From owner-svn-src-head@FreeBSD.ORG Mon Jan 5 09:58:37 2015 Return-Path: Delivered-To: svn-src-head@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [8.8.178.115]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by hub.freebsd.org (Postfix) with ESMTPS id C0192138; Mon, 5 Jan 2015 09:58:37 +0000 (UTC) Received: from svn.freebsd.org (svn.freebsd.org [IPv6:2001:1900:2254:2068::e6a:0]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (Client did not present a certificate) by mx1.freebsd.org (Postfix) with ESMTPS id AA0E46487D; Mon, 5 Jan 2015 09:58:37 +0000 (UTC) Received: from svn.freebsd.org ([127.0.1.70]) by svn.freebsd.org (8.14.9/8.14.9) with ESMTP id t059wbpF018614; Mon, 5 Jan 2015 09:58:37 GMT (envelope-from rwatson@FreeBSD.org) Received: (from rwatson@localhost) by svn.freebsd.org (8.14.9/8.14.9/Submit) id t059wXCd018594; Mon, 5 Jan 2015 09:58:33 GMT (envelope-from rwatson@FreeBSD.org) Message-Id: <201501050958.t059wXCd018594@svn.freebsd.org> X-Authentication-Warning: svn.freebsd.org: rwatson set sender to rwatson@FreeBSD.org using -f From: Robert Watson Date: Mon, 5 Jan 2015 09:58:33 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r276692 - in head/sys: dev/en dev/fatm dev/iscsi_initiator dev/patm kern net80211 netinet netinet6 sys X-SVN-Group: head MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: svn-src-head@freebsd.org X-Mailman-Version: 2.1.18-1 Precedence: list List-Id: SVN commit messages for the src tree for head/-current List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Mon, 05 Jan 2015 09:58:37 -0000 Author: rwatson Date: Mon Jan 5 09:58:32 2015 New Revision: 276692 URL: https://svnweb.freebsd.org/changeset/base/276692 Log: To ease changes to underlying mbuf structure and the mbuf allocator, reduce the knowledge of mbuf layout, and in particular constants such as M_EXT, MLEN, MHLEN, and so on, in mbuf consumers by unifying various alignment utility functions (M_ALIGN(), MH_ALIGN(), MEXT_ALIGN() in a single M_ALIGN() macro, implemented by a now-inlined m_align() function: - Move m_align() from uipc_mbuf.c to mbuf.h; mark as __inline. - Reimplement M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() using m_align(). - Update consumers around the tree to simply use M_ALIGN(). This change eliminates a number of cases where mbuf consumers must be aware of whether or not mbufs returned by the allocator use external storage, but also assumptions about the size of the returned mbuf. This will make it easier to introduce changes in how we use external storage, as well as features such as variable-size mbufs. Differential Revision: https://reviews.freebsd.org/D1436 Reviewed by: glebius, trasz, gnn, bz Sponsored by: EMC / Isilon Storage Division Modified: head/sys/dev/en/midway.c head/sys/dev/fatm/if_fatm.c head/sys/dev/iscsi_initiator/isc_soc.c head/sys/dev/patm/if_patm_rx.c head/sys/kern/uipc_mbuf.c head/sys/net80211/ieee80211_freebsd.c head/sys/netinet/if_ether.c head/sys/netinet/igmp.c head/sys/netinet/ip_carp.c head/sys/netinet/sctp_os_bsd.h head/sys/netinet/tcp_output.c head/sys/netinet6/ip6_output.c head/sys/netinet6/mld6.c head/sys/netinet6/nd6_nbr.c head/sys/sys/mbuf.h Modified: head/sys/dev/en/midway.c ============================================================================== --- head/sys/dev/en/midway.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/dev/en/midway.c Mon Jan 5 09:58:32 2015 (r276692) @@ -1935,7 +1935,7 @@ en_mget(struct en_softc *sc, u_int pktle m->m_pkthdr.rcvif = NULL; m->m_pkthdr.len = pktlen; m->m_len = EN_RX1BUF; - MH_ALIGN(m, EN_RX1BUF); + M_ALIGN(m, EN_RX1BUF); if (m->m_len >= totlen) { m->m_len = totlen; Modified: head/sys/dev/fatm/if_fatm.c ============================================================================== --- head/sys/dev/fatm/if_fatm.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/dev/fatm/if_fatm.c Mon Jan 5 09:58:32 2015 (r276692) @@ -1105,7 +1105,7 @@ fatm_supply_small_buffers(struct fatm_so LIST_INSERT_HEAD(&sc->rbuf_free, rb, link); break; } - MH_ALIGN(m, SMALL_BUFFER_LEN); + M_ALIGN(m, SMALL_BUFFER_LEN); error = bus_dmamap_load(sc->rbuf_tag, rb->map, m->m_data, SMALL_BUFFER_LEN, dmaload_helper, &phys, BUS_DMA_NOWAIT); Modified: head/sys/dev/iscsi_initiator/isc_soc.c ============================================================================== --- head/sys/dev/iscsi_initiator/isc_soc.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/dev/iscsi_initiator/isc_soc.c Mon Jan 5 09:58:32 2015 (r276692) @@ -110,7 +110,7 @@ isc_sendPDU(isc_session_t *sp, pduq_t *p | Add any AHS to the iSCSI hdr mbuf */ if((mh->m_len + pp->ahs_len) < MHLEN) { - MH_ALIGN(mh, mh->m_len + pp->ahs_len); + M_ALIGN(mh, mh->m_len + pp->ahs_len); bcopy(&pp->ipdu, mh->m_data, mh->m_len); bcopy(pp->ahs_addr, mh->m_data + mh->m_len, pp->ahs_len); mh->m_len += pp->ahs_len; @@ -119,7 +119,7 @@ isc_sendPDU(isc_session_t *sp, pduq_t *p panic("len AHS=%d too big, not impleneted yet", pp->ahs_len); } else { - MH_ALIGN(mh, mh->m_len); + M_ALIGN(mh, mh->m_len); bcopy(&pp->ipdu, mh->m_data, mh->m_len); } mh->m_pkthdr.len = mh->m_len; Modified: head/sys/dev/patm/if_patm_rx.c ============================================================================== --- head/sys/dev/patm/if_patm_rx.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/dev/patm/if_patm_rx.c Mon Jan 5 09:58:32 2015 (r276692) @@ -471,7 +471,7 @@ patm_rx_raw(struct patm_softc *sc, u_cha default: case PATM_RAW_CELL: m->m_len = m->m_pkthdr.len = 53; - MH_ALIGN(m, 53); + M_ALIGN(m, 53); dst = mtod(m, u_char *); *dst++ = *cell++; *dst++ = *cell++; @@ -483,7 +483,7 @@ patm_rx_raw(struct patm_softc *sc, u_cha case PATM_RAW_NOHEC: m->m_len = m->m_pkthdr.len = 52; - MH_ALIGN(m, 52); + M_ALIGN(m, 52); dst = mtod(m, u_char *); *dst++ = *cell++; *dst++ = *cell++; @@ -494,7 +494,7 @@ patm_rx_raw(struct patm_softc *sc, u_cha case PATM_RAW_CS: m->m_len = m->m_pkthdr.len = 64; - MH_ALIGN(m, 64); + M_ALIGN(m, 64); dst = mtod(m, u_char *); *dst++ = *cell++; *dst++ = *cell++; Modified: head/sys/kern/uipc_mbuf.c ============================================================================== --- head/sys/kern/uipc_mbuf.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/kern/uipc_mbuf.c Mon Jan 5 09:58:32 2015 (r276692) @@ -574,13 +574,8 @@ m_prepend(struct mbuf *m, int len, int h m_move_pkthdr(mn, m); mn->m_next = m; m = mn; - if(m->m_flags & M_PKTHDR) { - if (len < MHLEN) - MH_ALIGN(m, len); - } else { - if (len < MLEN) - M_ALIGN(m, len); - } + if (len < M_SIZE(m)) + M_ALIGN(m, len); m->m_len = len; return (m); } @@ -1226,7 +1221,7 @@ m_split(struct mbuf *m0, int len0, int w goto extpacket; if (remain > MHLEN) { /* m can't be the lead packet */ - MH_ALIGN(n, 0); + M_ALIGN(n, 0); n->m_next = m_split(m, len, wait); if (n->m_next == NULL) { (void) m_free(n); @@ -1236,7 +1231,7 @@ m_split(struct mbuf *m0, int len0, int w return (n); } } else - MH_ALIGN(n, remain); + M_ALIGN(n, remain); } else if (remain == 0) { n = m->m_next; m->m_next = NULL; @@ -1888,33 +1883,6 @@ m_mbuftouio(struct uio *uio, struct mbuf } /* - * 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. - */ -void -m_align(struct mbuf *m, int len) -{ -#ifdef INVARIANTS - const char *msg = "%s: not a virgin mbuf"; -#endif - int adjust; - - if (m->m_flags & M_EXT) { - KASSERT(m->m_data == m->m_ext.ext_buf, (msg, __func__)); - adjust = m->m_ext.ext_size - len; - } else if (m->m_flags & M_PKTHDR) { - KASSERT(m->m_data == m->m_pktdat, (msg, __func__)); - adjust = MHLEN - len; - } else { - KASSERT(m->m_data == m->m_dat, (msg, __func__)); - adjust = MLEN - len; - } - - m->m_data += adjust &~ (sizeof(long)-1); -} - -/* * 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 Modified: head/sys/net80211/ieee80211_freebsd.c ============================================================================== --- head/sys/net80211/ieee80211_freebsd.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/net80211/ieee80211_freebsd.c Mon Jan 5 09:58:32 2015 (r276692) @@ -420,7 +420,7 @@ ieee80211_getmgtframe(uint8_t **frm, int * frames which all fit in MHLEN. */ if (m != NULL) - MH_ALIGN(m, len); + M_ALIGN(m, len); } else { m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m != NULL) Modified: head/sys/netinet/if_ether.c ============================================================================== --- head/sys/netinet/if_ether.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/netinet/if_ether.c Mon Jan 5 09:58:32 2015 (r276692) @@ -261,7 +261,7 @@ arprequest(struct ifnet *ifp, const stru m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) + 2 * ifp->if_addrlen; m->m_pkthdr.len = m->m_len; - MH_ALIGN(m, m->m_len); + M_ALIGN(m, m->m_len); ah = mtod(m, struct arphdr *); bzero((caddr_t)ah, m->m_len); #ifdef MAC Modified: head/sys/netinet/igmp.c ============================================================================== --- head/sys/netinet/igmp.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/netinet/igmp.c Mon Jan 5 09:58:32 2015 (r276692) @@ -2212,7 +2212,7 @@ igmp_v1v2_queue_report(struct in_multi * m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (ENOMEM); - MH_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp)); + M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp)); m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp); @@ -2793,7 +2793,7 @@ igmp_v3_enqueue_group_record(struct ifqu if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m) - MH_ALIGN(m, IGMP_LEADINGSPACE); + M_ALIGN(m, IGMP_LEADINGSPACE); } if (m == NULL) return (-ENOMEM); @@ -2917,7 +2917,7 @@ igmp_v3_enqueue_group_record(struct ifqu if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m) - MH_ALIGN(m, IGMP_LEADINGSPACE); + M_ALIGN(m, IGMP_LEADINGSPACE); } if (m == NULL) return (-ENOMEM); @@ -3073,7 +3073,7 @@ igmp_v3_enqueue_filter_change(struct ifq if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m) - MH_ALIGN(m, IGMP_LEADINGSPACE); + M_ALIGN(m, IGMP_LEADINGSPACE); } if (m == NULL) { CTR1(KTR_IGMPV3, Modified: head/sys/netinet/ip_carp.c ============================================================================== --- head/sys/netinet/ip_carp.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/netinet/ip_carp.c Mon Jan 5 09:58:32 2015 (r276692) @@ -840,7 +840,7 @@ carp_send_ad_locked(struct carp_softc *s m->m_pkthdr.len = len; m->m_pkthdr.rcvif = NULL; m->m_len = len; - MH_ALIGN(m, m->m_len); + M_ALIGN(m, m->m_len); m->m_flags |= M_MCAST; ip = mtod(m, struct ip *); ip->ip_v = IPVERSION; @@ -892,7 +892,7 @@ carp_send_ad_locked(struct carp_softc *s m->m_pkthdr.len = len; m->m_pkthdr.rcvif = NULL; m->m_len = len; - MH_ALIGN(m, m->m_len); + M_ALIGN(m, m->m_len); m->m_flags |= M_MCAST; ip6 = mtod(m, struct ip6_hdr *); bzero(ip6, sizeof(*ip6)); Modified: head/sys/netinet/sctp_os_bsd.h ============================================================================== --- head/sys/netinet/sctp_os_bsd.h Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/netinet/sctp_os_bsd.h Mon Jan 5 09:58:32 2015 (r276692) @@ -304,11 +304,7 @@ typedef struct callout sctp_os_timer_t; #define SCTP_BUF_RECVIF(m) (m->m_pkthdr.rcvif) #define SCTP_BUF_PREPEND M_PREPEND -#define SCTP_ALIGN_TO_END(m, len) if(m->m_flags & M_PKTHDR) { \ - MH_ALIGN(m, len); \ - } else if ((m->m_flags & M_EXT) == 0) { \ - M_ALIGN(m, len); \ - } +#define SCTP_ALIGN_TO_END(m, len) M_ALIGN(m, len) /* We make it so if you have up to 4 threads * writing based on the default size of Modified: head/sys/netinet/tcp_output.c ============================================================================== --- head/sys/netinet/tcp_output.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/netinet/tcp_output.c Mon Jan 5 09:58:32 2015 (r276692) @@ -1007,7 +1007,7 @@ send: #ifdef INET6 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) && MHLEN >= hdrlen) { - MH_ALIGN(m, hdrlen); + M_ALIGN(m, hdrlen); } else #endif m->m_data += max_linkhdr; Modified: head/sys/netinet6/ip6_output.c ============================================================================== --- head/sys/netinet6/ip6_output.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/netinet6/ip6_output.c Mon Jan 5 09:58:32 2015 (r276692) @@ -2940,7 +2940,7 @@ ip6_splithdr(struct mbuf *m, struct ip6_ return ENOBUFS; } m_move_pkthdr(mh, m); - MH_ALIGN(mh, sizeof(*ip6)); + M_ALIGN(mh, sizeof(*ip6)); m->m_len -= sizeof(*ip6); m->m_data += sizeof(*ip6); mh->m_next = m; Modified: head/sys/netinet6/mld6.c ============================================================================== --- head/sys/netinet6/mld6.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/netinet6/mld6.c Mon Jan 5 09:58:32 2015 (r276692) @@ -1818,7 +1818,7 @@ mld_v1_transmit_report(struct in6_multi * that ether_output() does not need to allocate another mbuf * for the header in the most common case. */ - MH_ALIGN(mh, sizeof(struct ip6_hdr)); + M_ALIGN(mh, sizeof(struct ip6_hdr)); mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr); mh->m_len = sizeof(struct ip6_hdr); @@ -3179,7 +3179,7 @@ mld_v2_encap_report(struct ifnet *ifp, s m_freem(m); return (NULL); } - MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report)); + M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report)); mldreclen = m_length(m, NULL); CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen); Modified: head/sys/netinet6/nd6_nbr.c ============================================================================== --- head/sys/netinet6/nd6_nbr.c Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/netinet6/nd6_nbr.c Mon Jan 5 09:58:32 2015 (r276692) @@ -431,7 +431,7 @@ nd6_ns_output(struct ifnet *ifp, const s icmp6len = sizeof(*nd_ns); m->m_pkthdr.len = m->m_len = sizeof(*ip6) + icmp6len; - m->m_data += max_linkhdr; /* or MH_ALIGN() equivalent? */ + m->m_data += max_linkhdr; /* or M_ALIGN() equivalent? */ /* fill neighbor solicitation packet */ ip6 = mtod(m, struct ip6_hdr *); @@ -1003,7 +1003,7 @@ nd6_na_output_fib(struct ifnet *ifp, con icmp6len = sizeof(*nd_na); m->m_pkthdr.len = m->m_len = sizeof(struct ip6_hdr) + icmp6len; - m->m_data += max_linkhdr; /* or MH_ALIGN() equivalent? */ + m->m_data += max_linkhdr; /* or M_ALIGN() equivalent? */ /* fill neighbor advertisement packet */ ip6 = mtod(m, struct ip6_hdr *); Modified: head/sys/sys/mbuf.h ============================================================================== --- head/sys/sys/mbuf.h Mon Jan 5 05:30:07 2015 (r276691) +++ head/sys/sys/mbuf.h Mon Jan 5 09:58:32 2015 (r276692) @@ -810,42 +810,6 @@ m_last(struct mbuf *m) ("%s: attempted use of a free mbuf!", __func__)) /* - * Set the m_data pointer of a newly-allocated mbuf (m_get/MGET) to place an - * object of the specified size at the end of the mbuf, longword aligned. - */ -#define M_ALIGN(m, len) do { \ - KASSERT(!((m)->m_flags & (M_PKTHDR|M_EXT)), \ - ("%s: M_ALIGN not normal mbuf", __func__)); \ - KASSERT((m)->m_data == (m)->m_dat, \ - ("%s: M_ALIGN not a virgin mbuf", __func__)); \ - (m)->m_data += (MLEN - (len)) & ~(sizeof(long) - 1); \ -} while (0) - -/* - * As above, for mbufs allocated with m_gethdr/MGETHDR or initialized by - * M_DUP/MOVE_PKTHDR. - */ -#define MH_ALIGN(m, len) do { \ - KASSERT((m)->m_flags & M_PKTHDR && !((m)->m_flags & M_EXT), \ - ("%s: MH_ALIGN not PKTHDR mbuf", __func__)); \ - KASSERT((m)->m_data == (m)->m_pktdat, \ - ("%s: MH_ALIGN not a virgin mbuf", __func__)); \ - (m)->m_data += (MHLEN - (len)) & ~(sizeof(long) - 1); \ -} while (0) - -/* - * As above, for mbuf with external storage. - */ -#define MEXT_ALIGN(m, len) do { \ - KASSERT((m)->m_flags & M_EXT, \ - ("%s: MEXT_ALIGN not an M_EXT mbuf", __func__)); \ - KASSERT((m)->m_data == (m)->m_ext.ext_buf, \ - ("%s: MEXT_ALIGN not a virgin mbuf", __func__)); \ - (m)->m_data += ((m)->m_ext.ext_size - (len)) & \ - ~(sizeof(long) - 1); \ -} while (0) - -/* * Return the address of the start of the buffer associated with an mbuf, * handling external storage, packet-header mbufs, and regular data mbufs. */ @@ -864,6 +828,41 @@ m_last(struct mbuf *m) 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__)); + + if (m->m_flags & M_EXT) { + adjust = m->m_ext.ext_size - len; + } else if (m->m_flags & M_PKTHDR) { + adjust = MHLEN - len; + } else { + adjust = MLEN - 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. *