee80211_wme_state *wme = &ic->ic_wme; const struct wmeParams *wmep; struct wmeParams *chanp, *bssp; enum ieee80211_phymode mode; int i; int do_aggrmode = 0; /* * Set up the channel access parameters for the physical * device. First populate the configured settings. */ for (i = 0; i < WME_NUM_AC; i++) { chanp = &wme->wme_chanParams.cap_wmeParams[i]; wmep = &wme->wme_wmeChanParams.cap_wmeParams[i]; chanp->wmep_aifsn = wmep->wmep_aifsn; chanp->wmep_logcwmin = wmep->wmep_logcwmin; chanp->wmep_logcwmax = wmep->wmep_logcwmax; chanp->wmep_txopLimit = wmep->wmep_txopLimit; chanp = &wme->wme_bssChanParams.cap_wmeParams[i]; wmep = &wme->wme_wmeBssChanParams.cap_wmeParams[i]; chanp->wmep_aifsn = wmep->wmep_aifsn; chanp->wmep_logcwmin = wmep->wmep_logcwmin; chanp->wmep_logcwmax = wmep->wmep_logcwmax; chanp->wmep_txopLimit = wmep->wmep_txopLimit; } /* * Select mode; we can be called early in which case we * always use auto mode. We know we'll be called when * entering the RUN state with bsschan setup properly * so state will eventually get set correctly */ if (ic->ic_bsschan != IEEE80211_CHAN_ANYC) mode = ieee80211_chan2mode(ic->ic_bsschan); else mode = IEEE80211_MODE_AUTO; /* * This implements aggressive mode as found in certain * vendors' AP's. When there is significant high * priority (VI/VO) traffic in the BSS throttle back BE * traffic by using conservative parameters. Otherwise * BE uses aggressive params to optimize performance of * legacy/non-QoS traffic. */ /* Hostap? Only if aggressive mode is enabled */ if (vap->iv_opmode == IEEE80211_M_HOSTAP && (wme->wme_flags & WME_F_AGGRMODE) != 0) do_aggrmode = 1; /* * Station? Only if we're in a non-QoS BSS. */ else if ((vap->iv_opmode == IEEE80211_M_STA && (vap->iv_bss->ni_flags & IEEE80211_NODE_QOS) == 0)) do_aggrmode = 1; /* * IBSS? Only if we we have WME enabled. */ else if ((vap->iv_opmode == IEEE80211_M_IBSS) && (vap->iv_flags & IEEE80211_F_WME)) do_aggrmode = 1; /* * If WME is disabled on this VAP, default to aggressive mode * regardless of the configuration. */ if ((vap->iv_flags & IEEE80211_F_WME) == 0) do_aggrmode = 1; /* XXX WDS? */ /* XXX MBSS? */ if (do_aggrmode) { chanp = &wme->wme_chanParams.cap_wmeParams[WME_AC_BE]; bssp = &wme->wme_bssChanParams.cap_wmeParams[WME_AC_BE]; chanp->wmep_aifsn = bssp->wmep_aifsn = aggrParam[mode].aifsn; chanp->wmep_logcwmin = bssp->wmep_logcwmin = aggrParam[mode].logcwmin; chanp->wmep_logcwmax = bssp->wmep_logcwmax = aggrParam[mode].logcwmax; chanp->wmep_txopLimit = bssp->wmep_txopLimit = (vap->iv_flags & IEEE80211_F_BURST) ? aggrParam[mode].txopLimit : 0; IEEE80211_DPRINTF(vap, IEEE80211_MSG_WME, "update %s (chan+bss) [acm %u aifsn %u logcwmin %u " "logcwmax %u txop %u]\n", ieee80211_wme_acnames[WME_AC_BE], chanp->wmep_acm, chanp->wmep_aifsn, chanp->wmep_logcwmin, chanp->wmep_logcwmax, chanp->wmep_txopLimit); } /* * Change the contention window based on the number of associated * stations. If the number of associated stations is 1 and * aggressive mode is enabled, lower the contention window even * further. */ if (vap->iv_opmode == IEEE80211_M_HOSTAP && vap->iv_sta_assoc < 2 && (wme->wme_flags & WME_F_AGGRMODE) != 0) { static const uint8_t logCwMin[IEEE80211_MODE_MAX] = { [IEEE80211_MODE_AUTO] = 3, [IEEE80211_MODE_11A] = 3, [IEEE80211_MODE_11B] = 4, [IEEE80211_MODE_11G] = 3, [IEEE80211_MODE_FH] = 4, [IEEE80211_MODE_TURBO_A] = 3, [IEEE80211_MODE_TURBO_G] = 3, [IEEE80211_MODE_STURBO_A] = 3, [IEEE80211_MODE_HALF] = 3, [IEEE80211_MODE_QUARTER] = 3, [IEEE80211_MODE_11NA] = 3, [IEEE80211_MODE_11NG] = 3, [IEEE80211_MODE_VHT_2GHZ] = 3, [IEEE80211_MODE_VHT_5GHZ] = 3, }; chanp = &wme->wme_chanParams.cap_wmeParams[WME_AC_BE]; bssp = &wme->wme_bssChanParams.cap_wmeParams[WME_AC_BE]; chanp->wmep_logcwmin = bssp->wmep_logcwmin = logCwMin[mode]; IEEE80211_DPRINTF(vap, IEEE80211_MSG_WME, "update %s (chan+bss) logcwmin %u\n", ieee80211_wme_acnames[WME_AC_BE], chanp->wmep_logcwmin); } /* schedule the deferred WME update */ ieee80211_runtask(ic, &vap->iv_wme_task); IEEE80211_DPRINTF(vap, IEEE80211_MSG_WME, "%s: WME params updated, cap_info 0x%x\n", __func__, vap->iv_opmode == IEEE80211_M_STA ? wme->wme_wmeChanParams.cap_info : wme->wme_bssChanParams.cap_info); } void ieee80211_wme_updateparams(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; if (ic->ic_caps & IEEE80211_C_WME) { IEEE80211_LOCK(ic); ieee80211_wme_updateparams_locked(vap); IEEE80211_UNLOCK(ic); } } /* * Fetch the WME parameters for the given VAP. * * When net80211 grows p2p, etc support, this may return different * parameters for each VAP. */ void ieee80211_wme_vap_getparams(struct ieee80211vap *vap, struct chanAccParams *wp) { memcpy(wp, &vap->iv_ic->ic_wme.wme_chanParams, sizeof(*wp)); } /* * For NICs which only support one set of WME parameters (ie, softmac NICs) * there may be different VAP WME parameters but only one is "active". * This returns the "NIC" WME parameters for the currently active * context. */ void ieee80211_wme_ic_getparams(struct ieee80211com *ic, struct chanAccParams *wp) { memcpy(wp, &ic->ic_wme.wme_chanParams, sizeof(*wp)); } /* * Return whether to use QoS on a given WME queue. * * This is intended to be called from the transmit path of softmac drivers * which are setting NoAck bits in transmit descriptors. * * Ideally this would be set in some transmit field before the packet is * queued to the driver but net80211 isn't quite there yet. */ int ieee80211_wme_vap_ac_is_noack(struct ieee80211vap *vap, int ac) { /* Bounds/sanity check */ if (ac < 0 || ac >= WME_NUM_AC) return (0); /* Again, there's only one global context for now */ return (!! vap->iv_ic->ic_wme.wme_chanParams.cap_wmeParams[ac].wmep_noackPolicy); } static void parent_updown(void *arg, int npending) { struct ieee80211com *ic = arg; ic->ic_parent(ic); } static void update_mcast(void *arg, int npending) { struct ieee80211com *ic = arg; ic->ic_update_mcast(ic); } static void update_promisc(void *arg, int npending) { struct ieee80211com *ic = arg; ic->ic_update_promisc(ic); } static void update_channel(void *arg, int npending) { struct ieee80211com *ic = arg; ic->ic_set_channel(ic); ieee80211_radiotap_chan_change(ic); } static void update_chw(void *arg, int npending) { struct ieee80211com *ic = arg; /* * XXX should we defer the channel width _config_ update until now? */ ic->ic_update_chw(ic); } /* * Deferred WME parameter and beacon update. * * In preparation for per-VAP WME configuration, call the VAP * method if the VAP requires it. Otherwise, just call the * older global method. There isn't a per-VAP WME configuration * just yet so for now just use the global configuration. */ static void vap_update_wme(void *arg, int npending) { struct ieee80211vap *vap = arg; struct ieee80211com *ic = vap->iv_ic; struct ieee80211_wme_state *wme = &ic->ic_wme; /* Driver update */ if (vap->iv_wme_update != NULL) vap->iv_wme_update(vap, ic->ic_wme.wme_chanParams.cap_wmeParams); else ic->ic_wme.wme_update(ic); IEEE80211_LOCK(ic); /* * Arrange for the beacon update. * * XXX what about MBSS, WDS? */ if (vap->iv_opmode == IEEE80211_M_HOSTAP || vap->iv_opmode == IEEE80211_M_IBSS) { /* * Arrange for a beacon update and bump the parameter * set number so associated stations load the new values. */ wme->wme_bssChanParams.cap_info = (wme->wme_bssChanParams.cap_info+1) & WME_QOSINFO_COUNT; ieee80211_beacon_notify(vap, IEEE80211_BEACON_WME); } IEEE80211_UNLOCK(ic); } static void restart_vaps(void *arg, int npending) { struct ieee80211com *ic = arg; ieee80211_suspend_all(ic); ieee80211_resume_all(ic); } /* * Block until the parent is in a known state. This is * used after any operations that dispatch a task (e.g. * to auto-configure the parent device up/down). */ void ieee80211_waitfor_parent(struct ieee80211com *ic) { taskqueue_block(ic->ic_tq); ieee80211_draintask(ic, &ic->ic_parent_task); ieee80211_draintask(ic, &ic->ic_mcast_task); ieee80211_draintask(ic, &ic->ic_promisc_task); ieee80211_draintask(ic, &ic->ic_chan_task); ieee80211_draintask(ic, &ic->ic_bmiss_task); ieee80211_draintask(ic, &ic->ic_chw_task); taskqueue_unblock(ic->ic_tq); } /* * Check to see whether the current channel needs reset. * * Some devices don't handle being given an invalid channel * in their operating mode very well (eg wpi(4) will throw a * firmware exception.) * * Return 0 if we're ok, 1 if the channel needs to be reset. * * See PR kern/202502. */ static int ieee80211_start_check_reset_chan(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; if ((vap->iv_opmode == IEEE80211_M_IBSS && IEEE80211_IS_CHAN_NOADHOC(ic->ic_curchan)) || (vap->iv_opmode == IEEE80211_M_HOSTAP && IEEE80211_IS_CHAN_NOHOSTAP(ic->ic_curchan))) return (1); return (0); } /* * Reset the curchan to a known good state. */ static void ieee80211_start_reset_chan(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; ic->ic_curchan = &ic->ic_channels[0]; } /* * Start a vap running. If this is the first vap to be * set running on the underlying device then we * automatically bring the device up. */ void ieee80211_start_locked(struct ieee80211vap *vap) { struct ifnet *ifp = vap->iv_ifp; struct ieee80211com *ic = vap->iv_ic; IEEE80211_LOCK_ASSERT(ic); IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE | IEEE80211_MSG_DEBUG, "start running, %d vaps running\n", ic->ic_nrunning); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { /* * Mark us running. Note that it's ok to do this first; * if we need to bring the parent device up we defer that * to avoid dropping the com lock. We expect the device * to respond to being marked up by calling back into us * through ieee80211_start_all at which point we'll come * back in here and complete the work. */ ifp->if_drv_flags |= IFF_DRV_RUNNING; ieee80211_notify_ifnet_change(vap); /* * We are not running; if this we are the first vap * to be brought up auto-up the parent if necessary. */ if (ic->ic_nrunning++ == 0) { /* reset the channel to a known good channel */ if (ieee80211_start_check_reset_chan(vap)) ieee80211_start_reset_chan(vap); IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE | IEEE80211_MSG_DEBUG, "%s: up parent %s\n", __func__, ic->ic_name); ieee80211_runtask(ic, &ic->ic_parent_task); return; } } /* * If the parent is up and running, then kick the * 802.11 state machine as appropriate. */ if (vap->iv_roaming != IEEE80211_ROAMING_MANUAL) { if (vap->iv_opmode == IEEE80211_M_STA) { #if 0 /* XXX bypasses scan too easily; disable for now */ /* * Try to be intelligent about clocking the state * machine. If we're currently in RUN state then * we should be able to apply any new state/parameters * simply by re-associating. Otherwise we need to * re-scan to select an appropriate ap. */ if (vap->iv_state >= IEEE80211_S_RUN) ieee80211_new_state_locked(vap, IEEE80211_S_ASSOC, 1); else #endif ieee80211_new_state_locked(vap, IEEE80211_S_SCAN, 0); } else { /* * For monitor+wds mode there's nothing to do but * start running. Otherwise if this is the first * vap to be brought up, start a scan which may be * preempted if the station is locked to a particular * channel. */ vap->iv_flags_ext |= IEEE80211_FEXT_REINIT; if (vap->iv_opmode == IEEE80211_M_MONITOR || vap->iv_opmode == IEEE80211_M_WDS) ieee80211_new_state_locked(vap, IEEE80211_S_RUN, -1); else ieee80211_new_state_locked(vap, IEEE80211_S_SCAN, 0); } } } /* * Start a single vap. */ void ieee80211_init(void *arg) { struct ieee80211vap *vap = arg; IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE | IEEE80211_MSG_DEBUG, "%s\n", __func__); IEEE80211_LOCK(vap->iv_ic); ieee80211_start_locked(vap); IEEE80211_UNLOCK(vap->iv_ic); } /* * Start all runnable vap's on a device. */ void ieee80211_start_all(struct ieee80211com *ic) { struct ieee80211vap *vap; IEEE80211_LOCK(ic); TAILQ_FOREACH(vap, &ic->ic_vaps, iv_next) { struct ifnet *ifp = vap->iv_ifp; if (IFNET_IS_UP_RUNNING(ifp)) /* NB: avoid recursion */ ieee80211_start_locked(vap); } IEEE80211_UNLOCK(ic); } /* * Stop a vap. We force it down using the state machine * then mark it's ifnet not running. If this is the last * vap running on the underlying device then we close it * too to insure it will be properly initialized when the * next vap is brought up. */ void ieee80211_stop_locked(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; struct ifnet *ifp = vap->iv_ifp; IEEE80211_LOCK_ASSERT(ic); IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE | IEEE80211_MSG_DEBUG, "stop running, %d vaps running\n", ic->ic_nrunning); ieee80211_new_state_locked(vap, IEEE80211_S_INIT, -1); if (ifp->if_drv_flags & IFF_DRV_RUNNING) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; /* mark us stopped */ ieee80211_notify_ifnet_change(vap); if (--ic->ic_nrunning == 0) { IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE | IEEE80211_MSG_DEBUG, "down parent %s\n", ic->ic_name); ieee80211_runtask(ic, &ic->ic_parent_task); } } } void ieee80211_stop(struct ieee80211vap *vap) { struct ieee80211com *ic = vap->iv_ic; IEEE80211_LOCK(ic); ieee80211_stop_locked(vap); IEEE80211_UNLOCK(ic); } /* * Stop all vap's running on a device. */ void ieee80211_stop_all(struct ieee80211com *ic) { struct ieee80211vap *vap; IEEE80211_LOCK(ic); TAILQ_FOREACH(vap, &ic->ic_vaps, iv_next) { struct ifnet *ifp = vap->iv_ifp; if (IFNET_IS_UP_RUNNING(ifp)) /* NB: avoid recursion */ ieee80211_stop_locked(vap); } IEEE80211_UNLOCK(ic); ieee80211_waitfor_parent(ic); } /* * Stop all vap's running on a device and arrange * for those that were running to be resumed. */ void ieee80211_suspend_all(struct ieee80211com *ic) { struct ieee80211vap *vap; IEEE80211_LOCK(ic); TAILQ_FOREACH(vap, &ic->ic_vaps, iv_next) { struct ifnet *ifp = vap->iv_ifp; if (IFNET_IS_UP_RUNNING(ifp)) { /* NB: avoid recursion */ vap->iv_flags_ext |= IEEE80211_FEXT_RESUME; ieee80211_stop_locked(vap); } } IEEE80211_UNLOCK(ic); ieee80211_waitfor_parent(ic); } /* * Start all vap's marked for resume. */ void ieee80211_resume_all(struct ieee80211com *ic) { struct ieee80211vap *vap; IEEE80211_LOCK(ic); TAILQ_FOREACH(vap, &ic->ic_vaps, iv_next) { struct ifnet *ifp = vap->iv_ifp; if (!IFNET_IS_UP_RUNNING(ifp) && (vap->iv_flags_ext & IEEE80211_FEXT_RESUME)) { vap->iv_flags_ext &= ~IEEE80211_FEXT_RESUME; ieee80211_start_locked(vap); } } IEEE80211_UNLOCK(ic); } /* * Restart all vap's running on a device. */ void ieee80211_restart_all(struct ieee80211com *ic) { /* * NB: do not use ieee80211_runtask here, we will * block & drain net80211 taskqueue. */ taskqueue_enqueue(taskqueue_thread, &ic->ic_restart_task); } void ieee80211_beacon_miss(struct ieee80211com *ic) { IEEE80211_LOCK(ic); if ((ic->ic_flags & IEEE80211_F_SCAN) == 0) { /* Process in a taskq, the handler may reenter the driver */ ieee80211_runtask(ic, &ic->ic_bmiss_task); } IEEE80211_UNLOCK(ic); } static void beacon_miss(void *arg, int npending) { struct ieee80211com *ic = arg; struct ieee80211vap *vap; IEEE80211_LOCK(ic); TAILQ_FOREACH(vap, &ic->ic_vaps, iv_next) { /* * We only pass events through for sta vap's in RUN+ state; * may be too restrictive but for now this saves all the * handlers duplicating these checks. */ if (vap->iv_opmode == IEEE80211_M_STA && vap->iv_state >= IEEE80211_S_RUN && vap->iv_bmiss != NULL) vap->iv_bmiss(vap); } IEEE80211_UNLOCK(ic); } static void beacon_swmiss(void *arg, int npending) { struct ieee80211vap *vap = arg; struct ieee80211com *ic = vap->iv_ic; IEEE80211_LOCK(ic); if (vap->iv_state >= IEEE80211_S_RUN) { /* XXX Call multiple times if npending > zero? */ vap->iv_bmiss(vap); } IEEE80211_UNLOCK(ic); } /* * Software beacon miss handling. Check if any beacons * were received in the last period. If not post a * beacon miss; otherwise reset the counter. */ void ieee80211_swbmiss(void *arg) { struct ieee80211vap *vap = arg; struct ieee80211com *ic = vap->iv_ic; IEEE80211_LOCK_ASSERT(ic); KASSERT(vap->iv_state >= IEEE80211_S_RUN, ("wrong state %d", vap->iv_state)); if (ic->ic_flags & IEEE80211_F_SCAN) { /* * If scanning just ignore and reset state. If we get a * bmiss after coming out of scan because we haven't had * time to receive a beacon then we should probe the AP * before posting a real bmiss (unless iv_bmiss_max has * been artifiically lowered). A cleaner solution might * be to disable the timer on scan start/end but to handle * case of multiple sta vap's we'd need to disable the * timers of all affected vap's. */ vap->iv_swbmiss_count = 0; } else if (vap->iv_swbmiss_count == 0) { if (vap->iv_bmiss != NULL) ieee80211_runtask(ic, &vap->iv_swbmiss_task); } else vap->iv_swbmiss_count = 0; callout_reset(&vap->iv_swbmiss, vap->iv_swbmiss_period, ieee80211_swbmiss, vap); } /* * Start an 802.11h channel switch. We record the parameters, * mark the operation pending, notify each vap through the * beacon update mechanism so it can update the beacon frame * contents, and then switch vap's to CSA state to block outbound * traffic. Devices that handle CSA directly can use the state * switch to do the right thing so long as they call * ieee80211_csa_completeswitch when it's time to complete the * channel change. Devices that depend on the net80211 layer can * use ieee80211_beacon_update to handle the countdown and the * channel switch. */ void ieee80211_csa_startswitch(struct ieee80211com *ic, struct ieee80211_channel *c, int mode, int count) { struct ieee80211vap *vap; IEEE80211_LOCK_ASSERT(ic); ic->ic_csa_newchan = c; ic->ic_csa_mode = mode; ic->ic_csa_count = count; ic->ic_flags |= IEEE80211_F_CSAPENDING; TAILQ_FOREACH(vap, &ic->ic_vaps, iv_next) { if (vap->iv_opmode == IEEE80211_M_HOSTAP || vap->iv_opmode == IEEE80211_M_IBSS || vap->iv_opmode == IEEE80211_M_MBSS) ieee80211_beacon_notify(vap, IEEE80211_BEACON_CSA); /* switch to CSA state to block outbound traffic */ if (vap->iv_state == IEEE80211_S_RUN) ieee80211_new_state_locked(vap, IEEE80211_S_CSA, 0); } ieee80211_notify_csa(ic, c, mode, count); } /* * Complete the channel switch by transitioning all CSA VAPs to RUN. * This is called by both the completion and cancellation functions * so each VAP is placed back in the RUN state and can thus transmit. */ static void csa_completeswitch(struct ieee80211com *ic) { struct ieee80211vap *vap; ic->ic_csa_newchan = NULL; ic->ic_flags &= ~IEEE80211_F_CSAPENDING; TAILQ_FOREACH(vap, &ic->ic_vaps, iv_next) if (vap->iv_state == IEEE80211_S_CSA) ieee80211_new_state_locked(vap, IEEE80211_S_RUN, 0); } /* * Complete an 802.11h channel switch started by ieee80211_csa_startswitch. * We clear state and move all vap's in CSA state to RUN state * so they can again transmit. * * Although this may not be completely correct, update the BSS channel * for each VAP to the newly configured channel. The setcurchan sets * the current operating channel for the interface (so the radio does * switch over) but the VAP BSS isn't updated, leading to incorrectly * reported information via ioctl. */ void ieee80211_csa_completeswitch(struct ieee80211com *ic) { struct ieee80211vap *vap; IEEE80211_LOCK_ASSERT(ic); KASSERT(ic->ic_flags & IEEE80211_F_CSAPENDING, ("csa not pending")); ieee80211_setcurchan(ic, ic->ic_csa_newchan); TAILQ_FOREACH(vap, &ic->ic_vaps, iv_next) if (vap->iv_state == IEEE80211_S_CSA) vap->iv_bss->ni_chan = ic->ic_curchan; csa_completeswitch(ic); } /* * Cancel an 802.11h channel switch started by ieee80211_csa_startswitch. * We clear state and move all vap's in CSA state to RUN state * so they can again transmit. */ void ieee80211_csa_cancelswitch(struct ieee80211com *ic) { IEEE80211_LOCK_ASSERT(ic); csa_completeswitch(ic); } /* * Complete a DFS CAC started by ieee80211_dfs_cac_start. * We clear state and move all vap's in CAC state to RUN state. */ void ieee80211_cac_completeswitch(struct ieee80211vap *vap0) { struct ieee80211com *ic = vap0->iv_ic; struct ieee80211vap *vap; IEEE80211_LOCK(ic); /* * Complete CAC state change for lead vap first; then * clock all the other vap's waiting. */ KASSERT(vap0->iv_state == IEEE80211_S_CAC, ("wrong state %d", vap0->iv_state)); ieee80211_new_state_locked(vap0, IEEE80211_S_RUN, 0); TAILQ_FOREACH(vap, &ic->ic_vaps, iv_next) if (vap->iv_state == IEEE80211_S_CAC && vap != vap0) ieee80211_new_state_locked(vap, IEEE80211_S_RUN, 0); IEEE80211_UNLOCK(ic); } /* * Force all vap's other than the specified vap to the INIT state * and mark them as waiting for a scan to complete. These vaps * will be brought up when the scan completes and the scanning vap * reaches RUN state by wakeupwaiting. */ static void markwaiting(struct ieee80211vap *vap0) { struct ieee80211com *ic = vap0->iv_ic; struct ieee80211vap *vap; IEEE80211_LOCK_ASSERT(ic); /* * A vap list entry can not disappear since we are running on the * taskqueue and a vap destroy will queue and drain another state * change task. */ TAILQ_FOREACH(vap, &ic->ic_vaps, iv_next) { if (vap == vap0) continue; if (vap->iv_state != IEEE80211_S_INIT) { /* NB: iv_newstate may drop the lock */ vap->iv_newstate(vap, IEEE80211_S_INIT, 0); IEEE80211_LOCK_ASSERT(ic); vap->iv_flags_ext |= IEEE80211_FEXT_SCANWAIT; } } } /* * Wakeup all vap's waiting for a scan to complete. This is the * companion to markwaiting (above) and is used to coordinate * multiple vaps scanning. * This is called from the state taskqueue. */ static void wakeupwaiting(struct ieee80211vap *vap0) { struct ieee80211com *ic = vap0->iv_ic; struct ieee80211vap *vap; IEEE80211_LOCK_ASSERT(ic); /* * A vap list entry can not disappear since we are running on the * taskqueue and a vap destroy will queue and drain another state * change task. */ TAILQ_FOREACH(vap, &ic->ic_vaps, iv_next) { if (vap == vap0) continue; if (vap->iv_flags_ext & IEEE80211_FEXT_SCANWAIT) { vap->iv_flags_ext &= ~IEEE80211_FEXT_SCANWAIT; /* NB: sta's cannot go INIT->RUN */ /* NB: iv_newstate may drop the lock */ /* * This is problematic if the interface has OACTIVE * set. Only the deferred ieee80211_newstate_cb() * will end up actually /clearing/ the OACTIVE * flag on a state transition to RUN from a non-RUN * state. * * But, we're not actually deferring this callback; * and when the deferred call occurs it shows up as * a RUN->RUN transition! So the flag isn't/wasn't * cleared! * * I'm also not sure if it's correct to actually * do the transitions here fully through the deferred * paths either as other things can be invoked as * part of that state machine. * * So just keep this in mind when looking at what * the markwaiting/wakeupwaiting routines are doing * and how they invoke vap state changes. */ vap->iv_newstate(vap, vap->iv_opmode == IEEE80211_M_STA ? IEEE80211_S_SCAN : IEEE80211_S_RUN, 0); IEEE80211_LOCK_ASSERT(ic); } } } /* * Handle post state change work common to all operating modes. */ static void ieee80211_newstate_cb(void *xvap, int npending) { struct ieee80211vap *vap = xvap; struct ieee80211com *ic = vap->iv_ic; enum ieee80211_state nstate, ostate; int arg, rc; IEEE80211_LOCK(ic); nstate = vap->iv_nstate; arg = vap->iv_nstate_arg; if (vap->iv_flags_ext & IEEE80211_FEXT_REINIT) { /* * We have been requested to drop back to the INIT before * proceeding to the new state. */ /* Deny any state changes while we are here. */ vap->iv_nstate = IEEE80211_S_INIT; IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE, "%s: %s -> %s arg %d\n", __func__, ieee80211_state_name[vap->iv_state], ieee80211_state_name[vap->iv_nstate], arg); vap->iv_newstate(vap, vap->iv_nstate, 0); IEEE80211_LOCK_ASSERT(ic); vap->iv_flags_ext &= ~(IEEE80211_FEXT_REINIT | IEEE80211_FEXT_STATEWAIT); /* enqueue new state transition after cancel_scan() task */ ieee80211_new_state_locked(vap, nstate, arg); goto done; } ostate = vap->iv_state; if (nstate == IEEE80211_S_SCAN && ostate != IEEE80211_S_INIT) { /* * SCAN was forced; e.g. on beacon miss. Force other running * vap's to INIT state and mark them as waiting for the scan to * complete. This insures they don't interfere with our * scanning. Since we are single threaded the vaps can not * transition again while we are executing. * * XXX not always right, assumes ap follows sta */ markwaiting(vap); } IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE, "%s: %s -> %s arg %d\n", __func__, ieee80211_state_name[ostate], ieee80211_state_name[nstate], arg); rc = vap->iv_newstate(vap, nstate, arg); IEEE80211_LOCK_ASSERT(ic); vap->iv_flags_ext &= ~IEEE80211_FEXT_STATEWAIT; if (rc != 0) { /* State transition failed */ KASSERT(rc != EINPROGRESS, ("iv_newstate was deferred")); KASSERT(nstate != IEEE80211_S_INIT, ("INIT state change failed")); IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE, "%s: %s returned error %d\n", __func__, ieee80211_state_name[nstate], rc); goto done; } /* * Handle the case of a RUN->RUN transition occuring when STA + AP * VAPs occur on the same radio. * * The mark and wakeup waiting routines call iv_newstate() directly, * but they do not end up deferring state changes here. * Thus, although the VAP newstate method sees a transition * of RUN->INIT->RUN, the deferred path here only sees a RUN->RUN * transition. If OACTIVE is set then it is never cleared. * * So, if we're here and the state is RUN, just clear OACTIVE. * At some point if the markwaiting/wakeupwaiting paths end up * also invoking the deferred state updates then this will * be no-op code - and also if OACTIVE is finally retired, it'll * also be no-op code. */ if (nstate == IEEE80211_S_RUN) { /* * OACTIVE may be set on the vap if the upper layer * tried to transmit (e.g. IPv6 NDP) before we reach * RUN state. Clear it and restart xmit. * * Note this can also happen as a result of SLEEP->RUN * (i.e. coming out of power save mode). * * Historically this was done only for a state change * but is needed earlier; see next comment. The 2nd half * of the work is still only done in case of an actual * state change below. */ /* * Unblock the VAP queue; a RUN->RUN state can happen * on a STA+AP setup on the AP vap. See wakeupwaiting(). */ vap->iv_ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; /* * XXX TODO Kick-start a VAP queue - this should be a method! */ } /* No actual transition, skip post processing */ if (ostate == nstate) goto done; if (nstate == IEEE80211_S_RUN) { /* bring up any vaps waiting on us */ wakeupwaiting(vap); } else if (nstate == IEEE80211_S_INIT) { /* * Flush the scan cache if we did the last scan (XXX?) * and flush any frames on send queues from this vap. * Note the mgt q is used only for legacy drivers and * will go away shortly. */ ieee80211_scan_flush(vap); /* * XXX TODO: ic/vap queue flush */ } done: IEEE80211_UNLOCK(ic); } /* * Public interface for initiating a state machine change. * This routine single-threads the request and coordinates * the scheduling of multiple vaps for the purpose of selecting * an operating channel. Specifically the following scenarios * are handled: * o only one vap can be selecting a channel so on transition to * SCAN state if another vap is already scanning then * mark the caller for later processing and return without * doing anything (XXX? expectations by caller of synchronous operation) * o only one vap can be doing CAC of a channel so on transition to * CAC state if another vap is already scanning for radar then * mark the caller for later processing and return without * doing anything (XXX? expectations by caller of synchronous operation) * o if another vap is already running when a request is made * to SCAN then an operating channel has been chosen; bypass * the scan and just join the channel * * Note that the state change call is done through the iv_newstate * method pointer so any driver routine gets invoked. The driver * will normally call back into operating mode-specific * ieee80211_newstate routines (below) unless it needs to completely * bypass the state machine (e.g. because the firmware has it's * own idea how things should work). Bypassing the net80211 layer * is usually a mistake and indicates lack of proper integration * with the net80211 layer. */ int ieee80211_new_state_locked(struct ieee80211vap *vap, enum ieee80211_state nstate, int arg) { struct ieee80211com *ic = vap->iv_ic; struct ieee80211vap *vp; enum ieee80211_state ostate; int nrunning, nscanning; IEEE80211_LOCK_ASSERT(ic); if (vap->iv_flags_ext & IEEE80211_FEXT_STATEWAIT) { if (vap->iv_nstate == IEEE80211_S_INIT || ((vap->iv_state == IEEE80211_S_INIT || (vap->iv_flags_ext & IEEE80211_FEXT_REINIT)) && vap->iv_nstate == IEEE80211_S_SCAN && nstate > IEEE80211_S_SCAN)) { /* * XXX The vap is being stopped/started, * do not allow any other state changes * until this is completed. */ IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE, "%s: %s -> %s (%s) transition discarded\n", __func__, ieee80211_state_name[vap->iv_state], ieee80211_state_name[nstate], ieee80211_state_name[vap->iv_nstate]); return -1; } else if (vap->iv_state != vap->iv_nstate) { #if 0 /* Warn if the previous state hasn't completed. */ IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE, "%s: pending %s -> %s transition lost\n", __func__, ieee80211_state_name[vap->iv_state], ieee80211_state_name[vap->iv_nstate]); #else /* XXX temporarily enable to identify issues */ if_printf(vap->iv_ifp, "%s: pending %s -> %s transition lost\n", __func__, ieee80211_state_name[vap->iv_state], ieee80211_state_name[vap->iv_nstate]); #endif } } nrunning = nscanning = 0; /* XXX can track this state instead of calculating */ TAILQ_FOREACH(vp, &ic->ic_vaps, iv_next) { if (vp != vap) { if (vp->iv_state >= IEEE80211_S_RUN) nrunning++; /* XXX doesn't handle bg scan */ /* NB: CAC+AUTH+ASSOC treated like SCAN */ else if (vp->iv_state > IEEE80211_S_INIT) nscanning++; } } ostate = vap->iv_state; IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE, "%s: %s -> %s (arg %d) (nrunning %d nscanning %d)\n", __func__, ieee80211_state_name[ostate], ieee80211_state_name[nstate], arg, nrunning, nscanning); switch (nstate) { case IEEE80211_S_SCAN: if (ostate == IEEE80211_S_INIT) { /* * INIT -> SCAN happens on initial bringup. */ KASSERT(!(nscanning && nrunning), ("%d scanning and %d running", nscanning, nrunning)); if (nscanning) { /* * Someone is scanning, defer our state * change until the work has completed. */ IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE, "%s: defer %s -> %s\n", __func__, ieee80211_state_name[ostate], ieee80211_state_name[nstate]); vap->iv_flags_ext |= IEEE80211_FEXT_SCANWAIT; return 0; } if (nrunning) { /* * Someone is operating; just join the channel * they have chosen. */ /* XXX kill arg? */ /* XXX check each opmode, adhoc? */ if (vap->iv_opmode == IEEE80211_M_STA) nstate = IEEE80211_S_SCAN; else nstate = IEEE80211_S_RUN; #ifdef IEEE80211_DEBUG if (nstate != IEEE80211_S_SCAN) { IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE, "%s: override, now %s -> %s\n", __func__, ieee80211_state_name[ostate], ieee80211_state_name[nstate]); } #endif } } break; case IEEE80211_S_RUN: if (vap->iv_opmode == IEEE80211_M_WDS && (vap->iv_flags_ext & IEEE80211_FEXT_WDSLEGACY) && nscanning) { /* * Legacy WDS with someone else scanning; don't * go online until that completes as we should * follow the other vap to the channel they choose. */ IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE, "%s: defer %s -> %s (legacy WDS)\n", __func__, ieee80211_state_name[ostate], ieee80211_state_name[nstate]); vap->iv_flags_ext |= IEEE80211_FEXT_SCANWAIT; return 0; } if (vap->iv_opmode == IEEE80211_M_HOSTAP && IEEE80211_IS_CHAN_DFS(ic->ic_bsschan) && (vap->iv_flags_ext & IEEE80211_FEXT_DFS) && !IEEE80211_IS_CHAN_CACDONE(ic->ic_bsschan)) { /* * This is a DFS channel, transition to CAC state * instead of RUN. This allows us to initiate * Channel Availability Check (CAC) as specified * by 11h/DFS. */ nstate = IEEE80211_S_CAC; IEEE80211_DPRINTF(vap, IEEE80211_MSG_STATE, "%s: override %s -> %s (DFS)\n", __func__, ieee80211_state_name[ostate], ieee80211_state_name[nstate]); } break; case IEEE80211_S_INIT: /* cancel any scan in progress */ ieee80211_cancel_scan(vap); if (ostate == IEEE80211_S_INIT ) { /* XXX don't believe this */ /* INIT -> INIT. nothing to do */ vap->iv_flags_ext &= ~IEEE80211_FEXT_SCANWAIT; } /* fall thru... */ default: break; } /* defer the state change to a thread */ vap->iv_nstate = nstate; vap->iv_nstate_arg = arg; vap->iv_flags_ext |= IEEE80211_FEXT_STATEWAIT; ieee80211_runtask(ic, &vap->iv_nstate_task); return EINPROGRESS; } int ieee80211_new_state(struct ieee80211vap *vap, enum ieee80211_state nstate, int arg) { struct ieee80211com *ic = vap->iv_ic; int rc; IEEE80211_LOCK(ic); rc = ieee80211_new_state_locked(vap, nstate, arg); IEEE80211_UNLOCK(ic); return rc; } diff --git a/sys/netinet/tcp_subr.c b/sys/netinet/tcp_subr.c index 1c04de080623..f2dc1b0b2836 100644 --- a/sys/netinet/tcp_subr.c +++ b/sys/netinet/tcp_subr.c @@ -1,4154 +1,4149 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_kern_tls.h" #include "opt_tcpdebug.h" #include #include #include #include #include #ifdef TCP_HHOOK #include #endif #include #ifdef TCP_HHOOK #include #endif #ifdef KERN_TLS #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #include #include #include #include #endif #include #ifdef INVARIANTS #define TCPSTATES #endif #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #ifdef TCPPCAP #include #endif #ifdef TCPDEBUG #include #endif #ifdef INET6 #include #endif #ifdef TCP_OFFLOAD #include #endif #include #include #include #include #include #include VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; #ifdef INET6 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; #endif #ifdef NETFLIX_EXP_DETECTION /* Sack attack detection thresholds and such */ SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Sack Attack detection thresholds"); int32_t tcp_force_detection = 0; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, CTLFLAG_RW, &tcp_force_detection, 0, "Do we force detection even if the INP has it off?"); int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, CTLFLAG_RW, &tcp_sack_to_ack_thresh, 700, "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, CTLFLAG_RW, &tcp_sack_to_move_thresh, 600, "Percentage of sack moves we must see above (10.1 percent is 101)"); int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, CTLFLAG_RW, &tcp_restoral_thresh, 550, "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); int32_t tcp_sad_decay_val = 800; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, CTLFLAG_RW, &tcp_sad_decay_val, 800, "The decay percentage (10.1 percent equals 101 )"); int32_t tcp_map_minimum = 500; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, CTLFLAG_RW, &tcp_map_minimum, 500, "Number of Map enteries before we start detection"); int32_t tcp_attack_on_turns_on_logging = 0; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, CTLFLAG_RW, &tcp_attack_on_turns_on_logging, 0, "When we have a positive hit on attack, do we turn on logging?"); int32_t tcp_sad_pacing_interval = 2000; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, CTLFLAG_RW, &tcp_sad_pacing_interval, 2000, "What is the minimum pacing interval for a classified attacker?"); int32_t tcp_sad_low_pps = 100; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, CTLFLAG_RW, &tcp_sad_low_pps, 100, "What is the input pps that below which we do not decay?"); #endif uint32_t tcp_ack_war_time_window = 1000; SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow, CTLFLAG_RW, &tcp_ack_war_time_window, 1000, "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?"); uint32_t tcp_ack_war_cnt = 5; SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, CTLFLAG_RW, &tcp_ack_war_cnt, 5, "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?"); struct rwlock tcp_function_lock; static int sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) { int error, new; new = V_tcp_mssdflt; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new < TCP_MINMSS) error = EINVAL; else V_tcp_mssdflt = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", "Default TCP Maximum Segment Size"); #ifdef INET6 static int sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) { int error, new; new = V_tcp_v6mssdflt; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new < TCP_MINMSS) error = EINVAL; else V_tcp_v6mssdflt = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", "Default TCP Maximum Segment Size for IPv6"); #endif /* INET6 */ /* * Minimum MSS we accept and use. This prevents DoS attacks where * we are forced to a ridiculous low MSS like 20 and send hundreds * of packets instead of one. The effect scales with the available * bandwidth and quickly saturates the CPU and network interface * with packet generation and sending. Set to zero to disable MINMSS * checking. This setting prevents us from sending too small packets. */ VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_minmss), 0, "Minimum TCP Maximum Segment Size"); VNET_DEFINE(int, tcp_do_rfc1323) = 1; SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc1323), 0, "Enable rfc1323 (high performance TCP) extensions"); /* * As of June 2021, several TCP stacks violate RFC 7323 from September 2014. * Some stacks negotiate TS, but never send them after connection setup. Some * stacks negotiate TS, but don't send them when sending keep-alive segments. * These include modern widely deployed TCP stacks. * Therefore tolerating violations for now... */ VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_tolerate_missing_ts), 0, "Tolerate missing TCP timestamps"); VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_ts_offset_per_conn), 0, "Initialize TCP timestamps per connection instead of per host pair"); /* How many connections are pacing */ static volatile uint32_t number_of_tcp_connections_pacing = 0; static uint32_t shadow_num_connections = 0; static int tcp_pacing_limit = 10000; SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW, &tcp_pacing_limit, 1000, "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)"); SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD, &shadow_num_connections, 0, "Number of TCP connections being paced"); static int tcp_log_debug = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); static int tcp_tcbhashsize; SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); static int do_tcpdrain = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, "Enable tcp_drain routine for extra help when low on mbufs"); SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; #define V_icmp_may_rst VNET(icmp_may_rst) SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmp_may_rst), 0, "Certain ICMP unreachable messages may abort connections in SYN_SENT"); VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_isn_reseed_interval), 0, "Seconds between reseeding of ISN secret"); static int tcp_soreceive_stream; SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); VNET_DEFINE(uma_zone_t, sack_hole_zone); #define V_sack_hole_zone VNET(sack_hole_zone) VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ static int sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_tcp_map_entries_limit; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { /* only allow "0" and value > minimum */ if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) error = EINVAL; else V_tcp_map_entries_limit = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(tcp_map_entries_limit), 0, &sysctl_net_inet_tcp_map_limit_check, "IU", "Total sendmap entries limit"); VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_map_split_limit), 0, "Total sendmap split entries limit"); #ifdef TCP_HHOOK VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); #endif #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); #define V_ts_offset_secret VNET(ts_offset_secret) static int tcp_default_fb_init(struct tcpcb *tp); static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); static int tcp_default_handoff_ok(struct tcpcb *tp); static struct inpcb *tcp_notify(struct inpcb *, int); static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); static struct inpcb *tcp_mtudisc(struct inpcb *, int); static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, const void *ip6hdr); static struct tcp_function_block tcp_def_funcblk = { .tfb_tcp_block_name = "freebsd", .tfb_tcp_output = tcp_default_output, .tfb_tcp_do_segment = tcp_do_segment, .tfb_tcp_ctloutput = tcp_default_ctloutput, .tfb_tcp_handoff_ok = tcp_default_handoff_ok, .tfb_tcp_fb_init = tcp_default_fb_init, .tfb_tcp_fb_fini = tcp_default_fb_fini, }; static int tcp_fb_cnt = 0; struct tcp_funchead t_functions; static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; void tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp) { TCPSTAT_INC(tcps_dsack_count); tp->t_dsack_pack++; if (tlp == 0) { if (SEQ_GT(end, start)) { tp->t_dsack_bytes += (end - start); TCPSTAT_ADD(tcps_dsack_bytes, (end - start)); } else { tp->t_dsack_tlp_bytes += (start - end); TCPSTAT_ADD(tcps_dsack_bytes, (start - end)); } } else { if (SEQ_GT(end, start)) { tp->t_dsack_bytes += (end - start); TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start)); } else { tp->t_dsack_tlp_bytes += (start - end); TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end)); } } } static struct tcp_function_block * find_tcp_functions_locked(struct tcp_function_set *fs) { struct tcp_function *f; struct tcp_function_block *blk=NULL; TAILQ_FOREACH(f, &t_functions, tf_next) { if (strcmp(f->tf_name, fs->function_set_name) == 0) { blk = f->tf_fb; break; } } return(blk); } static struct tcp_function_block * find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) { struct tcp_function_block *rblk=NULL; struct tcp_function *f; TAILQ_FOREACH(f, &t_functions, tf_next) { if (f->tf_fb == blk) { rblk = blk; if (s) { *s = f; } break; } } return (rblk); } struct tcp_function_block * find_and_ref_tcp_functions(struct tcp_function_set *fs) { struct tcp_function_block *blk; rw_rlock(&tcp_function_lock); blk = find_tcp_functions_locked(fs); if (blk) refcount_acquire(&blk->tfb_refcnt); rw_runlock(&tcp_function_lock); return(blk); } struct tcp_function_block * find_and_ref_tcp_fb(struct tcp_function_block *blk) { struct tcp_function_block *rblk; rw_rlock(&tcp_function_lock); rblk = find_tcp_fb_locked(blk, NULL); if (rblk) refcount_acquire(&rblk->tfb_refcnt); rw_runlock(&tcp_function_lock); return(rblk); } /* Find a matching alias for the given tcp_function_block. */ int find_tcp_function_alias(struct tcp_function_block *blk, struct tcp_function_set *fs) { struct tcp_function *f; int found; found = 0; rw_rlock(&tcp_function_lock); TAILQ_FOREACH(f, &t_functions, tf_next) { if ((f->tf_fb == blk) && (strncmp(f->tf_name, blk->tfb_tcp_block_name, TCP_FUNCTION_NAME_LEN_MAX) != 0)) { /* Matching function block with different name. */ strncpy(fs->function_set_name, f->tf_name, TCP_FUNCTION_NAME_LEN_MAX); found = 1; break; } } /* Null terminate the string appropriately. */ if (found) { fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; } else { fs->function_set_name[0] = '\0'; } rw_runlock(&tcp_function_lock); return (found); } static struct tcp_function_block * find_and_ref_tcp_default_fb(void) { struct tcp_function_block *rblk; rw_rlock(&tcp_function_lock); rblk = tcp_func_set_ptr; refcount_acquire(&rblk->tfb_refcnt); rw_runlock(&tcp_function_lock); return (rblk); } void tcp_switch_back_to_default(struct tcpcb *tp) { struct tcp_function_block *tfb; KASSERT(tp->t_fb != &tcp_def_funcblk, ("%s: called by the built-in default stack", __func__)); /* * Release the old stack. This function will either find a new one * or panic. */ if (tp->t_fb->tfb_tcp_fb_fini != NULL) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); refcount_release(&tp->t_fb->tfb_refcnt); /* * Now, we'll find a new function block to use. * Start by trying the current user-selected * default, unless this stack is the user-selected * default. */ tfb = find_and_ref_tcp_default_fb(); if (tfb == tp->t_fb) { refcount_release(&tfb->tfb_refcnt); tfb = NULL; } /* Does the stack accept this connection? */ if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && (*tfb->tfb_tcp_handoff_ok)(tp)) { refcount_release(&tfb->tfb_refcnt); tfb = NULL; } /* Try to use that stack. */ if (tfb != NULL) { /* Initialize the new stack. If it succeeds, we are done. */ tp->t_fb = tfb; if (tp->t_fb->tfb_tcp_fb_init == NULL || (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) return; /* * Initialization failed. Release the reference count on * the stack. */ refcount_release(&tfb->tfb_refcnt); } /* * If that wasn't feasible, use the built-in default * stack which is not allowed to reject anyone. */ tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); if (tfb == NULL) { /* there always should be a default */ panic("Can't refer to tcp_def_funcblk"); } if (tfb->tfb_tcp_handoff_ok != NULL) { if ((*tfb->tfb_tcp_handoff_ok) (tp)) { /* The default stack cannot say no */ panic("Default stack rejects a new session?"); } } tp->t_fb = tfb; if (tp->t_fb->tfb_tcp_fb_init != NULL && (*tp->t_fb->tfb_tcp_fb_init)(tp)) { /* The default stack cannot fail */ panic("Default stack initialization failed"); } } static bool tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp, const struct sockaddr *sa, void *ctx) { struct ip *iph; #ifdef INET6 struct ip6_hdr *ip6; #endif struct udphdr *uh; struct tcphdr *th; int thlen; uint16_t port; TCPSTAT_INC(tcps_tunneled_pkts); if ((m->m_flags & M_PKTHDR) == 0) { /* Can't handle one that is not a pkt hdr */ TCPSTAT_INC(tcps_tunneled_errs); goto out; } thlen = sizeof(struct tcphdr); if (m->m_len < off + sizeof(struct udphdr) + thlen && (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) { TCPSTAT_INC(tcps_tunneled_errs); goto out; } iph = mtod(m, struct ip *); uh = (struct udphdr *)((caddr_t)iph + off); th = (struct tcphdr *)(uh + 1); thlen = th->th_off << 2; if (m->m_len < off + sizeof(struct udphdr) + thlen) { m = m_pullup(m, off + sizeof(struct udphdr) + thlen); if (m == NULL) { TCPSTAT_INC(tcps_tunneled_errs); goto out; } else { iph = mtod(m, struct ip *); uh = (struct udphdr *)((caddr_t)iph + off); th = (struct tcphdr *)(uh + 1); } } m->m_pkthdr.tcp_tun_port = port = uh->uh_sport; bcopy(th, uh, m->m_len - off); m->m_len -= sizeof(struct udphdr); m->m_pkthdr.len -= sizeof(struct udphdr); /* * We use the same algorithm for * both UDP and TCP for c-sum. So * the code in tcp_input will skip * the checksum. So we do nothing * with the flag (m->m_pkthdr.csum_flags). */ switch (iph->ip_v) { #ifdef INET case IPVERSION: iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr)); tcp_input_with_port(&m, &off, IPPROTO_TCP, port); break; #endif #ifdef INET6 case IPV6_VERSION >> 4: ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr)); tcp6_input_with_port(&m, &off, IPPROTO_TCP, port); break; #endif default: goto out; break; } return (true); out: m_freem(m); return (true); } static int sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) { int error=ENOENT; struct tcp_function_set fs; struct tcp_function_block *blk; memset(&fs, 0, sizeof(fs)); rw_rlock(&tcp_function_lock); blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); if (blk) { /* Found him */ strcpy(fs.function_set_name, blk->tfb_tcp_block_name); fs.pcbcnt = blk->tfb_refcnt; } rw_runlock(&tcp_function_lock); error = sysctl_handle_string(oidp, fs.function_set_name, sizeof(fs.function_set_name), req); /* Check for error or no change */ if (error != 0 || req->newptr == NULL) return(error); rw_wlock(&tcp_function_lock); blk = find_tcp_functions_locked(&fs); if ((blk == NULL) || (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { error = ENOENT; goto done; } tcp_func_set_ptr = blk; done: rw_wunlock(&tcp_function_lock); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0, sysctl_net_inet_default_tcp_functions, "A", "Set/get the default TCP functions"); static int sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) { int error, cnt, linesz; struct tcp_function *f; char *buffer, *cp; size_t bufsz, outsz; bool alias; cnt = 0; rw_rlock(&tcp_function_lock); TAILQ_FOREACH(f, &t_functions, tf_next) { cnt++; } rw_runlock(&tcp_function_lock); bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; buffer = malloc(bufsz, M_TEMP, M_WAITOK); error = 0; cp = buffer; linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', "Alias", "PCB count"); cp += linesz; bufsz -= linesz; outsz = linesz; rw_rlock(&tcp_function_lock); TAILQ_FOREACH(f, &t_functions, tf_next) { alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", f->tf_fb->tfb_tcp_block_name, (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', alias ? f->tf_name : "-", f->tf_fb->tfb_refcnt); if (linesz >= bufsz) { error = EOVERFLOW; break; } cp += linesz; bufsz -= linesz; outsz += linesz; } rw_runlock(&tcp_function_lock); if (error == 0) error = sysctl_handle_string(oidp, buffer, outsz + 1, req); free(buffer, M_TEMP); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, NULL, 0, sysctl_net_inet_list_available, "A", "list available TCP Function sets"); VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT; #ifdef INET VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL; #define V_udp4_tun_socket VNET(udp4_tun_socket) #endif #ifdef INET6 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL; #define V_udp6_tun_socket VNET(udp6_tun_socket) #endif static void tcp_over_udp_stop(void) { /* * This function assumes sysctl caller holds inp_rinfo_lock() * for writing! */ #ifdef INET if (V_udp4_tun_socket != NULL) { soclose(V_udp4_tun_socket); V_udp4_tun_socket = NULL; } #endif #ifdef INET6 if (V_udp6_tun_socket != NULL) { soclose(V_udp6_tun_socket); V_udp6_tun_socket = NULL; } #endif } static int tcp_over_udp_start(void) { uint16_t port; int ret; #ifdef INET struct sockaddr_in sin; #endif #ifdef INET6 struct sockaddr_in6 sin6; #endif /* * This function assumes sysctl caller holds inp_info_rlock() * for writing! */ port = V_tcp_udp_tunneling_port; if (ntohs(port) == 0) { /* Must have a port set */ return (EINVAL); } #ifdef INET if (V_udp4_tun_socket != NULL) { /* Already running -- must stop first */ return (EALREADY); } #endif #ifdef INET6 if (V_udp6_tun_socket != NULL) { /* Already running -- must stop first */ return (EALREADY); } #endif #ifdef INET if ((ret = socreate(PF_INET, &V_udp4_tun_socket, SOCK_DGRAM, IPPROTO_UDP, curthread->td_ucred, curthread))) { tcp_over_udp_stop(); return (ret); } /* Call the special UDP hook. */ if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket, tcp_recv_udp_tunneled_packet, tcp_ctlinput_viaudp, NULL))) { tcp_over_udp_stop(); return (ret); } /* Ok, we have a socket, bind it to the port. */ memset(&sin, 0, sizeof(struct sockaddr_in)); sin.sin_len = sizeof(struct sockaddr_in); sin.sin_family = AF_INET; sin.sin_port = htons(port); if ((ret = sobind(V_udp4_tun_socket, (struct sockaddr *)&sin, curthread))) { tcp_over_udp_stop(); return (ret); } #endif #ifdef INET6 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket, SOCK_DGRAM, IPPROTO_UDP, curthread->td_ucred, curthread))) { tcp_over_udp_stop(); return (ret); } /* Call the special UDP hook. */ if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket, tcp_recv_udp_tunneled_packet, tcp6_ctlinput_viaudp, NULL))) { tcp_over_udp_stop(); return (ret); } /* Ok, we have a socket, bind it to the port. */ memset(&sin6, 0, sizeof(struct sockaddr_in6)); sin6.sin6_len = sizeof(struct sockaddr_in6); sin6.sin6_family = AF_INET6; sin6.sin6_port = htons(port); if ((ret = sobind(V_udp6_tun_socket, (struct sockaddr *)&sin6, curthread))) { tcp_over_udp_stop(); return (ret); } #endif return (0); } static int sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS) { int error; uint32_t old, new; old = V_tcp_udp_tunneling_port; new = old; error = sysctl_handle_int(oidp, &new, 0, req); if ((error == 0) && (req->newptr != NULL)) { if ((new < TCP_TUNNELING_PORT_MIN) || (new > TCP_TUNNELING_PORT_MAX)) { error = EINVAL; } else { V_tcp_udp_tunneling_port = new; if (old != 0) { tcp_over_udp_stop(); } if (new != 0) { error = tcp_over_udp_start(); } } } return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &VNET_NAME(tcp_udp_tunneling_port), 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU", "Tunneling port for tcp over udp"); VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT; static int sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS) { int error, new; new = V_tcp_udp_tunneling_overhead; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if ((new < TCP_TUNNELING_OVERHEAD_MIN) || (new > TCP_TUNNELING_OVERHEAD_MAX)) error = EINVAL; else V_tcp_udp_tunneling_overhead = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &VNET_NAME(tcp_udp_tunneling_overhead), 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU", "MSS reduction when using tcp over udp"); /* * Exports one (struct tcp_function_info) for each alias/name. */ static int sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) { int cnt, error; struct tcp_function *f; struct tcp_function_info tfi; /* * We don't allow writes. */ if (req->newptr != NULL) return (EINVAL); /* * Wire the old buffer so we can directly copy the functions to * user space without dropping the lock. */ if (req->oldptr != NULL) { error = sysctl_wire_old_buffer(req, 0); if (error) return (error); } /* * Walk the list and copy out matching entries. If INVARIANTS * is compiled in, also walk the list to verify the length of * the list matches what we have recorded. */ rw_rlock(&tcp_function_lock); cnt = 0; #ifndef INVARIANTS if (req->oldptr == NULL) { cnt = tcp_fb_cnt; goto skip_loop; } #endif TAILQ_FOREACH(f, &t_functions, tf_next) { #ifdef INVARIANTS cnt++; #endif if (req->oldptr != NULL) { bzero(&tfi, sizeof(tfi)); tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; tfi.tfi_id = f->tf_fb->tfb_id; (void)strlcpy(tfi.tfi_alias, f->tf_name, sizeof(tfi.tfi_alias)); (void)strlcpy(tfi.tfi_name, f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); /* * Don't stop on error, as that is the * mechanism we use to accumulate length * information if the buffer was too short. */ } } KASSERT(cnt == tcp_fb_cnt, ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); #ifndef INVARIANTS skip_loop: #endif rw_runlock(&tcp_function_lock); if (req->oldptr == NULL) error = SYSCTL_OUT(req, NULL, (cnt + 1) * sizeof(struct tcp_function_info)); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", "List TCP function block name-to-ID mappings"); /* * tfb_tcp_handoff_ok() function for the default stack. * Note that we'll basically try to take all comers. */ static int tcp_default_handoff_ok(struct tcpcb *tp) { return (0); } /* * tfb_tcp_fb_init() function for the default stack. * * This handles making sure we have appropriate timers set if you are * transitioning a socket that has some amount of setup done. * * The init() fuction from the default can *never* return non-zero i.e. * it is required to always succeed since it is the stack of last resort! */ static int tcp_default_fb_init(struct tcpcb *tp) { struct socket *so; INP_WLOCK_ASSERT(tp->t_inpcb); KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, ("%s: connection %p in unexpected state %d", __func__, tp, tp->t_state)); /* * Nothing to do for ESTABLISHED or LISTEN states. And, we don't * know what to do for unexpected states (which includes TIME_WAIT). */ if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) return (0); /* * Make sure some kind of transmission timer is set if there is * outstanding data. */ so = tp->t_inpcb->inp_socket; if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || tcp_timer_active(tp, TT_PERSIST))) { /* * If the session has established and it looks like it should * be in the persist state, set the persist timer. Otherwise, * set the retransmit timer. */ if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && (int32_t)(tp->snd_nxt - tp->snd_una) < (int32_t)sbavail(&so->so_snd)) tcp_setpersist(tp); else tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); } /* All non-embryonic sessions get a keepalive timer. */ if (!tcp_timer_active(tp, TT_KEEP)) tcp_timer_activate(tp, TT_KEEP, TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : TP_KEEPINIT(tp)); /* * Make sure critical variables are initialized * if transitioning while in Recovery. */ if IN_FASTRECOVERY(tp->t_flags) { if (tp->sackhint.recover_fs == 0) tp->sackhint.recover_fs = max(1, tp->snd_nxt - tp->snd_una); } return (0); } /* * tfb_tcp_fb_fini() function for the default stack. * * This changes state as necessary (or prudent) to prepare for another stack * to assume responsibility for the connection. */ static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) { INP_WLOCK_ASSERT(tp->t_inpcb); return; } /* * Target size of TCP PCB hash tables. Must be a power of two. * * Note that this can be overridden by the kernel environment * variable net.inet.tcp.tcbhashsize */ #ifndef TCBHASHSIZE #define TCBHASHSIZE 0 #endif /* * XXX * Callouts should be moved into struct tcp directly. They are currently * separate because the tcpcb structure is exported to userland for sysctl * parsing purposes, which do not know about callouts. */ struct tcpcb_mem { struct tcpcb tcb; struct tcp_timer tt; struct cc_var ccv; #ifdef TCP_HHOOK struct osd osd; #endif }; VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); #define V_tcpcb_zone VNET(tcpcb_zone) MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); static struct mtx isn_mtx; #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) #define ISN_LOCK() mtx_lock(&isn_mtx) #define ISN_UNLOCK() mtx_unlock(&isn_mtx) INPCBSTORAGE_DEFINE(tcpcbstor, "tcpinp", "tcp_inpcb", "tcp", "tcphash"); /* * Take a value and get the next power of 2 that doesn't overflow. * Used to size the tcp_inpcb hash buckets. */ static int maketcp_hashsize(int size) { int hashsize; /* * auto tune. * get the next power of 2 higher than maxsockets. */ hashsize = 1 << fls(size); /* catch overflow, and just go one power of 2 smaller */ if (hashsize < size) { hashsize = 1 << (fls(size) - 1); } return (hashsize); } static volatile int next_tcp_stack_id = 1; /* * Register a TCP function block with the name provided in the names * array. (Note that this function does NOT automatically register * blk->tfb_tcp_block_name as a stack name. Therefore, you should * explicitly include blk->tfb_tcp_block_name in the list of names if * you wish to register the stack with that name.) * * Either all name registrations will succeed or all will fail. If * a name registration fails, the function will update the num_names * argument to point to the array index of the name that encountered * the failure. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, const char *names[], int *num_names) { struct tcp_function *n; struct tcp_function_set fs; int error, i; KASSERT(names != NULL && *num_names > 0, ("%s: Called with 0-length name list", __func__)); KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); KASSERT(rw_initialized(&tcp_function_lock), ("%s: called too early", __func__)); if ((blk->tfb_tcp_output == NULL) || (blk->tfb_tcp_do_segment == NULL) || (blk->tfb_tcp_ctloutput == NULL) || (strlen(blk->tfb_tcp_block_name) == 0)) { /* * These functions are required and you * need a name. */ *num_names = 0; return (EINVAL); } if (blk->tfb_tcp_timer_stop_all || blk->tfb_tcp_timer_activate || blk->tfb_tcp_timer_active || blk->tfb_tcp_timer_stop) { /* * If you define one timer function you * must have them all. */ if ((blk->tfb_tcp_timer_stop_all == NULL) || (blk->tfb_tcp_timer_activate == NULL) || (blk->tfb_tcp_timer_active == NULL) || (blk->tfb_tcp_timer_stop == NULL)) { *num_names = 0; return (EINVAL); } } if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { *num_names = 0; return (EINVAL); } refcount_init(&blk->tfb_refcnt, 0); blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); for (i = 0; i < *num_names; i++) { n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); if (n == NULL) { error = ENOMEM; goto cleanup; } n->tf_fb = blk; (void)strlcpy(fs.function_set_name, names[i], sizeof(fs.function_set_name)); rw_wlock(&tcp_function_lock); if (find_tcp_functions_locked(&fs) != NULL) { /* Duplicate name space not allowed */ rw_wunlock(&tcp_function_lock); free(n, M_TCPFUNCTIONS); error = EALREADY; goto cleanup; } (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); TAILQ_INSERT_TAIL(&t_functions, n, tf_next); tcp_fb_cnt++; rw_wunlock(&tcp_function_lock); } return(0); cleanup: /* * Deregister the names we just added. Because registration failed * for names[i], we don't need to deregister that name. */ *num_names = i; rw_wlock(&tcp_function_lock); while (--i >= 0) { TAILQ_FOREACH(n, &t_functions, tf_next) { if (!strncmp(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX)) { TAILQ_REMOVE(&t_functions, n, tf_next); tcp_fb_cnt--; n->tf_fb = NULL; free(n, M_TCPFUNCTIONS); break; } } } rw_wunlock(&tcp_function_lock); return (error); } /* * Register a TCP function block using the name provided in the name * argument. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, int wait) { const char *name_list[1]; int num_names, rv; num_names = 1; if (name != NULL) name_list[0] = name; else name_list[0] = blk->tfb_tcp_block_name; rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); return (rv); } /* * Register a TCP function block using the name defined in * blk->tfb_tcp_block_name. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions(struct tcp_function_block *blk, int wait) { return (register_tcp_functions_as_name(blk, NULL, wait)); } /* * Deregister all names associated with a function block. This * functionally removes the function block from use within the system. * * When called with a true quiesce argument, mark the function block * as being removed so no more stacks will use it and determine * whether the removal would succeed. * * When called with a false quiesce argument, actually attempt the * removal. * * When called with a force argument, attempt to switch all TCBs to * use the default stack instead of returning EBUSY. * * Returns 0 on success (or if the removal would succeed, or an error * code on failure. */ int deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, bool force) { struct tcp_function *f; if (blk == &tcp_def_funcblk) { /* You can't un-register the default */ return (EPERM); } rw_wlock(&tcp_function_lock); if (blk == tcp_func_set_ptr) { /* You can't free the current default */ rw_wunlock(&tcp_function_lock); return (EBUSY); } /* Mark the block so no more stacks can use it. */ blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; /* * If TCBs are still attached to the stack, attempt to switch them * to the default stack. */ if (force && blk->tfb_refcnt) { struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, INPLOOKUP_WLOCKPCB); struct inpcb *inp; struct tcpcb *tp; VNET_ITERATOR_DECL(vnet_iter); rw_wunlock(&tcp_function_lock); VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); while ((inp = inp_next(&inpi)) != NULL) { if (inp->inp_flags & INP_TIMEWAIT) continue; tp = intotcpcb(inp); if (tp == NULL || tp->t_fb != blk) continue; tcp_switch_back_to_default(tp); } CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); rw_wlock(&tcp_function_lock); } if (blk->tfb_refcnt) { /* TCBs still attached. */ rw_wunlock(&tcp_function_lock); return (EBUSY); } if (quiesce) { /* Skip removal. */ rw_wunlock(&tcp_function_lock); return (0); } /* Remove any function names that map to this function block. */ while (find_tcp_fb_locked(blk, &f) != NULL) { TAILQ_REMOVE(&t_functions, f, tf_next); tcp_fb_cnt--; f->tf_fb = NULL; free(f, M_TCPFUNCTIONS); } rw_wunlock(&tcp_function_lock); return (0); } static void tcp_drain(void) { struct epoch_tracker et; VNET_ITERATOR_DECL(vnet_iter); if (!do_tcpdrain) return; NET_EPOCH_ENTER(et); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, INPLOOKUP_WLOCKPCB); struct inpcb *inpb; struct tcpcb *tcpb; /* * Walk the tcpbs, if existing, and flush the reassembly queue, * if there is one... * XXX: The "Net/3" implementation doesn't imply that the TCP * reassembly queue should be flushed, but in a situation * where we're really low on mbufs, this is potentially * useful. */ while ((inpb = inp_next(&inpi)) != NULL) { if (inpb->inp_flags & INP_TIMEWAIT) continue; if ((tcpb = intotcpcb(inpb)) != NULL) { tcp_reass_flush(tcpb); tcp_clean_sackreport(tcpb); #ifdef TCP_BLACKBOX tcp_log_drain(tcpb); #endif #ifdef TCPPCAP if (tcp_pcap_aggressive_free) { /* Free the TCP PCAP queues. */ tcp_pcap_drain(&(tcpb->t_inpkts)); tcp_pcap_drain(&(tcpb->t_outpkts)); } #endif } } CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); NET_EPOCH_EXIT(et); } static void tcp_vnet_init(void *arg __unused) { #ifdef TCP_HHOOK if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register helper hook\n", __func__); if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register helper hook\n", __func__); #endif #ifdef STATS if (tcp_stats_init()) printf("%s: WARNING: unable to initialise TCP stats\n", __func__); #endif in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize, tcp_tcbhashsize); /* * These have to be type stable for the benefit of the timers. */ V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_zone_set_max(V_tcpcb_zone, maxsockets); uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); tcp_tw_init(); syncache_init(); tcp_hc_init(); TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); tcp_fastopen_init(); COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); V_tcp_msl = TCPTV_MSL; } VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_vnet_init, NULL); static void tcp_init(void *arg __unused) { const char *tcbhash_tuneable; int hashsize; tcp_reass_global_init(); /* XXX virtualize those below? */ tcp_delacktime = TCPTV_DELACK; tcp_keepinit = TCPTV_KEEP_INIT; tcp_keepidle = TCPTV_KEEP_IDLE; tcp_keepintvl = TCPTV_KEEPINTVL; tcp_maxpersistidle = TCPTV_KEEP_IDLE; tcp_rexmit_initial = TCPTV_RTOBASE; if (tcp_rexmit_initial < 1) tcp_rexmit_initial = 1; tcp_rexmit_min = TCPTV_MIN; if (tcp_rexmit_min < 1) tcp_rexmit_min = 1; tcp_persmin = TCPTV_PERSMIN; tcp_persmax = TCPTV_PERSMAX; tcp_rexmit_slop = TCPTV_CPU_VAR; tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; /* Setup the tcp function block list */ TAILQ_INIT(&t_functions); rw_init(&tcp_function_lock, "tcp_func_lock"); register_tcp_functions(&tcp_def_funcblk, M_WAITOK); #ifdef TCP_BLACKBOX /* Initialize the TCP logging data. */ tcp_log_init(); #endif arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); if (tcp_soreceive_stream) { #ifdef INET tcp_protosw.pr_soreceive = soreceive_stream; #endif #ifdef INET6 tcp6_protosw.pr_soreceive = soreceive_stream; #endif /* INET6 */ } #ifdef INET6 -#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) + max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); #else /* INET6 */ -#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) + max_protohdr_grow(sizeof(struct tcpiphdr)); #endif /* INET6 */ - if (max_protohdr < TCP_MINPROTOHDR) - max_protohdr = TCP_MINPROTOHDR; - if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) - panic("tcp_init"); -#undef TCP_MINPROTOHDR ISN_LOCK_INIT(); EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, SHUTDOWN_PRI_DEFAULT); EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); tcp_would_have_but = counter_u64_alloc(M_WAITOK); tcp_comp_total = counter_u64_alloc(M_WAITOK); tcp_uncomp_total = counter_u64_alloc(M_WAITOK); tcp_bad_csums = counter_u64_alloc(M_WAITOK); #ifdef TCPPCAP tcp_pcap_init(); #endif hashsize = TCBHASHSIZE; tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); if (hashsize == 0) { /* * Auto tune the hash size based on maxsockets. * A perfect hash would have a 1:1 mapping * (hashsize = maxsockets) however it's been * suggested that O(2) average is better. */ hashsize = maketcp_hashsize(maxsockets / 4); /* * Our historical default is 512, * do not autotune lower than this. */ if (hashsize < 512) hashsize = 512; if (bootverbose) printf("%s: %s auto tuned to %d\n", __func__, tcbhash_tuneable, hashsize); } /* * We require a hashsize to be a power of two. * Previously if it was not a power of two we would just reset it * back to 512, which could be a nasty surprise if you did not notice * the error message. * Instead what we do is clip it to the closest power of two lower * than the specified hash value. */ if (!powerof2(hashsize)) { int oldhashsize = hashsize; hashsize = maketcp_hashsize(hashsize); /* prevent absurdly low value */ if (hashsize < 16) hashsize = 16; printf("%s: WARNING: TCB hash size not a power of 2, " "clipped from %d to %d.\n", __func__, oldhashsize, hashsize); } tcp_tcbhashsize = hashsize; #ifdef INET IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput); #endif #ifdef INET6 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput); #endif } SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL); #ifdef VIMAGE static void tcp_destroy(void *unused __unused) { int n; #ifdef TCP_HHOOK int error; #endif /* * All our processes are gone, all our sockets should be cleaned * up, which means, we should be past the tcp_discardcb() calls. * Sleep to let all tcpcb timers really disappear and cleanup. */ for (;;) { INP_INFO_WLOCK(&V_tcbinfo); n = V_tcbinfo.ipi_count; INP_INFO_WUNLOCK(&V_tcbinfo); if (n == 0) break; pause("tcpdes", hz / 10); } tcp_hc_destroy(); syncache_destroy(); tcp_tw_destroy(); in_pcbinfo_destroy(&V_tcbinfo); /* tcp_discardcb() clears the sack_holes up. */ uma_zdestroy(V_sack_hole_zone); uma_zdestroy(V_tcpcb_zone); /* * Cannot free the zone until all tcpcbs are released as we attach * the allocations to them. */ tcp_fastopen_destroy(); COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); VNET_PCPUSTAT_FREE(tcpstat); #ifdef TCP_HHOOK error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); if (error != 0) { printf("%s: WARNING: unable to deregister helper hook " "type=%d, id=%d: error %d returned\n", __func__, HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); } error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); if (error != 0) { printf("%s: WARNING: unable to deregister helper hook " "type=%d, id=%d: error %d returned\n", __func__, HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); } #endif } VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); #endif void tcp_fini(void *xtp) { } /* * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. * tcp_template used to store this data in mbufs, but we now recopy it out * of the tcpcb each time to conserve mbufs. */ void tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) { struct tcphdr *th = (struct tcphdr *)tcp_ptr; INP_WLOCK_ASSERT(inp); #ifdef INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { struct ip6_hdr *ip6; ip6 = (struct ip6_hdr *)ip_ptr; ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | (inp->inp_flow & IPV6_FLOWINFO_MASK); ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | (IPV6_VERSION & IPV6_VERSION_MASK); if (port == 0) ip6->ip6_nxt = IPPROTO_TCP; else ip6->ip6_nxt = IPPROTO_UDP; ip6->ip6_plen = htons(sizeof(struct tcphdr)); ip6->ip6_src = inp->in6p_laddr; ip6->ip6_dst = inp->in6p_faddr; } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET { struct ip *ip; ip = (struct ip *)ip_ptr; ip->ip_v = IPVERSION; ip->ip_hl = 5; ip->ip_tos = inp->inp_ip_tos; ip->ip_len = 0; ip->ip_id = 0; ip->ip_off = 0; ip->ip_ttl = inp->inp_ip_ttl; ip->ip_sum = 0; if (port == 0) ip->ip_p = IPPROTO_TCP; else ip->ip_p = IPPROTO_UDP; ip->ip_src = inp->inp_laddr; ip->ip_dst = inp->inp_faddr; } #endif /* INET */ th->th_sport = inp->inp_lport; th->th_dport = inp->inp_fport; th->th_seq = 0; th->th_ack = 0; th->th_off = 5; tcp_set_flags(th, 0); th->th_win = 0; th->th_urp = 0; th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ } /* * Create template to be used to send tcp packets on a connection. * Allocates an mbuf and fills in a skeletal tcp/ip header. The only * use for this function is in keepalives, which use tcp_respond. */ struct tcptemp * tcpip_maketemplate(struct inpcb *inp) { struct tcptemp *t; t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); if (t == NULL) return (NULL); tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); return (t); } /* * Send a single message to the TCP at address specified by * the given TCP/IP header. If m == NULL, then we make a copy * of the tcpiphdr at th and send directly to the addressed host. * This is used to force keep alive messages out using the TCP * template for a connection. If flags are given then we send * a message back to the TCP which originated the segment th, * and discard the mbuf containing it and any other attached mbufs. * * In any case the ack and sequence number of the transmitted * segment are as specified by the parameters. * * NOTE: If m != NULL, then th must point to *inside* the mbuf. */ void tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) { struct tcpopt to; struct inpcb *inp; struct ip *ip; struct mbuf *optm; struct udphdr *uh = NULL; struct tcphdr *nth; struct tcp_log_buffer *lgb; u_char *optp; #ifdef INET6 struct ip6_hdr *ip6; int isipv6; #endif /* INET6 */ int optlen, tlen, win, ulen; bool incl_opts; uint16_t port; int output_ret; #ifdef INVARIANTS int thflags = tcp_get_flags(th); #endif KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); NET_EPOCH_ASSERT(); #ifdef INET6 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); ip6 = ipgen; #endif /* INET6 */ ip = ipgen; if (tp != NULL) { inp = tp->t_inpcb; KASSERT(inp != NULL, ("tcp control block w/o inpcb")); INP_LOCK_ASSERT(inp); } else inp = NULL; if (m != NULL) { #ifdef INET6 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) port = m->m_pkthdr.tcp_tun_port; else #endif if (ip && (ip->ip_p == IPPROTO_UDP)) port = m->m_pkthdr.tcp_tun_port; else port = 0; } else port = tp->t_port; incl_opts = false; win = 0; if (tp != NULL) { if (!(flags & TH_RST)) { win = sbspace(&inp->inp_socket->so_rcv); if (win > TCP_MAXWIN << tp->rcv_scale) win = TCP_MAXWIN << tp->rcv_scale; } if ((tp->t_flags & TF_NOOPT) == 0) incl_opts = true; } if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return; m->m_data += max_linkhdr; #ifdef INET6 if (isipv6) { bcopy((caddr_t)ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr)); ip6 = mtod(m, struct ip6_hdr *); nth = (struct tcphdr *)(ip6 + 1); if (port) { /* Insert a UDP header */ uh = (struct udphdr *)nth; uh->uh_sport = htons(V_tcp_udp_tunneling_port); uh->uh_dport = port; nth = (struct tcphdr *)(uh + 1); } } else #endif /* INET6 */ { bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); ip = mtod(m, struct ip *); nth = (struct tcphdr *)(ip + 1); if (port) { /* Insert a UDP header */ uh = (struct udphdr *)nth; uh->uh_sport = htons(V_tcp_udp_tunneling_port); uh->uh_dport = port; nth = (struct tcphdr *)(uh + 1); } } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); flags = TH_ACK; } else if ((!M_WRITABLE(m)) || (port != 0)) { struct mbuf *n; /* Can't reuse 'm', allocate a new mbuf. */ n = m_gethdr(M_NOWAIT, MT_DATA); if (n == NULL) { m_freem(m); return; } if (!m_dup_pkthdr(n, m, M_NOWAIT)) { m_freem(m); m_freem(n); return; } n->m_data += max_linkhdr; /* m_len is set later */ #define xchg(a,b,type) { type t; t=a; a=b; b=t; } #ifdef INET6 if (isipv6) { bcopy((caddr_t)ip6, mtod(n, caddr_t), sizeof(struct ip6_hdr)); ip6 = mtod(n, struct ip6_hdr *); xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); nth = (struct tcphdr *)(ip6 + 1); if (port) { /* Insert a UDP header */ uh = (struct udphdr *)nth; uh->uh_sport = htons(V_tcp_udp_tunneling_port); uh->uh_dport = port; nth = (struct tcphdr *)(uh + 1); } } else #endif /* INET6 */ { bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); ip = mtod(n, struct ip *); xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); nth = (struct tcphdr *)(ip + 1); if (port) { /* Insert a UDP header */ uh = (struct udphdr *)nth; uh->uh_sport = htons(V_tcp_udp_tunneling_port); uh->uh_dport = port; nth = (struct tcphdr *)(uh + 1); } } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); xchg(nth->th_dport, nth->th_sport, uint16_t); th = nth; m_freem(m); m = n; } else { /* * reuse the mbuf. * XXX MRT We inherit the FIB, which is lucky. */ m_freem(m->m_next); m->m_next = NULL; m->m_data = (caddr_t)ipgen; /* m_len is set later */ #ifdef INET6 if (isipv6) { xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); nth = (struct tcphdr *)(ip + 1); } if (th != nth) { /* * this is usually a case when an extension header * exists between the IPv6 header and the * TCP header. */ nth->th_sport = th->th_sport; nth->th_dport = th->th_dport; } xchg(nth->th_dport, nth->th_sport, uint16_t); #undef xchg } tlen = 0; #ifdef INET6 if (isipv6) tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET tlen = sizeof (struct tcpiphdr); #endif if (port) tlen += sizeof (struct udphdr); #ifdef INVARIANTS m->m_len = 0; KASSERT(M_TRAILINGSPACE(m) >= tlen, ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", m, tlen, (long)M_TRAILINGSPACE(m))); #endif m->m_len = tlen; to.to_flags = 0; if (incl_opts) { /* Make sure we have room. */ if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { m->m_next = m_get(M_NOWAIT, MT_DATA); if (m->m_next) { optp = mtod(m->m_next, u_char *); optm = m->m_next; } else incl_opts = false; } else { optp = (u_char *) (nth + 1); optm = m; } } if (incl_opts) { /* Timestamps. */ if (tp->t_flags & TF_RCVD_TSTMP) { to.to_tsval = tcp_ts_getticks() + tp->ts_offset; to.to_tsecr = tp->ts_recent; to.to_flags |= TOF_TS; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* TCP-MD5 (RFC2385). */ if (tp->t_flags & TF_SIGNATURE) to.to_flags |= TOF_SIGNATURE; #endif /* Add the options. */ tlen += optlen = tcp_addoptions(&to, optp); /* Update m_len in the correct mbuf. */ optm->m_len += optlen; } else optlen = 0; #ifdef INET6 if (isipv6) { if (uh) { ulen = tlen - sizeof(struct ip6_hdr); uh->uh_ulen = htons(ulen); } ip6->ip6_flow = 0; ip6->ip6_vfc = IPV6_VERSION; if (port) ip6->ip6_nxt = IPPROTO_UDP; else ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_plen = htons(tlen - sizeof(*ip6)); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { if (uh) { ulen = tlen - sizeof(struct ip); uh->uh_ulen = htons(ulen); } ip->ip_len = htons(tlen); ip->ip_ttl = V_ip_defttl; if (port) { ip->ip_p = IPPROTO_UDP; } else { ip->ip_p = IPPROTO_TCP; } if (V_path_mtu_discovery) ip->ip_off |= htons(IP_DF); } #endif m->m_pkthdr.len = tlen; m->m_pkthdr.rcvif = NULL; #ifdef MAC if (inp != NULL) { /* * Packet is associated with a socket, so allow the * label of the response to reflect the socket label. */ INP_LOCK_ASSERT(inp); mac_inpcb_create_mbuf(inp, m); } else { /* * Packet is not associated with a socket, so possibly * update the label in place. */ mac_netinet_tcp_reply(m); } #endif nth->th_seq = htonl(seq); nth->th_ack = htonl(ack); nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; tcp_set_flags(nth, flags); if (tp != NULL) nth->th_win = htons((u_short) (win >> tp->rcv_scale)); else nth->th_win = htons((u_short)win); nth->th_urp = 0; #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (to.to_flags & TOF_SIGNATURE) { if (!TCPMD5_ENABLED() || TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { m_freem(m); return; } } #endif #ifdef INET6 if (isipv6) { if (port) { m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); nth->th_sum = 0; } else { m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); nth->th_sum = in6_cksum_pseudo(ip6, tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); } ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : NULL, NULL); } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET { if (port) { uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(ulen + IPPROTO_UDP)); m->m_pkthdr.csum_flags = CSUM_UDP; m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); nth->th_sum = 0; } else { m->m_pkthdr.csum_flags = CSUM_TCP; m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); } } #endif /* INET */ #ifdef TCPDEBUG if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); #endif TCP_PROBE3(debug__output, tp, th, m); if (flags & TH_RST) TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); lgb = NULL; if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) { if (INP_WLOCKED(inp)) { union tcp_log_stackspecific log; struct timeval tv; memset(&log.u_bbr, 0, sizeof(log.u_bbr)); log.u_bbr.inhpts = tp->t_inpcb->inp_in_hpts; log.u_bbr.flex8 = 4; log.u_bbr.pkts_out = tp->t_maxseg; log.u_bbr.timeStamp = tcp_get_usecs(&tv); log.u_bbr.delivered = 0; lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT, ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv); } else { /* * We can not log the packet, since we only own the * read lock, but a write lock is needed. The read lock * is not upgraded to a write lock, since only getting * the read lock was done intentionally to improve the * handling of SYN flooding attacks. * This happens only for pure SYN segments received in * the initial CLOSED state, or received in a more * advanced state than listen and the UDP encapsulation * port is unexpected. * The incoming SYN segments do not really belong to * the TCP connection and the handling does not change * the state of the TCP connection. Therefore, the * sending of the RST segments is not logged. Please * note that also the incoming SYN segments are not * logged. * * The following code ensures that the above description * is and stays correct. */ KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN && (tp->t_state == TCPS_CLOSED || (tp->t_state > TCPS_LISTEN && tp->t_port != port)), ("%s: Logging of TCP segment with flags 0x%b and " "UDP encapsulation port %u skipped in state %s", __func__, thflags, PRINT_TH_FLAGS, ntohs(port), tcpstates[tp->t_state])); } } #ifdef INET6 if (isipv6) { TCP_PROBE5(send, NULL, tp, ip6, tp, nth); output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); } #endif /* INET6 */ #if defined(INET) && defined(INET6) else #endif #ifdef INET { TCP_PROBE5(send, NULL, tp, ip, tp, nth); output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); } #endif if (lgb != NULL) lgb->tlb_errno = output_ret; } /* * Create a new TCP control block, making an * empty reassembly queue and hooking it to the argument * protocol control block. The `inp' parameter must have * come from the zone allocator set up in tcp_init(). */ struct tcpcb * tcp_newtcpcb(struct inpcb *inp) { struct tcpcb_mem *tm; struct tcpcb *tp; #ifdef INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); if (tm == NULL) return (NULL); tp = &tm->tcb; /* Initialise cc_var struct for this tcpcb. */ tp->ccv = &tm->ccv; tp->ccv->type = IPPROTO_TCP; tp->ccv->ccvc.tcp = tp; rw_rlock(&tcp_function_lock); tp->t_fb = tcp_func_set_ptr; refcount_acquire(&tp->t_fb->tfb_refcnt); rw_runlock(&tcp_function_lock); /* * Use the current system default CC algorithm. */ cc_attach(tp, CC_DEFAULT_ALGO()); /* * The tcpcb will hold a reference on its inpcb until tcp_discardcb() * is called. */ in_pcbref(inp); /* Reference for tcpcb */ tp->t_inpcb = inp; if (CC_ALGO(tp)->cb_init != NULL) if (CC_ALGO(tp)->cb_init(tp->ccv, NULL) > 0) { cc_detach(tp); if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); in_pcbrele_wlocked(inp); refcount_release(&tp->t_fb->tfb_refcnt); uma_zfree(V_tcpcb_zone, tm); return (NULL); } #ifdef TCP_HHOOK tp->osd = &tm->osd; if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); in_pcbrele_wlocked(inp); refcount_release(&tp->t_fb->tfb_refcnt); uma_zfree(V_tcpcb_zone, tm); return (NULL); } #endif #ifdef VIMAGE tp->t_vnet = inp->inp_vnet; #endif tp->t_timers = &tm->tt; TAILQ_INIT(&tp->t_segq); tp->t_maxseg = #ifdef INET6 isipv6 ? V_tcp_v6mssdflt : #endif /* INET6 */ V_tcp_mssdflt; /* Set up our timeouts. */ callout_init(&tp->t_timers->tt_rexmt, 1); callout_init(&tp->t_timers->tt_persist, 1); callout_init(&tp->t_timers->tt_keep, 1); callout_init(&tp->t_timers->tt_2msl, 1); callout_init(&tp->t_timers->tt_delack, 1); if (V_tcp_do_rfc1323) tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); if (V_tcp_do_sack) tp->t_flags |= TF_SACK_PERMIT; TAILQ_INIT(&tp->snd_holes); /* * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives * reasonable initial retransmit time. */ tp->t_srtt = TCPTV_SRTTBASE; tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; tp->t_rttmin = tcp_rexmit_min; tp->t_rxtcur = tcp_rexmit_initial; tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->t_rcvtime = ticks; /* * IPv4 TTL initialization is necessary for an IPv6 socket as well, * because the socket may be bound to an IPv6 wildcard address, * which may match an IPv4-mapped IPv6 address. */ inp->inp_ip_ttl = V_ip_defttl; inp->inp_ppcb = tp; #ifdef TCPPCAP /* * Init the TCP PCAP queues. */ tcp_pcap_tcpcb_init(tp); #endif #ifdef TCP_BLACKBOX /* Initialize the per-TCPCB log data. */ tcp_log_tcpcbinit(tp); #endif tp->t_pacing_rate = -1; if (tp->t_fb->tfb_tcp_fb_init) { if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) { refcount_release(&tp->t_fb->tfb_refcnt); in_pcbrele_wlocked(inp); uma_zfree(V_tcpcb_zone, tm); return (NULL); } } #ifdef STATS if (V_tcp_perconn_stats_enable == 1) tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); #endif if (V_tcp_do_lrd) tp->t_flags |= TF_LRD; return (tp); /* XXX */ } /* * Drop a TCP connection, reporting * the specified error. If connection is synchronized, * then send a RST to peer. */ struct tcpcb * tcp_drop(struct tcpcb *tp, int errno) { struct socket *so = tp->t_inpcb->inp_socket; NET_EPOCH_ASSERT(); INP_WLOCK_ASSERT(tp->t_inpcb); if (TCPS_HAVERCVDSYN(tp->t_state)) { tcp_state_change(tp, TCPS_CLOSED); /* Don't use tcp_output() here due to possible recursion. */ (void)tcp_output_nodrop(tp); TCPSTAT_INC(tcps_drops); } else TCPSTAT_INC(tcps_conndrops); if (errno == ETIMEDOUT && tp->t_softerror) errno = tp->t_softerror; so->so_error = errno; return (tcp_close(tp)); } void tcp_discardcb(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; INP_WLOCK_ASSERT(inp); /* * Make sure that all of our timers are stopped before we delete the * PCB. * * If stopping a timer fails, we schedule a discard function in same * callout, and the last discard function called will take care of * deleting the tcpcb. */ tp->t_timers->tt_draincnt = 0; tcp_timer_stop(tp, TT_REXMT); tcp_timer_stop(tp, TT_PERSIST); tcp_timer_stop(tp, TT_KEEP); tcp_timer_stop(tp, TT_2MSL); tcp_timer_stop(tp, TT_DELACK); if (tp->t_fb->tfb_tcp_timer_stop_all) { /* * Call the stop-all function of the methods, * this function should call the tcp_timer_stop() * method with each of the function specific timeouts. * That stop will be called via the tfb_tcp_timer_stop() * which should use the async drain function of the * callout system (see tcp_var.h). */ tp->t_fb->tfb_tcp_timer_stop_all(tp); } /* free the reassembly queue, if any */ tcp_reass_flush(tp); #ifdef TCP_OFFLOAD /* Disconnect offload device, if any. */ if (tp->t_flags & TF_TOE) tcp_offload_detach(tp); #endif tcp_free_sackholes(tp); #ifdef TCPPCAP /* Free the TCP PCAP queues. */ tcp_pcap_drain(&(tp->t_inpkts)); tcp_pcap_drain(&(tp->t_outpkts)); #endif /* Allow the CC algorithm to clean up after itself. */ if (CC_ALGO(tp)->cb_destroy != NULL) CC_ALGO(tp)->cb_destroy(tp->ccv); CC_DATA(tp) = NULL; /* Detach from the CC algorithm */ cc_detach(tp); #ifdef TCP_HHOOK khelp_destroy_osd(tp->osd); #endif #ifdef STATS stats_blob_destroy(tp->t_stats); #endif CC_ALGO(tp) = NULL; inp->inp_ppcb = NULL; if (tp->t_timers->tt_draincnt == 0) { bool released __diagused; released = tcp_freecb(tp); KASSERT(!released, ("%s: inp %p should not have been released " "here", __func__, inp)); } } bool tcp_freecb(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; struct socket *so = inp->inp_socket; #ifdef INET6 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif INP_WLOCK_ASSERT(inp); MPASS(tp->t_timers->tt_draincnt == 0); /* We own the last reference on tcpcb, let's free it. */ #ifdef TCP_BLACKBOX tcp_log_tcpcbfini(tp); #endif TCPSTATES_DEC(tp->t_state); if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); /* * If we got enough samples through the srtt filter, * save the rtt and rttvar in the routing entry. * 'Enough' is arbitrarily defined as 4 rtt samples. * 4 samples is enough for the srtt filter to converge * to within enough % of the correct value; fewer samples * and we could save a bogus rtt. The danger is not high * as tcp quickly recovers from everything. * XXX: Works very well but needs some more statistics! * * XXXRRS: Updating must be after the stack fini() since * that may be converting some internal representation of * say srtt etc into the general one used by other stacks. * Lets also at least protect against the so being NULL * as RW stated below. */ if ((tp->t_rttupdated >= 4) && (so != NULL)) { struct hc_metrics_lite metrics; uint32_t ssthresh; bzero(&metrics, sizeof(metrics)); /* * Update the ssthresh always when the conditions below * are satisfied. This gives us better new start value * for the congestion avoidance for new connections. * ssthresh is only set if packet loss occurred on a session. * * XXXRW: 'so' may be NULL here, and/or socket buffer may be * being torn down. Ideally this code would not use 'so'. */ ssthresh = tp->snd_ssthresh; if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { /* * convert the limit from user data bytes to * packets then to packet data bytes. */ ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; if (ssthresh < 2) ssthresh = 2; ssthresh *= (tp->t_maxseg + #ifdef INET6 (isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : #endif sizeof (struct tcpiphdr) #ifdef INET6 ) #endif ); } else ssthresh = 0; metrics.rmx_ssthresh = ssthresh; metrics.rmx_rtt = tp->t_srtt; metrics.rmx_rttvar = tp->t_rttvar; metrics.rmx_cwnd = tp->snd_cwnd; metrics.rmx_sendpipe = 0; metrics.rmx_recvpipe = 0; tcp_hc_update(&inp->inp_inc, &metrics); } refcount_release(&tp->t_fb->tfb_refcnt); uma_zfree(V_tcpcb_zone, tp); return (in_pcbrele_wlocked(inp)); } /* * Attempt to close a TCP control block, marking it as dropped, and freeing * the socket if we hold the only reference. */ struct tcpcb * tcp_close(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; struct socket *so; INP_WLOCK_ASSERT(inp); #ifdef TCP_OFFLOAD if (tp->t_state == TCPS_LISTEN) tcp_offload_listen_stop(tp); #endif /* * This releases the TFO pending counter resource for TFO listen * sockets as well as passively-created TFO sockets that transition * from SYN_RECEIVED to CLOSED. */ if (tp->t_tfo_pending) { tcp_fastopen_decrement_counter(tp->t_tfo_pending); tp->t_tfo_pending = NULL; } #ifdef TCPHPTS tcp_hpts_remove(inp); #endif in_pcbdrop(inp); TCPSTAT_INC(tcps_closed); if (tp->t_state != TCPS_CLOSED) tcp_state_change(tp, TCPS_CLOSED); KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); so = inp->inp_socket; soisdisconnected(so); if (inp->inp_flags & INP_SOCKREF) { inp->inp_flags &= ~INP_SOCKREF; INP_WUNLOCK(inp); sorele(so); return (NULL); } return (tp); } /* * Notify a tcp user of an asynchronous error; * store error as soft error, but wake up user * (for now, won't do anything until can select for soft error). * * Do not wake up user since there currently is no mechanism for * reporting soft errors (yet - a kqueue filter may be added). */ static struct inpcb * tcp_notify(struct inpcb *inp, int error) { struct tcpcb *tp; INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return (inp); tp = intotcpcb(inp); KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); /* * Ignore some errors if we are hooked up. * If connection hasn't completed, has retransmitted several times, * and receives a second error, give up now. This is better * than waiting a long time to establish a connection that * can never complete. */ if (tp->t_state == TCPS_ESTABLISHED && (error == EHOSTUNREACH || error == ENETUNREACH || error == EHOSTDOWN)) { if (inp->inp_route.ro_nh) { NH_FREE(inp->inp_route.ro_nh); inp->inp_route.ro_nh = (struct nhop_object *)NULL; } return (inp); } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && tp->t_softerror) { tp = tcp_drop(tp, error); if (tp != NULL) return (inp); else return (NULL); } else { tp->t_softerror = error; return (inp); } #if 0 wakeup( &so->so_timeo); sorwakeup(so); sowwakeup(so); #endif } static int tcp_pcblist(SYSCTL_HANDLER_ARGS) { struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, INPLOOKUP_RLOCKPCB); struct xinpgen xig; struct inpcb *inp; int error; if (req->newptr != NULL) return (EPERM); if (req->oldptr == NULL) { int n; n = V_tcbinfo.ipi_count + counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); n += imax(n / 8, 10); req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); return (0); } if ((error = sysctl_wire_old_buffer(req, 0)) != 0) return (error); bzero(&xig, sizeof(xig)); xig.xig_len = sizeof xig; xig.xig_count = V_tcbinfo.ipi_count + counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); xig.xig_gen = V_tcbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) return (error); error = syncache_pcblist(req); if (error) return (error); while ((inp = inp_next(&inpi)) != NULL) { if (inp->inp_gencnt <= xig.xig_gen) { int crerr; /* * XXX: This use of cr_cansee(), introduced with * TCP state changes, is not quite right, but for * now, better than nothing. */ if (inp->inp_flags & INP_TIMEWAIT) { if (intotw(inp) != NULL) crerr = cr_cansee(req->td->td_ucred, intotw(inp)->tw_cred); else crerr = EINVAL; /* Skip this inp. */ } else crerr = cr_canseeinpcb(req->td->td_ucred, inp); if (crerr == 0) { struct xtcpcb xt; tcp_inptoxtp(inp, &xt); error = SYSCTL_OUT(req, &xt, sizeof xt); if (error) { INP_RUNLOCK(inp); break; } else continue; } } } if (!error) { /* * Give the user an updated idea of our state. * If the generation differs from what we told * her before, she knows that something happened * while we were processing this request, and it * might be necessary to retry. */ xig.xig_gen = V_tcbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; xig.xig_count = V_tcbinfo.ipi_count + counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); error = SYSCTL_OUT(req, &xig, sizeof xig); } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, NULL, 0, tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); #ifdef INET static int tcp_getcred(SYSCTL_HANDLER_ARGS) { struct xucred xuc; struct sockaddr_in addrs[2]; struct epoch_tracker et; struct inpcb *inp; int error; error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); NET_EPOCH_ENTER(et); inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); NET_EPOCH_EXIT(et); if (inp != NULL) { if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 0, 0, tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); #endif /* INET */ #ifdef INET6 static int tcp6_getcred(SYSCTL_HANDLER_ARGS) { struct epoch_tracker et; struct xucred xuc; struct sockaddr_in6 addrs[2]; struct inpcb *inp; int error; #ifdef INET int mapped = 0; #endif error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { return (error); } if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { #ifdef INET if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) mapped = 1; else #endif return (EINVAL); } NET_EPOCH_ENTER(et); #ifdef INET if (mapped == 1) inp = in_pcblookup(&V_tcbinfo, *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], addrs[1].sin6_port, *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); else #endif inp = in6_pcblookup(&V_tcbinfo, &addrs[1].sin6_addr, addrs[1].sin6_port, &addrs[0].sin6_addr, addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); NET_EPOCH_EXIT(et); if (inp != NULL) { if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 0, 0, tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); #endif /* INET6 */ #ifdef INET /* Path MTU to try next when a fragmentation-needed message is received. */ static inline int tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) { int mtu = ntohs(icp->icmp_nextmtu); /* If no alternative MTU was proposed, try the next smaller one. */ if (!mtu) mtu = ip_next_mtu(ntohs(ip->ip_len), 1); if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) mtu = V_tcp_minmss + sizeof(struct tcpiphdr); return (mtu); } static void tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port) { struct ip *ip = vip; struct tcphdr *th; struct in_addr faddr; struct inpcb *inp; struct tcpcb *tp; struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; struct icmp *icp; struct in_conninfo inc; tcp_seq icmp_tcp_seq; int mtu; faddr = ((struct sockaddr_in *)sa)->sin_addr; if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) return; if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc_notify; else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || cmd == PRC_TIMXCEED_INTRANS) && ip) notify = tcp_drop_syn_sent; /* * Hostdead is ugly because it goes linearly through all PCBs. * XXX: We never get this from ICMP, otherwise it makes an * excellent DoS attack on machines with many connections. */ else if (cmd == PRC_HOSTDEAD) ip = NULL; else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) return; if (ip == NULL) { in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); return; } icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL); if (inp != NULL && PRC_IS_REDIRECT(cmd)) { /* signal EHOSTDOWN, as it flushes the cached route */ inp = (*notify)(inp, EHOSTDOWN); goto out; } icmp_tcp_seq = th->th_seq; if (inp != NULL) { if (!(inp->inp_flags & INP_TIMEWAIT) && !(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket == NULL)) { tp = intotcpcb(inp); #ifdef TCP_OFFLOAD if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) { /* * MTU discovery for offloaded connections. Let * the TOE driver verify seq# and process it. */ mtu = tcp_next_pmtu(icp, ip); tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); goto out; } #endif if (tp->t_port != port) { goto out; } if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { if (cmd == PRC_MSGSIZE) { /* * MTU discovery: we got a needfrag and * will potentially try a lower MTU. */ mtu = tcp_next_pmtu(icp, ip); /* * Only process the offered MTU if it * is smaller than the current one. */ if (mtu < tp->t_maxseg + sizeof(struct tcpiphdr)) { bzero(&inc, sizeof(inc)); inc.inc_faddr = faddr; inc.inc_fibnum = inp->inp_inc.inc_fibnum; tcp_hc_updatemtu(&inc, mtu); inp = tcp_mtudisc(inp, mtu); } } else inp = (*notify)(inp, inetctlerrmap[cmd]); } } } else { bzero(&inc, sizeof(inc)); inc.inc_fport = th->th_dport; inc.inc_lport = th->th_sport; inc.inc_faddr = faddr; inc.inc_laddr = ip->ip_src; syncache_unreach(&inc, icmp_tcp_seq, port); } out: if (inp != NULL) INP_WUNLOCK(inp); } void tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) { tcp_ctlinput_with_port(cmd, sa, vip, htons(0)); } void tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused) { /* Its a tunneled TCP over UDP icmp */ struct ip *outer_ip, *inner_ip; struct icmp *icmp; struct udphdr *udp; struct tcphdr *th, ttemp; int i_hlen, o_len; uint16_t port; inner_ip = (struct ip *)vip; icmp = (struct icmp *)((caddr_t)inner_ip - (sizeof(struct icmp) - sizeof(struct ip))); outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); i_hlen = inner_ip->ip_hl << 2; o_len = ntohs(outer_ip->ip_len); if (o_len < (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { /* Not enough data present */ return; } /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { return; } port = udp->uh_dport; th = (struct tcphdr *)(udp + 1); memcpy(&ttemp, th, sizeof(struct tcphdr)); memcpy(udp, &ttemp, sizeof(struct tcphdr)); /* Now adjust down the size of the outer IP header */ o_len -= sizeof(struct udphdr); outer_ip->ip_len = htons(o_len); /* Now call in to the normal handling code */ tcp_ctlinput_with_port(cmd, sa, vip, port); } #endif /* INET */ #ifdef INET6 static inline int tcp6_next_pmtu(const struct icmp6_hdr *icmp6) { int mtu = ntohl(icmp6->icmp6_mtu); /* * If no alternative MTU was proposed, or the proposed MTU was too * small, set to the min. */ if (mtu < IPV6_MMTU) mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */ return (mtu); } static void tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port) { struct in6_addr *dst; struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; struct ip6_hdr *ip6; struct mbuf *m; struct inpcb *inp; struct tcpcb *tp; struct icmp6_hdr *icmp6; struct ip6ctlparam *ip6cp = NULL; const struct sockaddr_in6 *sa6_src = NULL; struct in_conninfo inc; struct tcp_ports { uint16_t th_sport; uint16_t th_dport; } t_ports; tcp_seq icmp_tcp_seq; unsigned int mtu; unsigned int off; if (sa->sa_family != AF_INET6 || sa->sa_len != sizeof(struct sockaddr_in6)) return; /* if the parameter is from icmp6, decode it. */ if (d != NULL) { ip6cp = (struct ip6ctlparam *)d; icmp6 = ip6cp->ip6c_icmp6; m = ip6cp->ip6c_m; ip6 = ip6cp->ip6c_ip6; off = ip6cp->ip6c_off; sa6_src = ip6cp->ip6c_src; dst = ip6cp->ip6c_finaldst; } else { m = NULL; ip6 = NULL; off = 0; /* fool gcc */ sa6_src = &sa6_any; dst = NULL; } if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc_notify; else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) notify = tcp_drop_syn_sent; /* * Hostdead is ugly because it goes linearly through all PCBs. * XXX: We never get this from ICMP, otherwise it makes an * excellent DoS attack on machines with many connections. */ else if (cmd == PRC_HOSTDEAD) ip6 = NULL; else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) return; if (ip6 == NULL) { in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); return; } /* Check if we can safely get the ports from the tcp hdr */ if (m == NULL || (m->m_pkthdr.len < (int32_t) (off + sizeof(struct tcp_ports)))) { return; } bzero(&t_ports, sizeof(struct tcp_ports)); m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); if (inp != NULL && PRC_IS_REDIRECT(cmd)) { /* signal EHOSTDOWN, as it flushes the cached route */ inp = (*notify)(inp, EHOSTDOWN); goto out; } off += sizeof(struct tcp_ports); if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { goto out; } m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); if (inp != NULL) { if (!(inp->inp_flags & INP_TIMEWAIT) && !(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket == NULL)) { tp = intotcpcb(inp); #ifdef TCP_OFFLOAD if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) { /* MTU discovery for offloaded connections. */ mtu = tcp6_next_pmtu(icmp6); tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); goto out; } #endif if (tp->t_port != port) { goto out; } if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { if (cmd == PRC_MSGSIZE) { /* * MTU discovery: * If we got a needfrag set the MTU * in the route to the suggested new * value (if given) and then notify. */ mtu = tcp6_next_pmtu(icmp6); bzero(&inc, sizeof(inc)); inc.inc_fibnum = M_GETFIB(m); inc.inc_flags |= INC_ISIPV6; inc.inc6_faddr = *dst; if (in6_setscope(&inc.inc6_faddr, m->m_pkthdr.rcvif, NULL)) goto out; /* * Only process the offered MTU if it * is smaller than the current one. */ if (mtu < tp->t_maxseg + sizeof (struct tcphdr) + sizeof (struct ip6_hdr)) { tcp_hc_updatemtu(&inc, mtu); tcp_mtudisc(inp, mtu); ICMP6STAT_INC(icp6s_pmtuchg); } } else inp = (*notify)(inp, inet6ctlerrmap[cmd]); } } } else { bzero(&inc, sizeof(inc)); inc.inc_fibnum = M_GETFIB(m); inc.inc_flags |= INC_ISIPV6; inc.inc_fport = t_ports.th_dport; inc.inc_lport = t_ports.th_sport; inc.inc6_faddr = *dst; inc.inc6_laddr = ip6->ip6_src; syncache_unreach(&inc, icmp_tcp_seq, port); } out: if (inp != NULL) INP_WUNLOCK(inp); } void tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) { tcp6_ctlinput_with_port(cmd, sa, d, htons(0)); } void tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused) { struct ip6ctlparam *ip6cp; struct mbuf *m; struct udphdr *udp; uint16_t port; ip6cp = (struct ip6ctlparam *)d; m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); if (m == NULL) { return; } udp = mtod(m, struct udphdr *); if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { return; } port = udp->uh_dport; m_adj(m, sizeof(struct udphdr)); if ((m->m_flags & M_PKTHDR) == 0) { ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); } /* Now call in to the normal handling code */ tcp6_ctlinput_with_port(cmd, sa, d, port); } #endif /* INET6 */ static uint32_t tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) { SIPHASH_CTX ctx; uint32_t hash[2]; KASSERT(len >= SIPHASH_KEY_LENGTH, ("%s: keylen %u too short ", __func__, len)); SipHash24_Init(&ctx); SipHash_SetKey(&ctx, (uint8_t *)key); SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); switch (inc->inc_flags & INC_ISIPV6) { #ifdef INET case 0: SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); break; #endif #ifdef INET6 case INC_ISIPV6: SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); break; #endif } SipHash_Final((uint8_t *)hash, &ctx); return (hash[0] ^ hash[1]); } uint32_t tcp_new_ts_offset(struct in_conninfo *inc) { struct in_conninfo inc_store, *local_inc; if (!V_tcp_ts_offset_per_conn) { memcpy(&inc_store, inc, sizeof(struct in_conninfo)); inc_store.inc_lport = 0; inc_store.inc_fport = 0; local_inc = &inc_store; } else { local_inc = inc; } return (tcp_keyed_hash(local_inc, V_ts_offset_secret, sizeof(V_ts_offset_secret))); } /* * Following is where TCP initial sequence number generation occurs. * * There are two places where we must use initial sequence numbers: * 1. In SYN-ACK packets. * 2. In SYN packets. * * All ISNs for SYN-ACK packets are generated by the syncache. See * tcp_syncache.c for details. * * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling * depends on this property. In addition, these ISNs should be * unguessable so as to prevent connection hijacking. To satisfy * the requirements of this situation, the algorithm outlined in * RFC 1948 is used, with only small modifications. * * Implementation details: * * Time is based off the system timer, and is corrected so that it * increases by one megabyte per second. This allows for proper * recycling on high speed LANs while still leaving over an hour * before rollover. * * As reading the *exact* system time is too expensive to be done * whenever setting up a TCP connection, we increment the time * offset in two ways. First, a small random positive increment * is added to isn_offset for each connection that is set up. * Second, the function tcp_isn_tick fires once per clock tick * and increments isn_offset as necessary so that sequence numbers * are incremented at approximately ISN_BYTES_PER_SECOND. The * random positive increments serve only to ensure that the same * exact sequence number is never sent out twice (as could otherwise * happen when a port is recycled in less than the system tick * interval.) * * net.inet.tcp.isn_reseed_interval controls the number of seconds * between seeding of isn_secret. This is normally set to zero, * as reseeding should not be necessary. * * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, * isn_offset_old, and isn_ctx is performed using the ISN lock. In * general, this means holding an exclusive (write) lock. */ #define ISN_BYTES_PER_SECOND 1048576 #define ISN_STATIC_INCREMENT 4096 #define ISN_RANDOM_INCREMENT (4096 - 1) #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); VNET_DEFINE_STATIC(int, isn_last); VNET_DEFINE_STATIC(int, isn_last_reseed); VNET_DEFINE_STATIC(u_int32_t, isn_offset); VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); #define V_isn_secret VNET(isn_secret) #define V_isn_last VNET(isn_last) #define V_isn_last_reseed VNET(isn_last_reseed) #define V_isn_offset VNET(isn_offset) #define V_isn_offset_old VNET(isn_offset_old) tcp_seq tcp_new_isn(struct in_conninfo *inc) { tcp_seq new_isn; u_int32_t projected_offset; ISN_LOCK(); /* Seed if this is the first use, reseed if requested. */ if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) < (u_int)ticks))) { arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); V_isn_last_reseed = ticks; } /* Compute the hash and return the ISN. */ new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, sizeof(V_isn_secret)); V_isn_offset += ISN_STATIC_INCREMENT + (arc4random() & ISN_RANDOM_INCREMENT); if (ticks != V_isn_last) { projected_offset = V_isn_offset_old + ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); if (SEQ_GT(projected_offset, V_isn_offset)) V_isn_offset = projected_offset; V_isn_offset_old = V_isn_offset; V_isn_last = ticks; } new_isn += V_isn_offset; ISN_UNLOCK(); return (new_isn); } /* * When a specific ICMP unreachable message is received and the * connection state is SYN-SENT, drop the connection. This behavior * is controlled by the icmp_may_rst sysctl. */ struct inpcb * tcp_drop_syn_sent(struct inpcb *inp, int errno) { struct tcpcb *tp; NET_EPOCH_ASSERT(); INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return (inp); tp = intotcpcb(inp); if (tp->t_state != TCPS_SYN_SENT) return (inp); if (IS_FASTOPEN(tp->t_flags)) tcp_fastopen_disable_path(tp); tp = tcp_drop(tp, errno); if (tp != NULL) return (inp); else return (NULL); } /* * When `need fragmentation' ICMP is received, update our idea of the MSS * based on the new value. Also nudge TCP to send something, since we * know the packet we just sent was dropped. * This duplicates some code in the tcp_mss() function in tcp_input.c. */ static struct inpcb * tcp_mtudisc_notify(struct inpcb *inp, int error) { return (tcp_mtudisc(inp, -1)); } static struct inpcb * tcp_mtudisc(struct inpcb *inp, int mtuoffer) { struct tcpcb *tp; struct socket *so; INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return (inp); tp = intotcpcb(inp); KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); so = inp->inp_socket; SOCKBUF_LOCK(&so->so_snd); /* If the mss is larger than the socket buffer, decrease the mss. */ if (so->so_snd.sb_hiwat < tp->t_maxseg) tp->t_maxseg = so->so_snd.sb_hiwat; SOCKBUF_UNLOCK(&so->so_snd); TCPSTAT_INC(tcps_mturesent); tp->t_rtttime = 0; tp->snd_nxt = tp->snd_una; tcp_free_sackholes(tp); tp->snd_recover = tp->snd_max; if (tp->t_flags & TF_SACK_PERMIT) EXIT_FASTRECOVERY(tp->t_flags); if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { /* * Conceptually the snd_nxt setting * and freeing sack holes should * be done by the default stacks * own tfb_tcp_mtu_chg(). */ tp->t_fb->tfb_tcp_mtu_chg(tp); } if (tcp_output(tp) < 0) return (NULL); else return (inp); } #ifdef INET /* * Look-up the routing entry to the peer of this inpcb. If no route * is found and it cannot be allocated, then return 0. This routine * is called by TCP routines that access the rmx structure and by * tcp_mss_update to get the peer/interface MTU. */ uint32_t tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) { struct nhop_object *nh; struct ifnet *ifp; uint32_t maxmtu = 0; KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); if (inc->inc_faddr.s_addr != INADDR_ANY) { nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); if (nh == NULL) return (0); ifp = nh->nh_ifp; maxmtu = nh->nh_mtu; /* Report additional interface capabilities. */ if (cap != NULL) { if (ifp->if_capenable & IFCAP_TSO4 && ifp->if_hwassist & CSUM_TSO) { cap->ifcap |= CSUM_TSO; cap->tsomax = ifp->if_hw_tsomax; cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; } } } return (maxmtu); } #endif /* INET */ #ifdef INET6 uint32_t tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) { struct nhop_object *nh; struct in6_addr dst6; uint32_t scopeid; struct ifnet *ifp; uint32_t maxmtu = 0; KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); if (inc->inc_flags & INC_IPV6MINMTU) return (IPV6_MMTU); if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); if (nh == NULL) return (0); ifp = nh->nh_ifp; maxmtu = nh->nh_mtu; /* Report additional interface capabilities. */ if (cap != NULL) { if (ifp->if_capenable & IFCAP_TSO6 && ifp->if_hwassist & CSUM_TSO) { cap->ifcap |= CSUM_TSO; cap->tsomax = ifp->if_hw_tsomax; cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; } } } return (maxmtu); } /* * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. * * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. * The right place to do that is ip6_setpktopt() that has just been * executed. By the way it just filled ip6po_minmtu for us. */ void tcp6_use_min_mtu(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; INP_WLOCK_ASSERT(inp); /* * In case of the IPV6_USE_MIN_MTU socket * option, the INC_IPV6MINMTU flag to announce * a corresponding MSS during the initial * handshake. If the TCP connection is not in * the front states, just reduce the MSS being * used. This avoids the sending of TCP * segments which will be fragmented at the * IPv6 layer. */ inp->inp_inc.inc_flags |= INC_IPV6MINMTU; if ((tp->t_state >= TCPS_SYN_SENT) && (inp->inp_inc.inc_flags & INC_ISIPV6)) { struct ip6_pktopts *opt; opt = inp->in6p_outputopts; if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && tp->t_maxseg > TCP6_MSS) tp->t_maxseg = TCP6_MSS; } } #endif /* INET6 */ /* * Calculate effective SMSS per RFC5681 definition for a given TCP * connection at its current state, taking into account SACK and etc. */ u_int tcp_maxseg(const struct tcpcb *tp) { u_int optlen; if (tp->t_flags & TF_NOOPT) return (tp->t_maxseg); /* * Here we have a simplified code from tcp_addoptions(), * without a proper loop, and having most of paddings hardcoded. * We might make mistakes with padding here in some edge cases, * but this is harmless, since result of tcp_maxseg() is used * only in cwnd and ssthresh estimations. */ if (TCPS_HAVEESTABLISHED(tp->t_state)) { if (tp->t_flags & TF_RCVD_TSTMP) optlen = TCPOLEN_TSTAMP_APPA; else optlen = 0; #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); #endif if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { optlen += TCPOLEN_SACKHDR; optlen += tp->rcv_numsacks * TCPOLEN_SACK; optlen = PADTCPOLEN(optlen); } } else { if (tp->t_flags & TF_REQ_TSTMP) optlen = TCPOLEN_TSTAMP_APPA; else optlen = PADTCPOLEN(TCPOLEN_MAXSEG); if (tp->t_flags & TF_REQ_SCALE) optlen += PADTCPOLEN(TCPOLEN_WINDOW); #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); #endif if (tp->t_flags & TF_SACK_PERMIT) optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); } #undef PAD optlen = min(optlen, TCP_MAXOLEN); return (tp->t_maxseg - optlen); } u_int tcp_fixed_maxseg(const struct tcpcb *tp) { int optlen; if (tp->t_flags & TF_NOOPT) return (tp->t_maxseg); /* * Here we have a simplified code from tcp_addoptions(), * without a proper loop, and having most of paddings hardcoded. * We only consider fixed options that we would send every * time I.e. SACK is not considered. This is important * for cc modules to figure out what the modulo of the * cwnd should be. */ #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) if (TCPS_HAVEESTABLISHED(tp->t_state)) { if (tp->t_flags & TF_RCVD_TSTMP) optlen = TCPOLEN_TSTAMP_APPA; else optlen = 0; #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) optlen += PAD(TCPOLEN_SIGNATURE); #endif } else { if (tp->t_flags & TF_REQ_TSTMP) optlen = TCPOLEN_TSTAMP_APPA; else optlen = PAD(TCPOLEN_MAXSEG); if (tp->t_flags & TF_REQ_SCALE) optlen += PAD(TCPOLEN_WINDOW); #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) optlen += PAD(TCPOLEN_SIGNATURE); #endif if (tp->t_flags & TF_SACK_PERMIT) optlen += PAD(TCPOLEN_SACK_PERMITTED); } #undef PAD optlen = min(optlen, TCP_MAXOLEN); return (tp->t_maxseg - optlen); } static int sysctl_drop(SYSCTL_HANDLER_ARGS) { /* addrs[0] is a foreign socket, addrs[1] is a local one. */ struct sockaddr_storage addrs[2]; struct inpcb *inp; struct tcpcb *tp; struct tcptw *tw; #ifdef INET struct sockaddr_in *fin = NULL, *lin = NULL; #endif struct epoch_tracker et; #ifdef INET6 struct sockaddr_in6 *fin6, *lin6; #endif int error; inp = NULL; #ifdef INET6 fin6 = lin6 = NULL; #endif error = 0; if (req->oldptr != NULL || req->oldlen != 0) return (EINVAL); if (req->newptr == NULL) return (EPERM); if (req->newlen < sizeof(addrs)) return (ENOMEM); error = SYSCTL_IN(req, &addrs, sizeof(addrs)); if (error) return (error); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: fin6 = (struct sockaddr_in6 *)&addrs[0]; lin6 = (struct sockaddr_in6 *)&addrs[1]; if (fin6->sin6_len != sizeof(struct sockaddr_in6) || lin6->sin6_len != sizeof(struct sockaddr_in6)) return (EINVAL); if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) return (EINVAL); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); #ifdef INET fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; #endif break; } error = sa6_embedscope(fin6, V_ip6_use_defzone); if (error) return (error); error = sa6_embedscope(lin6, V_ip6_use_defzone); if (error) return (error); break; #endif #ifdef INET case AF_INET: fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; if (fin->sin_len != sizeof(struct sockaddr_in) || lin->sin_len != sizeof(struct sockaddr_in)) return (EINVAL); break; #endif default: return (EINVAL); } NET_EPOCH_ENTER(et); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif #ifdef INET case AF_INET: inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif } if (inp != NULL) { if (inp->inp_flags & INP_TIMEWAIT) { /* * XXXRW: There currently exists a state where an * inpcb is present, but its timewait state has been * discarded. For now, don't allow dropping of this * type of inpcb. */ tw = intotw(inp); if (tw != NULL) tcp_twclose(tw, 0); else INP_WUNLOCK(inp); } else if ((inp->inp_flags & INP_DROPPED) == 0 && !SOLISTENING(inp->inp_socket)) { tp = intotcpcb(inp); tp = tcp_drop(tp, ECONNABORTED); if (tp != NULL) INP_WUNLOCK(inp); } else INP_WUNLOCK(inp); } else error = ESRCH; NET_EPOCH_EXIT(et); return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", "Drop TCP connection"); static int tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS) { return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo, &tcp_ctloutput_set)); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "", "Set socket option for TCP endpoint"); #ifdef KERN_TLS static int sysctl_switch_tls(SYSCTL_HANDLER_ARGS) { /* addrs[0] is a foreign socket, addrs[1] is a local one. */ struct sockaddr_storage addrs[2]; struct inpcb *inp; #ifdef INET struct sockaddr_in *fin = NULL, *lin = NULL; #endif struct epoch_tracker et; #ifdef INET6 struct sockaddr_in6 *fin6, *lin6; #endif int error; inp = NULL; #ifdef INET6 fin6 = lin6 = NULL; #endif error = 0; if (req->oldptr != NULL || req->oldlen != 0) return (EINVAL); if (req->newptr == NULL) return (EPERM); if (req->newlen < sizeof(addrs)) return (ENOMEM); error = SYSCTL_IN(req, &addrs, sizeof(addrs)); if (error) return (error); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: fin6 = (struct sockaddr_in6 *)&addrs[0]; lin6 = (struct sockaddr_in6 *)&addrs[1]; if (fin6->sin6_len != sizeof(struct sockaddr_in6) || lin6->sin6_len != sizeof(struct sockaddr_in6)) return (EINVAL); if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) return (EINVAL); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); #ifdef INET fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; #endif break; } error = sa6_embedscope(fin6, V_ip6_use_defzone); if (error) return (error); error = sa6_embedscope(lin6, V_ip6_use_defzone); if (error) return (error); break; #endif #ifdef INET case AF_INET: fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; if (fin->sin_len != sizeof(struct sockaddr_in) || lin->sin_len != sizeof(struct sockaddr_in)) return (EINVAL); break; #endif default: return (EINVAL); } NET_EPOCH_ENTER(et); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif #ifdef INET case AF_INET: inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif } NET_EPOCH_EXIT(et); if (inp != NULL) { if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 || inp->inp_socket == NULL) { error = ECONNRESET; INP_WUNLOCK(inp); } else { struct socket *so; so = inp->inp_socket; soref(so); error = ktls_set_tx_mode(so, arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); INP_WUNLOCK(inp); sorele(so); } } else error = ESRCH; return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", "Switch TCP connection to SW TLS"); SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", "Switch TCP connection to ifnet TLS"); #endif /* * Generate a standardized TCP log line for use throughout the * tcp subsystem. Memory allocation is done with M_NOWAIT to * allow use in the interrupt context. * * NB: The caller MUST free(s, M_TCPLOG) the returned string. * NB: The function may return NULL if memory allocation failed. * * Due to header inclusion and ordering limitations the struct ip * and ip6_hdr pointers have to be passed as void pointers. */ char * tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, const void *ip6hdr) { /* Is logging enabled? */ if (V_tcp_log_in_vain == 0) return (NULL); return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); } char * tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, const void *ip6hdr) { /* Is logging enabled? */ if (tcp_log_debug == 0) return (NULL); return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); } static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, const void *ip6hdr) { char *s, *sp; size_t size; #ifdef INET const struct ip *ip = (const struct ip *)ip4hdr; #endif #ifdef INET6 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr; #endif /* INET6 */ /* * The log line looks like this: * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2" */ size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + sizeof(PRINT_TH_FLAGS) + 1 + #ifdef INET6 2 * INET6_ADDRSTRLEN; #else 2 * INET_ADDRSTRLEN; #endif /* INET6 */ s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); if (s == NULL) return (NULL); strcat(s, "TCP: ["); sp = s + strlen(s); if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { inet_ntoa_r(inc->inc_faddr, sp); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); sp = s + strlen(s); inet_ntoa_r(inc->inc_laddr, sp); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(inc->inc_lport)); #ifdef INET6 } else if (inc) { ip6_sprintf(sp, &inc->inc6_faddr); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); sp = s + strlen(s); ip6_sprintf(sp, &inc->inc6_laddr); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(inc->inc_lport)); } else if (ip6 && th) { ip6_sprintf(sp, &ip6->ip6_src); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(th->th_sport)); sp = s + strlen(s); ip6_sprintf(sp, &ip6->ip6_dst); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(th->th_dport)); #endif /* INET6 */ #ifdef INET } else if (ip && th) { inet_ntoa_r(ip->ip_src, sp); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(th->th_sport)); sp = s + strlen(s); inet_ntoa_r(ip->ip_dst, sp); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(th->th_dport)); #endif /* INET */ } else { free(s, M_TCPLOG); return (NULL); } sp = s + strlen(s); if (th) sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS); if (*(s + size - 1) != '\0') panic("%s: string too long", __func__); return (s); } /* * A subroutine which makes it easy to track TCP state changes with DTrace. * This function shouldn't be called for t_state initializations that don't * correspond to actual TCP state transitions. */ void tcp_state_change(struct tcpcb *tp, int newstate) { #if defined(KDTRACE_HOOKS) int pstate = tp->t_state; #endif TCPSTATES_DEC(tp->t_state); TCPSTATES_INC(newstate); tp->t_state = newstate; TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); } /* * Create an external-format (``xtcpcb'') structure using the information in * the kernel-format tcpcb structure pointed to by tp. This is done to * reduce the spew of irrelevant information over this interface, to isolate * user code from changes in the kernel structure, and potentially to provide * information-hiding if we decide that some of this information should be * hidden from users. */ void tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) { struct tcpcb *tp = intotcpcb(inp); struct tcptw *tw = intotw(inp); sbintime_t now; bzero(xt, sizeof(*xt)); if (inp->inp_flags & INP_TIMEWAIT) { xt->t_state = TCPS_TIME_WAIT; xt->xt_encaps_port = tw->t_port; } else { xt->t_state = tp->t_state; xt->t_logstate = tp->t_logstate; xt->t_flags = tp->t_flags; xt->t_sndzerowin = tp->t_sndzerowin; xt->t_sndrexmitpack = tp->t_sndrexmitpack; xt->t_rcvoopack = tp->t_rcvoopack; xt->t_rcv_wnd = tp->rcv_wnd; xt->t_snd_wnd = tp->snd_wnd; xt->t_snd_cwnd = tp->snd_cwnd; xt->t_snd_ssthresh = tp->snd_ssthresh; xt->t_dsack_bytes = tp->t_dsack_bytes; xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; xt->t_dsack_pack = tp->t_dsack_pack; xt->t_maxseg = tp->t_maxseg; xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; now = getsbinuptime(); #define COPYTIMER(ttt) do { \ if (callout_active(&tp->t_timers->ttt)) \ xt->ttt = (tp->t_timers->ttt.c_time - now) / \ SBT_1MS; \ else \ xt->ttt = 0; \ } while (0) COPYTIMER(tt_delack); COPYTIMER(tt_rexmt); COPYTIMER(tt_persist); COPYTIMER(tt_keep); COPYTIMER(tt_2msl); #undef COPYTIMER xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; xt->xt_encaps_port = tp->t_port; bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, TCP_FUNCTION_NAME_LEN_MAX); bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX); #ifdef TCP_BLACKBOX (void)tcp_log_get_id(tp, xt->xt_logid); #endif } xt->xt_len = sizeof(struct xtcpcb); in_pcbtoxinpcb(inp, &xt->xt_inp); if (inp->inp_socket == NULL) xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; } void tcp_log_end_status(struct tcpcb *tp, uint8_t status) { uint32_t bit, i; if ((tp == NULL) || (status > TCP_EI_STATUS_MAX_VALUE) || (status == 0)) { /* Invalid */ return; } if (status > (sizeof(uint32_t) * 8)) { /* Should this be a KASSERT? */ return; } bit = 1U << (status - 1); if (bit & tp->t_end_info_status) { /* already logged */ return; } for (i = 0; i < TCP_END_BYTE_INFO; i++) { if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { tp->t_end_info_bytes[i] = status; tp->t_end_info_status |= bit; break; } } } int tcp_can_enable_pacing(void) { if ((tcp_pacing_limit == -1) || (tcp_pacing_limit > number_of_tcp_connections_pacing)) { atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); shadow_num_connections = number_of_tcp_connections_pacing; return (1); } else { return (0); } } static uint8_t tcp_pacing_warning = 0; void tcp_decrement_paced_conn(void) { uint32_t ret; ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); shadow_num_connections = number_of_tcp_connections_pacing; KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); if (ret == 0) { if (tcp_pacing_limit != -1) { printf("Warning all pacing is now disabled, count decrements invalidly!\n"); tcp_pacing_limit = 0; } else if (tcp_pacing_warning == 0) { printf("Warning pacing count is invalid, invalid decrement\n"); tcp_pacing_warning = 1; } } } diff --git a/sys/sys/mbuf.h b/sys/sys/mbuf.h index 601b1e92c172..fa72314f78ed 100644 --- a/sys/sys/mbuf.h +++ b/sys/sys/mbuf.h @@ -1,1703 +1,1705 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1993 * The Regents of the University of California. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)mbuf.h 8.5 (Berkeley) 2/19/95 * $FreeBSD$ */ #ifndef _SYS_MBUF_H_ #define _SYS_MBUF_H_ /* XXX: These includes suck. Sorry! */ #include #ifdef _KERNEL #include #include #include #ifdef WITNESS #include #endif #endif #ifdef _KERNEL #include #define MBUF_PROBE1(probe, arg0) \ SDT_PROBE1(sdt, , , probe, arg0) #define MBUF_PROBE2(probe, arg0, arg1) \ SDT_PROBE2(sdt, , , probe, arg0, arg1) #define MBUF_PROBE3(probe, arg0, arg1, arg2) \ SDT_PROBE3(sdt, , , probe, arg0, arg1, arg2) #define MBUF_PROBE4(probe, arg0, arg1, arg2, arg3) \ SDT_PROBE4(sdt, , , probe, arg0, arg1, arg2, arg3) #define MBUF_PROBE5(probe, arg0, arg1, arg2, arg3, arg4) \ SDT_PROBE5(sdt, , , probe, arg0, arg1, arg2, arg3, arg4) SDT_PROBE_DECLARE(sdt, , , m__init); SDT_PROBE_DECLARE(sdt, , , m__gethdr_raw); SDT_PROBE_DECLARE(sdt, , , m__gethdr); SDT_PROBE_DECLARE(sdt, , , m__get_raw); SDT_PROBE_DECLARE(sdt, , , m__get); SDT_PROBE_DECLARE(sdt, , , m__getcl); SDT_PROBE_DECLARE(sdt, , , m__getjcl); SDT_PROBE_DECLARE(sdt, , , m__clget); SDT_PROBE_DECLARE(sdt, , , m__cljget); SDT_PROBE_DECLARE(sdt, , , m__cljset); SDT_PROBE_DECLARE(sdt, , , m__free); SDT_PROBE_DECLARE(sdt, , , m__freem); #endif /* _KERNEL */ /* * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead. * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in * sys/param.h), which has no additional overhead and is used instead of the * internal data area; this is done when at least MINCLSIZE of data must be * stored. Additionally, it is possible to allocate a separate buffer * externally and attach it to the mbuf in a way similar to that of mbuf * clusters. * * NB: These calculation do not take actual compiler-induced alignment and * padding inside the complete struct mbuf into account. Appropriate * attention is required when changing members of struct mbuf. * * MLEN is data length in a normal mbuf. * MHLEN is data length in an mbuf with pktheader. * MINCLSIZE is a smallest amount of data that should be put into cluster. * * Compile-time assertions in uipc_mbuf.c test these values to ensure that * they are sensible. */ struct mbuf; #define MHSIZE offsetof(struct mbuf, m_dat) #define MPKTHSIZE offsetof(struct mbuf, m_pktdat) #define MLEN ((int)(MSIZE - MHSIZE)) #define MHLEN ((int)(MSIZE - MPKTHSIZE)) #define MINCLSIZE (MHLEN + 1) #define M_NODOM 255 #ifdef _KERNEL /*- * Macro for type conversion: convert mbuf pointer to data pointer of correct * type: * * mtod(m, t) -- Convert mbuf pointer to data pointer of correct type. * mtodo(m, o) -- Same as above but with offset 'o' into data. */ #define mtod(m, t) ((t)((m)->m_data)) #define mtodo(m, o) ((void *)(((m)->m_data) + (o))) /* * Argument structure passed to UMA routines during mbuf and packet * allocations. */ struct mb_args { int flags; /* Flags for mbuf being allocated */ short type; /* Type of mbuf being allocated */ }; #endif /* _KERNEL */ /* * Packet tag structure (see below for details). */ struct m_tag { SLIST_ENTRY(m_tag) m_tag_link; /* List of packet tags */ u_int16_t m_tag_id; /* Tag ID */ u_int16_t m_tag_len; /* Length of data */ u_int32_t m_tag_cookie; /* ABI/Module ID */ void (*m_tag_free)(struct m_tag *); }; /* * Static network interface owned tag. * Allocated through ifp->if_snd_tag_alloc(). */ struct if_snd_tag_sw; struct m_snd_tag { struct ifnet *ifp; /* network interface tag belongs to */ const struct if_snd_tag_sw *sw; volatile u_int refcount; }; /* * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set. * Size ILP32: 56 * LP64: 64 * Compile-time assertions in uipc_mbuf.c test these values to ensure that * they are correct. */ struct pkthdr { union { struct m_snd_tag *snd_tag; /* send tag, if any */ struct ifnet *rcvif; /* rcv interface */ struct { uint16_t rcvidx; /* rcv interface index ... */ uint16_t rcvgen; /* ... and generation count */ }; }; union { struct ifnet *leaf_rcvif; /* leaf rcv interface */ struct { uint16_t leaf_rcvidx; /* leaf rcv interface index ... */ uint16_t leaf_rcvgen; /* ... and generation count */ }; }; SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */ int32_t len; /* total packet length */ /* Layer crossing persistent information. */ uint32_t flowid; /* packet's 4-tuple system */ uint32_t csum_flags; /* checksum and offload features */ uint16_t fibnum; /* this packet should use this fib */ uint8_t numa_domain; /* NUMA domain of recvd pkt */ uint8_t rsstype; /* hash type */ #if !defined(__LP64__) uint32_t pad; /* pad for 64bit alignment */ #endif union { uint64_t rcv_tstmp; /* timestamp in ns */ struct { uint8_t l2hlen; /* layer 2 hdr len */ uint8_t l3hlen; /* layer 3 hdr len */ uint8_t l4hlen; /* layer 4 hdr len */ uint8_t l5hlen; /* layer 5 hdr len */ uint8_t inner_l2hlen; uint8_t inner_l3hlen; uint8_t inner_l4hlen; uint8_t inner_l5hlen; }; }; union { uint8_t eight[8]; uint16_t sixteen[4]; uint32_t thirtytwo[2]; uint64_t sixtyfour[1]; uintptr_t unintptr[1]; void *ptr; } PH_per; /* Layer specific non-persistent local storage for reassembly, etc. */ union { union { uint8_t eight[8]; uint16_t sixteen[4]; uint32_t thirtytwo[2]; uint64_t sixtyfour[1]; uintptr_t unintptr[1]; void *ptr; } PH_loc; /* Upon allocation: total packet memory consumption. */ u_int memlen; }; }; #define ether_vtag PH_per.sixteen[0] #define tcp_tun_port PH_per.sixteen[0] /* outbound */ #define vt_nrecs PH_per.sixteen[0] /* mld and v6-ND */ #define tso_segsz PH_per.sixteen[1] /* inbound after LRO */ #define lro_nsegs tso_segsz /* inbound after LRO */ #define csum_data PH_per.thirtytwo[1] /* inbound from hardware up */ #define lro_tcp_d_len PH_loc.sixteen[0] /* inbound during LRO (no reassembly) */ #define lro_tcp_d_csum PH_loc.sixteen[1] /* inbound during LRO (no reassembly) */ #define lro_tcp_h_off PH_loc.sixteen[2] /* inbound during LRO (no reassembly) */ #define lro_etype PH_loc.sixteen[3] /* inbound during LRO (no reassembly) */ /* Note PH_loc is used during IP reassembly (all 8 bytes as a ptr) */ /* * TLS records for TLS 1.0-1.2 can have the following header lengths: * - 5 (AES-CBC with implicit IV) * - 21 (AES-CBC with explicit IV) * - 13 (AES-GCM with 8 byte explicit IV) */ #define MBUF_PEXT_HDR_LEN 23 /* * TLS records for TLS 1.0-1.2 can have the following maximum trailer * lengths: * - 16 (AES-GCM) * - 36 (AES-CBC with SHA1 and up to 16 bytes of padding) * - 48 (AES-CBC with SHA2-256 and up to 16 bytes of padding) * - 64 (AES-CBC with SHA2-384 and up to 16 bytes of padding) */ #define MBUF_PEXT_TRAIL_LEN 64 #if defined(__LP64__) #define MBUF_PEXT_MAX_PGS (40 / sizeof(vm_paddr_t)) #else #define MBUF_PEXT_MAX_PGS (64 / sizeof(vm_paddr_t)) #endif #define MBUF_PEXT_MAX_BYTES \ (MBUF_PEXT_MAX_PGS * PAGE_SIZE + MBUF_PEXT_HDR_LEN + MBUF_PEXT_TRAIL_LEN) struct ktls_session; struct socket; /* * Description of external storage mapped into mbuf; valid only if M_EXT is * set. * Size ILP32: 28 * LP64: 48 * Compile-time assertions in uipc_mbuf.c test these values to ensure that * they are correct. */ typedef void m_ext_free_t(struct mbuf *); struct m_ext { union { /* * If EXT_FLAG_EMBREF is set, then we use refcount in the * mbuf, the 'ext_count' member. Otherwise, we have a * shadow copy and we use pointer 'ext_cnt'. The original * mbuf is responsible to carry the pointer to free routine * and its arguments. They aren't copied into shadows in * mb_dupcl() to avoid dereferencing next cachelines. */ volatile u_int ext_count; volatile u_int *ext_cnt; }; uint32_t ext_size; /* size of buffer, for ext_free */ uint32_t ext_type:8, /* type of external storage */ ext_flags:24; /* external storage mbuf flags */ union { struct { /* * Regular M_EXT mbuf: * o ext_buf always points to the external buffer. * o ext_free (below) and two optional arguments * ext_arg1 and ext_arg2 store the free context for * the external storage. They are set only in the * refcount carrying mbuf, the one with * EXT_FLAG_EMBREF flag, with exclusion for * EXT_EXTREF type, where the free context is copied * into all mbufs that use same external storage. */ char *ext_buf; /* start of buffer */ #define m_ext_copylen offsetof(struct m_ext, ext_arg2) void *ext_arg2; }; struct { /* * Multi-page M_EXTPG mbuf: * o extpg_pa - page vector. * o extpg_trail and extpg_hdr - TLS trailer and * header. * Uses ext_free and may also use ext_arg1. */ vm_paddr_t extpg_pa[MBUF_PEXT_MAX_PGS]; char extpg_trail[MBUF_PEXT_TRAIL_LEN]; char extpg_hdr[MBUF_PEXT_HDR_LEN]; /* Pretend these 3 fields are part of mbuf itself. */ #define m_epg_pa m_ext.extpg_pa #define m_epg_trail m_ext.extpg_trail #define m_epg_hdr m_ext.extpg_hdr #define m_epg_ext_copylen offsetof(struct m_ext, ext_free) }; }; /* * Free method and optional argument pointer, both * used by M_EXT and M_EXTPG. */ m_ext_free_t *ext_free; void *ext_arg1; }; /* * The core of the mbuf object along with some shortcut defines for practical * purposes. */ struct mbuf { /* * Header present at the beginning of every mbuf. * Size ILP32: 24 * LP64: 32 * Compile-time assertions in uipc_mbuf.c test these values to ensure * that they are correct. */ union { /* next buffer in chain */ struct mbuf *m_next; SLIST_ENTRY(mbuf) m_slist; STAILQ_ENTRY(mbuf) m_stailq; }; union { /* next chain in queue/record */ struct mbuf *m_nextpkt; SLIST_ENTRY(mbuf) m_slistpkt; STAILQ_ENTRY(mbuf) m_stailqpkt; }; caddr_t m_data; /* location of data */ int32_t m_len; /* amount of data in this mbuf */ uint32_t m_type:8, /* type of data in this mbuf */ m_flags:24; /* flags; see below */ #if !defined(__LP64__) uint32_t m_pad; /* pad for 64bit alignment */ #endif /* * A set of optional headers (packet header, external storage header) * and internal data storage. Historically, these arrays were sized * to MHLEN (space left after a packet header) and MLEN (space left * after only a regular mbuf header); they are now variable size in * order to support future work on variable-size mbufs. */ union { struct { union { /* M_PKTHDR set. */ struct pkthdr m_pkthdr; /* M_EXTPG set. * Multi-page M_EXTPG mbuf has its meta data * split between the below anonymous structure * and m_ext. It carries vector of pages, * optional header and trailer char vectors * and pointers to socket/TLS data. */ #define m_epg_startcopy m_epg_npgs #define m_epg_endcopy m_epg_stailq struct { /* Overall count of pages and count of * pages with I/O pending. */ uint8_t m_epg_npgs; uint8_t m_epg_nrdy; /* TLS header and trailer lengths. * The data itself resides in m_ext. */ uint8_t m_epg_hdrlen; uint8_t m_epg_trllen; /* Offset into 1st page and length of * data in the last page. */ uint16_t m_epg_1st_off; uint16_t m_epg_last_len; uint8_t m_epg_flags; #define EPG_FLAG_ANON 0x1 /* Data can be encrypted in place. */ #define EPG_FLAG_2FREE 0x2 /* Scheduled for free. */ uint8_t m_epg_record_type; uint8_t __spare[2]; int m_epg_enc_cnt; struct ktls_session *m_epg_tls; struct socket *m_epg_so; uint64_t m_epg_seqno; STAILQ_ENTRY(mbuf) m_epg_stailq; }; }; union { /* M_EXT or M_EXTPG set. */ struct m_ext m_ext; /* M_PKTHDR set, neither M_EXT nor M_EXTPG. */ char m_pktdat[0]; }; }; char m_dat[0]; /* !M_PKTHDR, !M_EXT */ }; }; #ifdef _KERNEL static inline int m_epg_pagelen(const struct mbuf *m, int pidx, int pgoff) { KASSERT(pgoff == 0 || pidx == 0, ("page %d with non-zero offset %d in %p", pidx, pgoff, m)); if (pidx == m->m_epg_npgs - 1) { return (m->m_epg_last_len); } else { return (PAGE_SIZE - pgoff); } } #ifdef INVARIANTS #define MCHECK(ex, msg) KASSERT((ex), \ ("Multi page mbuf %p with " #msg " at %s:%d", \ m, __FILE__, __LINE__)) /* * NB: This expects a non-empty buffer (npgs > 0 and * last_pg_len > 0). */ #define MBUF_EXT_PGS_ASSERT_SANITY(m) do { \ MCHECK(m->m_epg_npgs > 0, "no valid pages"); \ MCHECK(m->m_epg_npgs <= nitems(m->m_epg_pa), \ "too many pages"); \ MCHECK(m->m_epg_nrdy <= m->m_epg_npgs, \ "too many ready pages"); \ MCHECK(m->m_epg_1st_off < PAGE_SIZE, \ "too large page offset"); \ MCHECK(m->m_epg_last_len > 0, "zero last page length"); \ MCHECK(m->m_epg_last_len <= PAGE_SIZE, \ "too large last page length"); \ if (m->m_epg_npgs == 1) \ MCHECK(m->m_epg_1st_off + \ m->m_epg_last_len <= PAGE_SIZE, \ "single page too large"); \ MCHECK(m->m_epg_hdrlen <= sizeof(m->m_epg_hdr), \ "too large header length"); \ MCHECK(m->m_epg_trllen <= sizeof(m->m_epg_trail), \ "too large header length"); \ } while (0) #else #define MBUF_EXT_PGS_ASSERT_SANITY(m) do {} while (0) #endif #endif /* * mbuf flags of global significance and layer crossing. * Those of only protocol/layer specific significance are to be mapped * to M_PROTO[1-11] and cleared at layer handoff boundaries. * NB: Limited to the lower 24 bits. */ #define M_EXT 0x00000001 /* has associated external storage */ #define M_PKTHDR 0x00000002 /* start of record */ #define M_EOR 0x00000004 /* end of record */ #define M_RDONLY 0x00000008 /* associated data is marked read-only */ #define M_BCAST 0x00000010 /* send/received as link-level broadcast */ #define M_MCAST 0x00000020 /* send/received as link-level multicast */ #define M_PROMISC 0x00000040 /* packet was not for us */ #define M_VLANTAG 0x00000080 /* ether_vtag is valid */ #define M_EXTPG 0x00000100 /* has array of unmapped pages and TLS */ #define M_NOFREE 0x00000200 /* do not free mbuf, embedded in cluster */ #define M_TSTMP 0x00000400 /* rcv_tstmp field is valid */ #define M_TSTMP_HPREC 0x00000800 /* rcv_tstmp is high-prec, typically hw-stamped on port (useful for IEEE 1588 and 802.1AS) */ #define M_TSTMP_LRO 0x00001000 /* Time LRO pushed in pkt is valid in (PH_loc) */ #define M_PROTO1 0x00002000 /* protocol-specific */ #define M_PROTO2 0x00004000 /* protocol-specific */ #define M_PROTO3 0x00008000 /* protocol-specific */ #define M_PROTO4 0x00010000 /* protocol-specific */ #define M_PROTO5 0x00020000 /* protocol-specific */ #define M_PROTO6 0x00040000 /* protocol-specific */ #define M_PROTO7 0x00080000 /* protocol-specific */ #define M_PROTO8 0x00100000 /* protocol-specific */ #define M_PROTO9 0x00200000 /* protocol-specific */ #define M_PROTO10 0x00400000 /* protocol-specific */ #define M_PROTO11 0x00800000 /* protocol-specific */ /* * Flags to purge when crossing layers. */ #define M_PROTOFLAGS \ (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\ M_PROTO9|M_PROTO10|M_PROTO11) /* * Flags preserved when copying m_pkthdr. */ #define M_COPYFLAGS \ (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \ M_TSTMP_HPREC|M_TSTMP_LRO|M_PROTOFLAGS) /* * Flags preserved during demote. */ #define M_DEMOTEFLAGS \ (M_EXT | M_RDONLY | M_NOFREE | M_EXTPG) /* * Mbuf flag description for use with printf(9) %b identifier. */ #define M_FLAG_BITS \ "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \ "\7M_PROMISC\10M_VLANTAG\11M_EXTPG\12M_NOFREE\13M_TSTMP\14M_TSTMP_HPREC\15M_TSTMP_LRO" #define M_FLAG_PROTOBITS \ "\16M_PROTO1\17M_PROTO2\20M_PROTO3\21M_PROTO4" \ "\22M_PROTO5\23M_PROTO6\24M_PROTO7\25M_PROTO8\26M_PROTO9" \ "\27M_PROTO10\28M_PROTO11" #define M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS) /* * Network interface cards are able to hash protocol fields (such as IPv4 * addresses and TCP port numbers) classify packets into flows. These flows * can then be used to maintain ordering while delivering packets to the OS * via parallel input queues, as well as to provide a stateless affinity * model. NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set * m_flag fields to indicate how the hash should be interpreted by the * network stack. * * Most NICs support RSS, which provides ordering and explicit affinity, and * use the hash m_flag bits to indicate what header fields were covered by * the hash. M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non- * RSS cards or configurations that provide an opaque flow identifier, allowing * for ordering and distribution without explicit affinity. Additionally, * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash * properties. * * The meaning of the IPV6_EX suffix: * "o Home address from the home address option in the IPv6 destination * options header. If the extension header is not present, use the Source * IPv6 Address. * o IPv6 address that is contained in the Routing-Header-Type-2 from the * associated extension header. If the extension header is not present, * use the Destination IPv6 Address." * Quoted from: * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex */ #define M_HASHTYPE_HASHPROP 0x80 /* has hash properties */ #define M_HASHTYPE_INNER 0x40 /* calculated from inner headers */ #define M_HASHTYPE_HASH(t) (M_HASHTYPE_HASHPROP | (t)) /* Microsoft RSS standard hash types */ #define M_HASHTYPE_NONE 0 #define M_HASHTYPE_RSS_IPV4 M_HASHTYPE_HASH(1) /* IPv4 2-tuple */ #define M_HASHTYPE_RSS_TCP_IPV4 M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */ #define M_HASHTYPE_RSS_IPV6 M_HASHTYPE_HASH(3) /* IPv6 2-tuple */ #define M_HASHTYPE_RSS_TCP_IPV6 M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */ #define M_HASHTYPE_RSS_IPV6_EX M_HASHTYPE_HASH(5) /* IPv6 2-tuple + * ext hdrs */ #define M_HASHTYPE_RSS_TCP_IPV6_EX M_HASHTYPE_HASH(6) /* TCPv6 4-tuple + * ext hdrs */ #define M_HASHTYPE_RSS_UDP_IPV4 M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/ #define M_HASHTYPE_RSS_UDP_IPV6 M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/ #define M_HASHTYPE_RSS_UDP_IPV6_EX M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple + * ext hdrs */ #define M_HASHTYPE_OPAQUE 0x3f /* ordering, not affinity */ #define M_HASHTYPE_OPAQUE_HASH M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE) /* ordering+hash, not affinity*/ #define M_HASHTYPE_CLEAR(m) ((m)->m_pkthdr.rsstype = 0) #define M_HASHTYPE_GET(m) ((m)->m_pkthdr.rsstype & ~M_HASHTYPE_INNER) #define M_HASHTYPE_SET(m, v) ((m)->m_pkthdr.rsstype = (v)) #define M_HASHTYPE_TEST(m, v) (M_HASHTYPE_GET(m) == (v)) #define M_HASHTYPE_ISHASH(m) \ (((m)->m_pkthdr.rsstype & M_HASHTYPE_HASHPROP) != 0) #define M_HASHTYPE_SETINNER(m) do { \ (m)->m_pkthdr.rsstype |= M_HASHTYPE_INNER; \ } while (0) /* * External mbuf storage buffer types. */ #define EXT_CLUSTER 1 /* mbuf cluster */ #define EXT_SFBUF 2 /* sendfile(2)'s sf_buf */ #define EXT_JUMBOP 3 /* jumbo cluster page sized */ #define EXT_JUMBO9 4 /* jumbo cluster 9216 bytes */ #define EXT_JUMBO16 5 /* jumbo cluster 16184 bytes */ #define EXT_PACKET 6 /* mbuf+cluster from packet zone */ #define EXT_MBUF 7 /* external mbuf reference */ #define EXT_RXRING 8 /* data in NIC receive ring */ #define EXT_VENDOR1 224 /* for vendor-internal use */ #define EXT_VENDOR2 225 /* for vendor-internal use */ #define EXT_VENDOR3 226 /* for vendor-internal use */ #define EXT_VENDOR4 227 /* for vendor-internal use */ #define EXT_EXP1 244 /* for experimental use */ #define EXT_EXP2 245 /* for experimental use */ #define EXT_EXP3 246 /* for experimental use */ #define EXT_EXP4 247 /* for experimental use */ #define EXT_NET_DRV 252 /* custom ext_buf provided by net driver(s) */ #define EXT_MOD_TYPE 253 /* custom module's ext_buf type */ #define EXT_DISPOSABLE 254 /* can throw this buffer away w/page flipping */ #define EXT_EXTREF 255 /* has externally maintained ext_cnt ptr */ /* * Flags for external mbuf buffer types. * NB: limited to the lower 24 bits. */ #define EXT_FLAG_EMBREF 0x000001 /* embedded ext_count */ #define EXT_FLAG_EXTREF 0x000002 /* external ext_cnt, notyet */ #define EXT_FLAG_NOFREE 0x000010 /* don't free mbuf to pool, notyet */ #define EXT_FLAG_VENDOR1 0x010000 /* These flags are vendor */ #define EXT_FLAG_VENDOR2 0x020000 /* or submodule specific, */ #define EXT_FLAG_VENDOR3 0x040000 /* not used by mbuf code. */ #define EXT_FLAG_VENDOR4 0x080000 /* Set/read by submodule. */ #define EXT_FLAG_EXP1 0x100000 /* for experimental use */ #define EXT_FLAG_EXP2 0x200000 /* for experimental use */ #define EXT_FLAG_EXP3 0x400000 /* for experimental use */ #define EXT_FLAG_EXP4 0x800000 /* for experimental use */ /* * EXT flag description for use with printf(9) %b identifier. */ #define EXT_FLAG_BITS \ "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \ "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \ "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \ "\30EXT_FLAG_EXP4" /* * Flags indicating checksum, segmentation and other offload work to be * done, or already done, by hardware or lower layers. It is split into * separate inbound and outbound flags. * * Outbound flags that are set by upper protocol layers requesting lower * layers, or ideally the hardware, to perform these offloading tasks. * For outbound packets this field and its flags can be directly tested * against ifnet if_hwassist. Note that the outbound and the inbound flags do * not collide right now but they could be allowed to (as long as the flags are * scrubbed appropriately when the direction of an mbuf changes). CSUM_BITS * would also have to split into CSUM_BITS_TX and CSUM_BITS_RX. * * CSUM_INNER_ is the same as CSUM_ but it applies to the inner frame. * The CSUM_ENCAP_ bits identify the outer encapsulation. */ #define CSUM_IP 0x00000001 /* IP header checksum offload */ #define CSUM_IP_UDP 0x00000002 /* UDP checksum offload */ #define CSUM_IP_TCP 0x00000004 /* TCP checksum offload */ #define CSUM_IP_SCTP 0x00000008 /* SCTP checksum offload */ #define CSUM_IP_TSO 0x00000010 /* TCP segmentation offload */ #define CSUM_IP_ISCSI 0x00000020 /* iSCSI checksum offload */ #define CSUM_INNER_IP6_UDP 0x00000040 #define CSUM_INNER_IP6_TCP 0x00000080 #define CSUM_INNER_IP6_TSO 0x00000100 #define CSUM_IP6_UDP 0x00000200 /* UDP checksum offload */ #define CSUM_IP6_TCP 0x00000400 /* TCP checksum offload */ #define CSUM_IP6_SCTP 0x00000800 /* SCTP checksum offload */ #define CSUM_IP6_TSO 0x00001000 /* TCP segmentation offload */ #define CSUM_IP6_ISCSI 0x00002000 /* iSCSI checksum offload */ #define CSUM_INNER_IP 0x00004000 #define CSUM_INNER_IP_UDP 0x00008000 #define CSUM_INNER_IP_TCP 0x00010000 #define CSUM_INNER_IP_TSO 0x00020000 #define CSUM_ENCAP_VXLAN 0x00040000 /* VXLAN outer encapsulation */ #define CSUM_ENCAP_RSVD1 0x00080000 /* Inbound checksum support where the checksum was verified by hardware. */ #define CSUM_INNER_L3_CALC 0x00100000 #define CSUM_INNER_L3_VALID 0x00200000 #define CSUM_INNER_L4_CALC 0x00400000 #define CSUM_INNER_L4_VALID 0x00800000 #define CSUM_L3_CALC 0x01000000 /* calculated layer 3 csum */ #define CSUM_L3_VALID 0x02000000 /* checksum is correct */ #define CSUM_L4_CALC 0x04000000 /* calculated layer 4 csum */ #define CSUM_L4_VALID 0x08000000 /* checksum is correct */ #define CSUM_L5_CALC 0x10000000 /* calculated layer 5 csum */ #define CSUM_L5_VALID 0x20000000 /* checksum is correct */ #define CSUM_COALESCED 0x40000000 /* contains merged segments */ #define CSUM_SND_TAG 0x80000000 /* Packet header has send tag */ #define CSUM_FLAGS_TX (CSUM_IP | CSUM_IP_UDP | CSUM_IP_TCP | CSUM_IP_SCTP | \ CSUM_IP_TSO | CSUM_IP_ISCSI | CSUM_INNER_IP6_UDP | CSUM_INNER_IP6_TCP | \ CSUM_INNER_IP6_TSO | CSUM_IP6_UDP | CSUM_IP6_TCP | CSUM_IP6_SCTP | \ CSUM_IP6_TSO | CSUM_IP6_ISCSI | CSUM_INNER_IP | CSUM_INNER_IP_UDP | \ CSUM_INNER_IP_TCP | CSUM_INNER_IP_TSO | CSUM_ENCAP_VXLAN | \ CSUM_ENCAP_RSVD1 | CSUM_SND_TAG) #define CSUM_FLAGS_RX (CSUM_INNER_L3_CALC | CSUM_INNER_L3_VALID | \ CSUM_INNER_L4_CALC | CSUM_INNER_L4_VALID | CSUM_L3_CALC | CSUM_L3_VALID | \ CSUM_L4_CALC | CSUM_L4_VALID | CSUM_L5_CALC | CSUM_L5_VALID | \ CSUM_COALESCED) /* * CSUM flag description for use with printf(9) %b identifier. */ #define CSUM_BITS \ "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \ "\6CSUM_IP_ISCSI\7CSUM_INNER_IP6_UDP\10CSUM_INNER_IP6_TCP" \ "\11CSUM_INNER_IP6_TSO\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP" \ "\15CSUM_IP6_TSO\16CSUM_IP6_ISCSI\17CSUM_INNER_IP\20CSUM_INNER_IP_UDP" \ "\21CSUM_INNER_IP_TCP\22CSUM_INNER_IP_TSO\23CSUM_ENCAP_VXLAN" \ "\24CSUM_ENCAP_RSVD1\25CSUM_INNER_L3_CALC\26CSUM_INNER_L3_VALID" \ "\27CSUM_INNER_L4_CALC\30CSUM_INNER_L4_VALID\31CSUM_L3_CALC" \ "\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID\35CSUM_L5_CALC" \ "\36CSUM_L5_VALID\37CSUM_COALESCED\40CSUM_SND_TAG" /* CSUM flags compatibility mappings. */ #define CSUM_IP_CHECKED CSUM_L3_CALC #define CSUM_IP_VALID CSUM_L3_VALID #define CSUM_DATA_VALID CSUM_L4_VALID #define CSUM_PSEUDO_HDR CSUM_L4_CALC #define CSUM_SCTP_VALID CSUM_L4_VALID #define CSUM_DELAY_DATA (CSUM_TCP|CSUM_UDP) #define CSUM_DELAY_IP CSUM_IP /* Only v4, no v6 IP hdr csum */ #define CSUM_DELAY_DATA_IPV6 (CSUM_TCP_IPV6|CSUM_UDP_IPV6) #define CSUM_DATA_VALID_IPV6 CSUM_DATA_VALID #define CSUM_TCP CSUM_IP_TCP #define CSUM_UDP CSUM_IP_UDP #define CSUM_SCTP CSUM_IP_SCTP #define CSUM_TSO (CSUM_IP_TSO|CSUM_IP6_TSO) #define CSUM_INNER_TSO (CSUM_INNER_IP_TSO|CSUM_INNER_IP6_TSO) #define CSUM_UDP_IPV6 CSUM_IP6_UDP #define CSUM_TCP_IPV6 CSUM_IP6_TCP #define CSUM_SCTP_IPV6 CSUM_IP6_SCTP #define CSUM_TLS_MASK (CSUM_L5_CALC|CSUM_L5_VALID) #define CSUM_TLS_DECRYPTED CSUM_L5_CALC /* * mbuf types describing the content of the mbuf (including external storage). */ #define MT_NOTMBUF 0 /* USED INTERNALLY ONLY! Object is not mbuf */ #define MT_DATA 1 /* dynamic (data) allocation */ #define MT_HEADER MT_DATA /* packet header, use M_PKTHDR instead */ #define MT_VENDOR1 4 /* for vendor-internal use */ #define MT_VENDOR2 5 /* for vendor-internal use */ #define MT_VENDOR3 6 /* for vendor-internal use */ #define MT_VENDOR4 7 /* for vendor-internal use */ #define MT_SONAME 8 /* socket name */ #define MT_EXP1 9 /* for experimental use */ #define MT_EXP2 10 /* for experimental use */ #define MT_EXP3 11 /* for experimental use */ #define MT_EXP4 12 /* for experimental use */ #define MT_CONTROL 14 /* extra-data protocol message */ #define MT_EXTCONTROL 15 /* control message with externalized contents */ #define MT_OOBDATA 16 /* expedited data */ #define MT_NOINIT 255 /* Not a type but a flag to allocate a non-initialized mbuf */ /* * String names of mbuf-related UMA(9) and malloc(9) types. Exposed to * !_KERNEL so that monitoring tools can look up the zones with * libmemstat(3). */ #define MBUF_MEM_NAME "mbuf" #define MBUF_CLUSTER_MEM_NAME "mbuf_cluster" #define MBUF_PACKET_MEM_NAME "mbuf_packet" #define MBUF_JUMBOP_MEM_NAME "mbuf_jumbo_page" #define MBUF_JUMBO9_MEM_NAME "mbuf_jumbo_9k" #define MBUF_JUMBO16_MEM_NAME "mbuf_jumbo_16k" #define MBUF_TAG_MEM_NAME "mbuf_tag" #define MBUF_EXTREFCNT_MEM_NAME "mbuf_ext_refcnt" #define MBUF_EXTPGS_MEM_NAME "mbuf_extpgs" #ifdef _KERNEL union if_snd_tag_alloc_params; #ifdef WITNESS #define MBUF_CHECKSLEEP(how) do { \ if (how == M_WAITOK) \ WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, \ "Sleeping in \"%s\"", __func__); \ } while (0) #else #define MBUF_CHECKSLEEP(how) do {} while (0) #endif /* * Network buffer allocation API * * The rest of it is defined in kern/kern_mbuf.c */ extern uma_zone_t zone_mbuf; extern uma_zone_t zone_clust; extern uma_zone_t zone_pack; extern uma_zone_t zone_jumbop; extern uma_zone_t zone_jumbo9; extern uma_zone_t zone_jumbo16; extern uma_zone_t zone_extpgs; void mb_dupcl(struct mbuf *, struct mbuf *); void mb_free_ext(struct mbuf *); void mb_free_extpg(struct mbuf *); void mb_free_mext_pgs(struct mbuf *); struct mbuf *mb_alloc_ext_pgs(int, m_ext_free_t); struct mbuf *mb_alloc_ext_plus_pages(int, int); struct mbuf *mb_mapped_to_unmapped(struct mbuf *, int, int, int, struct mbuf **); int mb_unmapped_compress(struct mbuf *m); struct mbuf *mb_unmapped_to_ext(struct mbuf *m); void mb_free_notready(struct mbuf *m, int count); void m_adj(struct mbuf *, int); void m_adj_decap(struct mbuf *, int); int m_apply(struct mbuf *, int, int, int (*)(void *, void *, u_int), void *); int m_append(struct mbuf *, int, c_caddr_t); void m_cat(struct mbuf *, struct mbuf *); void m_catpkt(struct mbuf *, struct mbuf *); int m_clget(struct mbuf *m, int how); void *m_cljget(struct mbuf *m, int how, int size); struct mbuf *m_collapse(struct mbuf *, int, int); void m_copyback(struct mbuf *, int, int, c_caddr_t); void m_copydata(const struct mbuf *, int, int, caddr_t); struct mbuf *m_copym(struct mbuf *, int, int, int); struct mbuf *m_copypacket(struct mbuf *, int); void m_copy_pkthdr(struct mbuf *, struct mbuf *); struct mbuf *m_copyup(struct mbuf *, int, int); struct mbuf *m_defrag(struct mbuf *, int); void m_demote_pkthdr(struct mbuf *); void m_demote(struct mbuf *, int, int); struct mbuf *m_devget(char *, int, int, struct ifnet *, void (*)(char *, caddr_t, u_int)); void m_dispose_extcontrolm(struct mbuf *m); struct mbuf *m_dup(const struct mbuf *, int); int m_dup_pkthdr(struct mbuf *, const struct mbuf *, int); void m_extadd(struct mbuf *, char *, u_int, m_ext_free_t, void *, void *, int, int); u_int m_fixhdr(struct mbuf *); struct mbuf *m_fragment(struct mbuf *, int, int); void m_freem(struct mbuf *); void m_free_raw(struct mbuf *); struct mbuf *m_get2(int, int, short, int); struct mbuf *m_get3(int, int, short, int); struct mbuf *m_getjcl(int, short, int, int); struct mbuf *m_getm2(struct mbuf *, int, int, short, int); struct mbuf *m_getptr(struct mbuf *, int, int *); u_int m_length(struct mbuf *, struct mbuf **); int m_mbuftouio(struct uio *, const struct mbuf *, int); void m_move_pkthdr(struct mbuf *, struct mbuf *); int m_pkthdr_init(struct mbuf *, int); struct mbuf *m_prepend(struct mbuf *, int, int); void m_print(const struct mbuf *, int); struct mbuf *m_pulldown(struct mbuf *, int, int, int *); struct mbuf *m_pullup(struct mbuf *, int); int m_sanity(struct mbuf *, int); struct mbuf *m_split(struct mbuf *, int, int); struct mbuf *m_uiotombuf(struct uio *, int, int, int, int); int m_unmapped_uiomove(const struct mbuf *, int, struct uio *, int); struct mbuf *m_unshare(struct mbuf *, int); int m_snd_tag_alloc(struct ifnet *, union if_snd_tag_alloc_params *, struct m_snd_tag **); void m_snd_tag_init(struct m_snd_tag *, struct ifnet *, const struct if_snd_tag_sw *); void m_snd_tag_destroy(struct m_snd_tag *); void m_rcvif_serialize(struct mbuf *); struct ifnet *m_rcvif_restore(struct mbuf *); static __inline int m_gettype(int size) { int type; switch (size) { case MSIZE: type = EXT_MBUF; break; case MCLBYTES: type = EXT_CLUSTER; break; case MJUMPAGESIZE: type = EXT_JUMBOP; break; case MJUM9BYTES: type = EXT_JUMBO9; break; case MJUM16BYTES: type = EXT_JUMBO16; break; default: panic("%s: invalid cluster size %d", __func__, size); } return (type); } /* * Associated an external reference counted buffer with an mbuf. */ static __inline void m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt, m_ext_free_t freef, void *arg1, void *arg2) { KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__)); atomic_add_int(ref_cnt, 1); m->m_flags |= M_EXT; m->m_ext.ext_buf = buf; m->m_ext.ext_cnt = ref_cnt; m->m_data = m->m_ext.ext_buf; m->m_ext.ext_size = size; m->m_ext.ext_free = freef; m->m_ext.ext_arg1 = arg1; m->m_ext.ext_arg2 = arg2; m->m_ext.ext_type = EXT_EXTREF; m->m_ext.ext_flags = 0; } static __inline uma_zone_t m_getzone(int size) { uma_zone_t zone; switch (size) { case MCLBYTES: zone = zone_clust; break; case MJUMPAGESIZE: zone = zone_jumbop; break; case MJUM9BYTES: zone = zone_jumbo9; break; case MJUM16BYTES: zone = zone_jumbo16; break; default: panic("%s: invalid cluster size %d", __func__, size); } return (zone); } /* * Initialize an mbuf with linear storage. * * Inline because the consumer text overhead will be roughly the same to * initialize or call a function with this many parameters and M_PKTHDR * should go away with constant propagation for !MGETHDR. */ static __inline int m_init(struct mbuf *m, int how, short type, int flags) { int error; m->m_next = NULL; m->m_nextpkt = NULL; m->m_data = m->m_dat; m->m_len = 0; m->m_flags = flags; m->m_type = type; if (flags & M_PKTHDR) error = m_pkthdr_init(m, how); else error = 0; MBUF_PROBE5(m__init, m, how, type, flags, error); return (error); } static __inline struct mbuf * m_get_raw(int how, short type) { struct mbuf *m; struct mb_args args; args.flags = 0; args.type = type | MT_NOINIT; m = uma_zalloc_arg(zone_mbuf, &args, how); MBUF_PROBE3(m__get_raw, how, type, m); return (m); } static __inline struct mbuf * m_get(int how, short type) { struct mbuf *m; struct mb_args args; args.flags = 0; args.type = type; m = uma_zalloc_arg(zone_mbuf, &args, how); MBUF_PROBE3(m__get, how, type, m); return (m); } static __inline struct mbuf * m_gethdr_raw(int how, short type) { struct mbuf *m; struct mb_args args; args.flags = M_PKTHDR; args.type = type | MT_NOINIT; m = uma_zalloc_arg(zone_mbuf, &args, how); MBUF_PROBE3(m__gethdr_raw, how, type, m); return (m); } static __inline struct mbuf * m_gethdr(int how, short type) { struct mbuf *m; struct mb_args args; args.flags = M_PKTHDR; args.type = type; m = uma_zalloc_arg(zone_mbuf, &args, how); MBUF_PROBE3(m__gethdr, how, type, m); return (m); } static __inline struct mbuf * m_getcl(int how, short type, int flags) { struct mbuf *m; struct mb_args args; args.flags = flags; args.type = type; m = uma_zalloc_arg(zone_pack, &args, how); MBUF_PROBE4(m__getcl, how, type, flags, m); return (m); } /* * XXX: m_cljset() is a dangerous API. One must attach only a new, * unreferenced cluster to an mbuf(9). It is not possible to assert * that, so care can be taken only by users of the API. */ static __inline void m_cljset(struct mbuf *m, void *cl, int type) { int size; switch (type) { case EXT_CLUSTER: size = MCLBYTES; break; case EXT_JUMBOP: size = MJUMPAGESIZE; break; case EXT_JUMBO9: size = MJUM9BYTES; break; case EXT_JUMBO16: size = MJUM16BYTES; break; default: panic("%s: unknown cluster type %d", __func__, type); break; } m->m_data = m->m_ext.ext_buf = cl; m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL; m->m_ext.ext_size = size; m->m_ext.ext_type = type; m->m_ext.ext_flags = EXT_FLAG_EMBREF; m->m_ext.ext_count = 1; m->m_flags |= M_EXT; MBUF_PROBE3(m__cljset, m, cl, type); } static __inline void m_chtype(struct mbuf *m, short new_type) { m->m_type = new_type; } static __inline void m_clrprotoflags(struct mbuf *m) { while (m) { m->m_flags &= ~M_PROTOFLAGS; m = m->m_next; } } static __inline struct mbuf * m_last(struct mbuf *m) { while (m->m_next) m = m->m_next; return (m); } static inline u_int m_extrefcnt(struct mbuf *m) { KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__)); return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count : *m->m_ext.ext_cnt); } /* * mbuf, cluster, and external object allocation macros (for compatibility * purposes). */ #define M_MOVE_PKTHDR(to, from) m_move_pkthdr((to), (from)) #define MGET(m, how, type) ((m) = m_get((how), (type))) #define MGETHDR(m, how, type) ((m) = m_gethdr((how), (type))) #define MCLGET(m, how) m_clget((m), (how)) #define MEXTADD(m, buf, size, free, arg1, arg2, flags, type) \ m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2), \ (flags), (type)) #define m_getm(m, len, how, type) \ m_getm2((m), (len), (how), (type), M_PKTHDR) /* * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can * be both the local data payload, or an external buffer area, depending on * whether M_EXT is set). */ #define M_WRITABLE(m) (((m)->m_flags & (M_RDONLY | M_EXTPG)) == 0 && \ (!(((m)->m_flags & M_EXT)) || \ (m_extrefcnt(m) == 1))) /* Check if the supplied mbuf has a packet header, or else panic. */ #define M_ASSERTPKTHDR(m) \ KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR, \ ("%s: no mbuf packet header!", __func__)) /* Check if the supplied mbuf has no send tag, or else panic. */ #define M_ASSERT_NO_SND_TAG(m) \ KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR && \ ((m)->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0, \ ("%s: receive mbuf has send tag!", __func__)) /* Check if mbuf is multipage. */ #define M_ASSERTEXTPG(m) \ KASSERT(((m)->m_flags & (M_EXTPG|M_PKTHDR)) == M_EXTPG, \ ("%s: m %p is not multipage!", __func__, m)) /* * Ensure that the supplied mbuf is a valid, non-free mbuf. * * XXX: Broken at the moment. Need some UMA magic to make it work again. */ #define M_ASSERTVALID(m) \ KASSERT((((struct mbuf *)m)->m_flags & 0) == 0, \ ("%s: attempted use of a free mbuf!", __func__)) /* Check whether any mbuf in the chain is unmapped. */ #ifdef INVARIANTS #define M_ASSERTMAPPED(m) do { \ for (struct mbuf *__m = (m); __m != NULL; __m = __m->m_next) \ KASSERT((__m->m_flags & M_EXTPG) == 0, \ ("%s: chain %p contains an unmapped mbuf", __func__, (m)));\ } while (0) #else #define M_ASSERTMAPPED(m) do {} while (0) #endif /* * Return the address of the start of the buffer associated with an mbuf, * handling external storage, packet-header mbufs, and regular data mbufs. */ #define M_START(m) \ (((m)->m_flags & M_EXTPG) ? NULL : \ ((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf : \ ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] : \ &(m)->m_dat[0]) /* * Return the size of the buffer associated with an mbuf, handling external * storage, packet-header mbufs, and regular data mbufs. */ #define M_SIZE(m) \ (((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size : \ ((m)->m_flags & M_PKTHDR) ? MHLEN : \ MLEN) /* * Set the m_data pointer of a newly allocated mbuf to place an object of the * specified size at the end of the mbuf, longword aligned. * * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as * separate macros, each asserting that it was called at the proper moment. * This required callers to themselves test the storage type and call the * right one. Rather than require callers to be aware of those layout * decisions, we centralize here. */ static __inline void m_align(struct mbuf *m, int len) { #ifdef INVARIANTS const char *msg = "%s: not a virgin mbuf"; #endif int adjust; KASSERT(m->m_data == M_START(m), (msg, __func__)); adjust = M_SIZE(m) - len; m->m_data += adjust &~ (sizeof(long)-1); } #define M_ALIGN(m, len) m_align(m, len) #define MH_ALIGN(m, len) m_align(m, len) #define MEXT_ALIGN(m, len) m_align(m, len) /* * Compute the amount of space available before the current start of data in * an mbuf. * * The M_WRITABLE() is a temporary, conservative safety measure: the burden * of checking writability of the mbuf data area rests solely with the caller. * * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE() * for mbufs with external storage. We now allow mbuf-embedded data to be * read-only as well. */ #define M_LEADINGSPACE(m) \ (M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0) /* * Compute the amount of space available after the end of data in an mbuf. * * The M_WRITABLE() is a temporary, conservative safety measure: the burden * of checking writability of the mbuf data area rests solely with the caller. * * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE() * for mbufs with external storage. We now allow mbuf-embedded data to be * read-only as well. */ #define M_TRAILINGSPACE(m) \ (M_WRITABLE(m) ? \ ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len)) : 0) /* * Arrange to prepend space of size plen to mbuf m. If a new mbuf must be * allocated, how specifies whether to wait. If the allocation fails, the * original mbuf chain is freed and m is set to NULL. */ #define M_PREPEND(m, plen, how) do { \ struct mbuf **_mmp = &(m); \ struct mbuf *_mm = *_mmp; \ int _mplen = (plen); \ int __mhow = (how); \ \ MBUF_CHECKSLEEP(how); \ if (M_LEADINGSPACE(_mm) >= _mplen) { \ _mm->m_data -= _mplen; \ _mm->m_len += _mplen; \ } else \ _mm = m_prepend(_mm, _mplen, __mhow); \ if (_mm != NULL && _mm->m_flags & M_PKTHDR) \ _mm->m_pkthdr.len += _mplen; \ *_mmp = _mm; \ } while (0) /* * Change mbuf to new type. This is a relatively expensive operation and * should be avoided. */ #define MCHTYPE(m, t) m_chtype((m), (t)) /* Return the rcvif of a packet header. */ static __inline struct ifnet * m_rcvif(struct mbuf *m) { M_ASSERTPKTHDR(m); if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) return (NULL); return (m->m_pkthdr.rcvif); } /* Length to m_copy to copy all. */ #define M_COPYALL 1000000000 -extern int max_datalen; /* MHLEN - max_hdr */ -extern int max_hdr; /* Largest link + protocol header */ -extern int max_linkhdr; /* Largest link-level header */ -extern int max_protohdr; /* Largest protocol header */ +extern u_int max_linkhdr; /* Largest link-level header */ +extern u_int max_hdr; /* Largest link + protocol header */ +extern u_int max_protohdr; /* Largest protocol header */ +void max_linkhdr_grow(u_int); +void max_protohdr_grow(u_int); + extern int nmbclusters; /* Maximum number of clusters */ extern bool mb_use_ext_pgs; /* Use ext_pgs for sendfile */ /*- * Network packets may have annotations attached by affixing a list of * "packet tags" to the pkthdr structure. Packet tags are dynamically * allocated semi-opaque data structures that have a fixed header * (struct m_tag) that specifies the size of the memory block and a * pair that identifies it. The cookie is a 32-bit unique * unsigned value used to identify a module or ABI. By convention this value * is chosen as the date+time that the module is created, expressed as the * number of seconds since the epoch (e.g., using date -u +'%s'). The type * value is an ABI/module-specific value that identifies a particular * annotation and is private to the module. For compatibility with systems * like OpenBSD that define packet tags w/o an ABI/module cookie, the value * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find * compatibility shim functions and several tag types are defined below. * Users that do not require compatibility should use a private cookie value * so that packet tag-related definitions can be maintained privately. * * Note that the packet tag returned by m_tag_alloc has the default memory * alignment implemented by malloc. To reference private data one can use a * construct like: * * struct m_tag *mtag = m_tag_alloc(...); * struct foo *p = (struct foo *)(mtag+1); * * if the alignment of struct m_tag is sufficient for referencing members of * struct foo. Otherwise it is necessary to embed struct m_tag within the * private data structure to insure proper alignment; e.g., * * struct foo { * struct m_tag tag; * ... * }; * struct foo *p = (struct foo *) m_tag_alloc(...); * struct m_tag *mtag = &p->tag; */ /* * Persistent tags stay with an mbuf until the mbuf is reclaimed. Otherwise * tags are expected to ``vanish'' when they pass through a network * interface. For most interfaces this happens normally as the tags are * reclaimed when the mbuf is free'd. However in some special cases * reclaiming must be done manually. An example is packets that pass through * the loopback interface. Also, one must be careful to do this when * ``turning around'' packets (e.g., icmp_reflect). * * To mark a tag persistent bit-or this flag in when defining the tag id. * The tag will then be treated as described above. */ #define MTAG_PERSISTENT 0x800 #define PACKET_TAG_NONE 0 /* Nadda */ /* Packet tags for use with PACKET_ABI_COMPAT. */ #define PACKET_TAG_IPSEC_IN_DONE 1 /* IPsec applied, in */ #define PACKET_TAG_IPSEC_OUT_DONE 2 /* IPsec applied, out */ #define PACKET_TAG_IPSEC_IN_CRYPTO_DONE 3 /* NIC IPsec crypto done */ #define PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED 4 /* NIC IPsec crypto req'ed */ #define PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO 5 /* NIC notifies IPsec */ #define PACKET_TAG_IPSEC_PENDING_TDB 6 /* Reminder to do IPsec */ #define PACKET_TAG_BRIDGE 7 /* Bridge processing done */ #define PACKET_TAG_GIF 8 /* GIF processing done */ #define PACKET_TAG_GRE 9 /* GRE processing done */ #define PACKET_TAG_IN_PACKET_CHECKSUM 10 /* NIC checksumming done */ #define PACKET_TAG_ENCAP 11 /* Encap. processing */ #define PACKET_TAG_IPSEC_SOCKET 12 /* IPSEC socket ref */ #define PACKET_TAG_IPSEC_HISTORY 13 /* IPSEC history */ #define PACKET_TAG_IPV6_INPUT 14 /* IPV6 input processing */ #define PACKET_TAG_DUMMYNET 15 /* dummynet info */ #define PACKET_TAG_DIVERT 17 /* divert info */ #define PACKET_TAG_IPFORWARD 18 /* ipforward info */ #define PACKET_TAG_MACLABEL (19 | MTAG_PERSISTENT) /* MAC label */ #define PACKET_TAG_PF 21 /* PF/ALTQ information */ /* was PACKET_TAG_RTSOCKFAM 25 rtsock sa family */ #define PACKET_TAG_IPOPTIONS 27 /* Saved IP options */ #define PACKET_TAG_CARP 28 /* CARP info */ #define PACKET_TAG_IPSEC_NAT_T_PORTS 29 /* two uint16_t */ #define PACKET_TAG_ND_OUTGOING 30 /* ND outgoing */ /* Specific cookies and tags. */ /* Packet tag routines. */ struct m_tag *m_tag_alloc(uint32_t, uint16_t, int, int); void m_tag_delete(struct mbuf *, struct m_tag *); void m_tag_delete_chain(struct mbuf *, struct m_tag *); void m_tag_free_default(struct m_tag *); struct m_tag *m_tag_locate(struct mbuf *, uint32_t, uint16_t, struct m_tag *); struct m_tag *m_tag_copy(struct m_tag *, int); int m_tag_copy_chain(struct mbuf *, const struct mbuf *, int); void m_tag_delete_nonpersistent(struct mbuf *); /* * Initialize the list of tags associated with an mbuf. */ static __inline void m_tag_init(struct mbuf *m) { SLIST_INIT(&m->m_pkthdr.tags); } /* * Set up the contents of a tag. Note that this does not fill in the free * method; the caller is expected to do that. * * XXX probably should be called m_tag_init, but that was already taken. */ static __inline void m_tag_setup(struct m_tag *t, uint32_t cookie, uint16_t type, int len) { t->m_tag_id = type; t->m_tag_len = len; t->m_tag_cookie = cookie; } /* * Reclaim resources associated with a tag. */ static __inline void m_tag_free(struct m_tag *t) { (*t->m_tag_free)(t); } /* * Return the first tag associated with an mbuf. */ static __inline struct m_tag * m_tag_first(struct mbuf *m) { return (SLIST_FIRST(&m->m_pkthdr.tags)); } /* * Return the next tag in the list of tags associated with an mbuf. */ static __inline struct m_tag * m_tag_next(struct mbuf *m __unused, struct m_tag *t) { return (SLIST_NEXT(t, m_tag_link)); } /* * Prepend a tag to the list of tags associated with an mbuf. */ static __inline void m_tag_prepend(struct mbuf *m, struct m_tag *t) { SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link); } /* * Unlink a tag from the list of tags associated with an mbuf. */ static __inline void m_tag_unlink(struct mbuf *m, struct m_tag *t) { SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link); } /* These are for OpenBSD compatibility. */ #define MTAG_ABI_COMPAT 0 /* compatibility ABI */ static __inline struct m_tag * m_tag_get(uint16_t type, int length, int wait) { return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait)); } static __inline struct m_tag * m_tag_find(struct mbuf *m, uint16_t type, struct m_tag *start) { return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL : m_tag_locate(m, MTAG_ABI_COMPAT, type, start)); } static inline struct m_snd_tag * m_snd_tag_ref(struct m_snd_tag *mst) { refcount_acquire(&mst->refcount); return (mst); } static inline void m_snd_tag_rele(struct m_snd_tag *mst) { if (refcount_release(&mst->refcount)) m_snd_tag_destroy(mst); } static __inline struct mbuf * m_free(struct mbuf *m) { struct mbuf *n = m->m_next; MBUF_PROBE1(m__free, m); if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE)) m_tag_delete_chain(m, NULL); if (m->m_flags & M_PKTHDR && m->m_pkthdr.csum_flags & CSUM_SND_TAG) m_snd_tag_rele(m->m_pkthdr.snd_tag); if (m->m_flags & M_EXTPG) mb_free_extpg(m); else if (m->m_flags & M_EXT) mb_free_ext(m); else if ((m->m_flags & M_NOFREE) == 0) uma_zfree(zone_mbuf, m); return (n); } static __inline int rt_m_getfib(struct mbuf *m) { KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf.")); return (m->m_pkthdr.fibnum); } #define M_GETFIB(_m) rt_m_getfib(_m) #define M_SETFIB(_m, _fib) do { \ KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf.")); \ ((_m)->m_pkthdr.fibnum) = (_fib); \ } while (0) /* flags passed as first argument for "m_xxx_tcpip_hash()" */ #define MBUF_HASHFLAG_L2 (1 << 2) #define MBUF_HASHFLAG_L3 (1 << 3) #define MBUF_HASHFLAG_L4 (1 << 4) /* mbuf hashing helper routines */ uint32_t m_ether_tcpip_hash_init(void); uint32_t m_ether_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t); uint32_t m_infiniband_tcpip_hash_init(void); uint32_t m_infiniband_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t); #ifdef MBUF_PROFILING void m_profile(struct mbuf *m); #define M_PROFILE(m) m_profile(m) #else #define M_PROFILE(m) #endif struct mbufq { STAILQ_HEAD(, mbuf) mq_head; int mq_len; int mq_maxlen; }; static inline void mbufq_init(struct mbufq *mq, int maxlen) { STAILQ_INIT(&mq->mq_head); mq->mq_maxlen = maxlen; mq->mq_len = 0; } static inline struct mbuf * mbufq_flush(struct mbufq *mq) { struct mbuf *m; m = STAILQ_FIRST(&mq->mq_head); STAILQ_INIT(&mq->mq_head); mq->mq_len = 0; return (m); } static inline void mbufq_drain(struct mbufq *mq) { struct mbuf *m, *n; n = mbufq_flush(mq); while ((m = n) != NULL) { n = STAILQ_NEXT(m, m_stailqpkt); m_freem(m); } } static inline struct mbuf * mbufq_first(const struct mbufq *mq) { return (STAILQ_FIRST(&mq->mq_head)); } static inline struct mbuf * mbufq_last(const struct mbufq *mq) { return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt)); } static inline int mbufq_full(const struct mbufq *mq) { return (mq->mq_maxlen > 0 && mq->mq_len >= mq->mq_maxlen); } static inline int mbufq_len(const struct mbufq *mq) { return (mq->mq_len); } static inline int mbufq_enqueue(struct mbufq *mq, struct mbuf *m) { if (mbufq_full(mq)) return (ENOBUFS); STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt); mq->mq_len++; return (0); } static inline struct mbuf * mbufq_dequeue(struct mbufq *mq) { struct mbuf *m; m = STAILQ_FIRST(&mq->mq_head); if (m) { STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt); m->m_nextpkt = NULL; mq->mq_len--; } return (m); } static inline void mbufq_prepend(struct mbufq *mq, struct mbuf *m) { STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt); mq->mq_len++; } /* * Note: this doesn't enforce the maximum list size for dst. */ static inline void mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src) { mq_dst->mq_len += mq_src->mq_len; STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head); mq_src->mq_len = 0; } #ifdef _SYS_TIMESPEC_H_ static inline void mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts) { KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m)); KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0, ("mbuf %p no M_TSTMP or M_TSTMP_LRO", m)); ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000; ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000; } #endif #ifdef DEBUGNET /* Invoked from the debugnet client code. */ void debugnet_mbuf_drain(void); void debugnet_mbuf_start(void); void debugnet_mbuf_finish(void); void debugnet_mbuf_reinit(int nmbuf, int nclust, int clsize); #endif static inline bool mbuf_has_tls_session(struct mbuf *m) { if (m->m_flags & M_EXTPG) { if (m->m_epg_tls != NULL) { return (true); } } return (false); } #endif /* _KERNEL */ #endif /* !_SYS_MBUF_H_ */