Index: head/sys/dev/xen/netback/netback.c =================================================================== --- head/sys/dev/xen/netback/netback.c (revision 296241) +++ head/sys/dev/xen/netback/netback.c (revision 296242) @@ -1,2520 +1,2520 @@ /*- * Copyright (c) 2009-2011 Spectra Logic Corporation * 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, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR 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 DAMAGES. * * Authors: Justin T. Gibbs (Spectra Logic Corporation) * Alan Somers (Spectra Logic Corporation) * John Suykerbuyk (Spectra Logic Corporation) */ #include __FBSDID("$FreeBSD$"); /** * \file netback.c * * \brief Device driver supporting the vending of network access * from this FreeBSD domain to other domains. */ #include "opt_inet.h" #include "opt_inet6.h" #include "opt_sctp.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if __FreeBSD_version >= 700000 #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include /*--------------------------- Compile-time Tunables --------------------------*/ /*---------------------------------- Macros ----------------------------------*/ /** * Custom malloc type for all driver allocations. */ static MALLOC_DEFINE(M_XENNETBACK, "xnb", "Xen Net Back Driver Data"); #define XNB_SG 1 /* netback driver supports feature-sg */ #define XNB_GSO_TCPV4 0 /* netback driver supports feature-gso-tcpv4 */ #define XNB_RX_COPY 1 /* netback driver supports feature-rx-copy */ #define XNB_RX_FLIP 0 /* netback driver does not support feature-rx-flip */ #undef XNB_DEBUG #define XNB_DEBUG /* hardcode on during development */ #ifdef XNB_DEBUG #define DPRINTF(fmt, args...) \ printf("xnb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) #else #define DPRINTF(fmt, args...) do {} while (0) #endif /* Default length for stack-allocated grant tables */ #define GNTTAB_LEN (64) /* Features supported by all backends. TSO and LRO can be negotiated */ #define XNB_CSUM_FEATURES (CSUM_TCP | CSUM_UDP) #define NET_TX_RING_SIZE __RING_SIZE((netif_tx_sring_t *)0, PAGE_SIZE) #define NET_RX_RING_SIZE __RING_SIZE((netif_rx_sring_t *)0, PAGE_SIZE) /** * Two argument version of the standard macro. Second argument is a tentative * value of req_cons */ #define RING_HAS_UNCONSUMED_REQUESTS_2(_r, cons) ({ \ unsigned int req = (_r)->sring->req_prod - cons; \ unsigned int rsp = RING_SIZE(_r) - \ (cons - (_r)->rsp_prod_pvt); \ req < rsp ? req : rsp; \ }) #define virt_to_mfn(x) (vtophys(x) >> PAGE_SHIFT) #define virt_to_offset(x) ((x) & (PAGE_SIZE - 1)) /** * Predefined array type of grant table copy descriptors. Used to pass around * statically allocated memory structures. */ typedef struct gnttab_copy gnttab_copy_table[GNTTAB_LEN]; /*--------------------------- Forward Declarations ---------------------------*/ struct xnb_softc; struct xnb_pkt; static void xnb_attach_failed(struct xnb_softc *xnb, int err, const char *fmt, ...) __printflike(3,4); static int xnb_shutdown(struct xnb_softc *xnb); static int create_netdev(device_t dev); static int xnb_detach(device_t dev); static int xnb_ifmedia_upd(struct ifnet *ifp); static void xnb_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr); static void xnb_intr(void *arg); static int xnb_send(netif_rx_back_ring_t *rxb, domid_t otherend, const struct mbuf *mbufc, gnttab_copy_table gnttab); static int xnb_recv(netif_tx_back_ring_t *txb, domid_t otherend, struct mbuf **mbufc, struct ifnet *ifnet, gnttab_copy_table gnttab); static int xnb_ring2pkt(struct xnb_pkt *pkt, const netif_tx_back_ring_t *tx_ring, RING_IDX start); static void xnb_txpkt2rsp(const struct xnb_pkt *pkt, netif_tx_back_ring_t *ring, int error); static struct mbuf *xnb_pkt2mbufc(const struct xnb_pkt *pkt, struct ifnet *ifp); static int xnb_txpkt2gnttab(const struct xnb_pkt *pkt, - const struct mbuf *mbufc, + struct mbuf *mbufc, gnttab_copy_table gnttab, const netif_tx_back_ring_t *txb, domid_t otherend_id); static void xnb_update_mbufc(struct mbuf *mbufc, const gnttab_copy_table gnttab, int n_entries); static int xnb_mbufc2pkt(const struct mbuf *mbufc, struct xnb_pkt *pkt, RING_IDX start, int space); static int xnb_rxpkt2gnttab(const struct xnb_pkt *pkt, const struct mbuf *mbufc, gnttab_copy_table gnttab, const netif_rx_back_ring_t *rxb, domid_t otherend_id); static int xnb_rxpkt2rsp(const struct xnb_pkt *pkt, const gnttab_copy_table gnttab, int n_entries, netif_rx_back_ring_t *ring); static void xnb_stop(struct xnb_softc*); static int xnb_ioctl(struct ifnet*, u_long, caddr_t); static void xnb_start_locked(struct ifnet*); static void xnb_start(struct ifnet*); static void xnb_ifinit_locked(struct xnb_softc*); static void xnb_ifinit(void*); #ifdef XNB_DEBUG static int xnb_unit_test_main(SYSCTL_HANDLER_ARGS); static int xnb_dump_rings(SYSCTL_HANDLER_ARGS); #endif #if defined(INET) || defined(INET6) static void xnb_add_mbuf_cksum(struct mbuf *mbufc); #endif /*------------------------------ Data Structures -----------------------------*/ /** * Representation of a xennet packet. Simplified version of a packet as * stored in the Xen tx ring. Applicable to both RX and TX packets */ struct xnb_pkt{ /** * Array index of the first data-bearing (eg, not extra info) entry * for this packet */ RING_IDX car; /** * Array index of the second data-bearing entry for this packet. * Invalid if the packet has only one data-bearing entry. If the * packet has more than two data-bearing entries, then the second * through the last will be sequential modulo the ring size */ RING_IDX cdr; /** * Optional extra info. Only valid if flags contains * NETTXF_extra_info. Note that extra.type will always be * XEN_NETIF_EXTRA_TYPE_GSO. Currently, no known netfront or netback * driver will ever set XEN_NETIF_EXTRA_TYPE_MCAST_* */ netif_extra_info_t extra; /** Size of entire packet in bytes. */ uint16_t size; /** The size of the first entry's data in bytes */ uint16_t car_size; /** * Either NETTXF_ or NETRXF_ flags. Note that the flag values are * not the same for TX and RX packets */ uint16_t flags; /** * The number of valid data-bearing entries (either netif_tx_request's * or netif_rx_response's) in the packet. If this is 0, it means the * entire packet is invalid. */ uint16_t list_len; /** There was an error processing the packet */ uint8_t error; }; /** xnb_pkt method: initialize it */ static inline void xnb_pkt_initialize(struct xnb_pkt *pxnb) { bzero(pxnb, sizeof(*pxnb)); } /** xnb_pkt method: mark the packet as valid */ static inline void xnb_pkt_validate(struct xnb_pkt *pxnb) { pxnb->error = 0; }; /** xnb_pkt method: mark the packet as invalid */ static inline void xnb_pkt_invalidate(struct xnb_pkt *pxnb) { pxnb->error = 1; }; /** xnb_pkt method: Check whether the packet is valid */ static inline int xnb_pkt_is_valid(const struct xnb_pkt *pxnb) { return (! pxnb->error); } #ifdef XNB_DEBUG /** xnb_pkt method: print the packet's contents in human-readable format*/ static void __unused xnb_dump_pkt(const struct xnb_pkt *pkt) { if (pkt == NULL) { DPRINTF("Was passed a null pointer.\n"); return; } DPRINTF("pkt address= %p\n", pkt); DPRINTF("pkt->size=%d\n", pkt->size); DPRINTF("pkt->car_size=%d\n", pkt->car_size); DPRINTF("pkt->flags=0x%04x\n", pkt->flags); DPRINTF("pkt->list_len=%d\n", pkt->list_len); /* DPRINTF("pkt->extra"); TODO */ DPRINTF("pkt->car=%d\n", pkt->car); DPRINTF("pkt->cdr=%d\n", pkt->cdr); DPRINTF("pkt->error=%d\n", pkt->error); } #endif /* XNB_DEBUG */ static void xnb_dump_txreq(RING_IDX idx, const struct netif_tx_request *txreq) { if (txreq != NULL) { DPRINTF("netif_tx_request index =%u\n", idx); DPRINTF("netif_tx_request.gref =%u\n", txreq->gref); DPRINTF("netif_tx_request.offset=%hu\n", txreq->offset); DPRINTF("netif_tx_request.flags =%hu\n", txreq->flags); DPRINTF("netif_tx_request.id =%hu\n", txreq->id); DPRINTF("netif_tx_request.size =%hu\n", txreq->size); } } /** * \brief Configuration data for a shared memory request ring * used to communicate with the front-end client of this * this driver. */ struct xnb_ring_config { /** * Runtime structures for ring access. Unfortunately, TX and RX rings * use different data structures, and that cannot be changed since it * is part of the interdomain protocol. */ union{ netif_rx_back_ring_t rx_ring; netif_tx_back_ring_t tx_ring; } back_ring; /** * The device bus address returned by the hypervisor when * mapping the ring and required to unmap it when a connection * is torn down. */ uint64_t bus_addr; /** The pseudo-physical address where ring memory is mapped.*/ uint64_t gnt_addr; /** KVA address where ring memory is mapped. */ vm_offset_t va; /** * Grant table handles, one per-ring page, returned by the * hyperpervisor upon mapping of the ring and required to * unmap it when a connection is torn down. */ grant_handle_t handle; /** The number of ring pages mapped for the current connection. */ unsigned ring_pages; /** * The grant references, one per-ring page, supplied by the * front-end, allowing us to reference the ring pages in the * front-end's domain and to map these pages into our own domain. */ grant_ref_t ring_ref; }; /** * Per-instance connection state flags. */ typedef enum { /** Communication with the front-end has been established. */ XNBF_RING_CONNECTED = 0x01, /** * Front-end requests exist in the ring and are waiting for * xnb_xen_req objects to free up. */ XNBF_RESOURCE_SHORTAGE = 0x02, /** Connection teardown has started. */ XNBF_SHUTDOWN = 0x04, /** A thread is already performing shutdown processing. */ XNBF_IN_SHUTDOWN = 0x08 } xnb_flag_t; /** * Types of rings. Used for array indices and to identify a ring's control * data structure type */ typedef enum{ XNB_RING_TYPE_TX = 0, /* ID of TX rings, used for array indices */ XNB_RING_TYPE_RX = 1, /* ID of RX rings, used for array indices */ XNB_NUM_RING_TYPES } xnb_ring_type_t; /** * Per-instance configuration data. */ struct xnb_softc { /** NewBus device corresponding to this instance. */ device_t dev; /* Media related fields */ /** Generic network media state */ struct ifmedia sc_media; /** Media carrier info */ struct ifnet *xnb_ifp; /** Our own private carrier state */ unsigned carrier; /** Device MAC Address */ uint8_t mac[ETHER_ADDR_LEN]; /* Xen related fields */ /** * \brief The netif protocol abi in effect. * * There are situations where the back and front ends can * have a different, native abi (e.g. intel x86_64 and * 32bit x86 domains on the same machine). The back-end * always accomodates the front-end's native abi. That * value is pulled from the XenStore and recorded here. */ int abi; /** * Name of the bridge to which this VIF is connected, if any * This field is dynamically allocated by xenbus and must be free()ed * when no longer needed */ char *bridge; /** The interrupt driven even channel used to signal ring events. */ evtchn_port_t evtchn; /** Xen device handle.*/ long handle; /** Handle to the communication ring event channel. */ xen_intr_handle_t xen_intr_handle; /** * \brief Cached value of the front-end's domain id. * * This value is used at once for each mapped page in * a transaction. We cache it to avoid incuring the * cost of an ivar access every time this is needed. */ domid_t otherend_id; /** * Undocumented frontend feature. Has something to do with * scatter/gather IO */ uint8_t can_sg; /** Undocumented frontend feature */ uint8_t gso; /** Undocumented frontend feature */ uint8_t gso_prefix; /** Can checksum TCP/UDP over IPv4 */ uint8_t ip_csum; /* Implementation related fields */ /** * Preallocated grant table copy descriptor for RX operations. * Access must be protected by rx_lock */ gnttab_copy_table rx_gnttab; /** * Preallocated grant table copy descriptor for TX operations. * Access must be protected by tx_lock */ gnttab_copy_table tx_gnttab; /** * Resource representing allocated physical address space * associated with our per-instance kva region. */ struct resource *pseudo_phys_res; /** Resource id for allocated physical address space. */ int pseudo_phys_res_id; /** Ring mapping and interrupt configuration data. */ struct xnb_ring_config ring_configs[XNB_NUM_RING_TYPES]; /** * Global pool of kva used for mapping remote domain ring * and I/O transaction data. */ vm_offset_t kva; /** Psuedo-physical address corresponding to kva. */ uint64_t gnt_base_addr; /** Various configuration and state bit flags. */ xnb_flag_t flags; /** Mutex protecting per-instance data in the receive path. */ struct mtx rx_lock; /** Mutex protecting per-instance data in the softc structure. */ struct mtx sc_lock; /** Mutex protecting per-instance data in the transmit path. */ struct mtx tx_lock; /** The size of the global kva pool. */ int kva_size; /** Name of the interface */ char if_name[IFNAMSIZ]; }; /*---------------------------- Debugging functions ---------------------------*/ #ifdef XNB_DEBUG static void __unused xnb_dump_gnttab_copy(const struct gnttab_copy *entry) { if (entry == NULL) { printf("NULL grant table pointer\n"); return; } if (entry->flags & GNTCOPY_dest_gref) printf("gnttab dest ref=\t%u\n", entry->dest.u.ref); else printf("gnttab dest gmfn=\t%"PRI_xen_pfn"\n", entry->dest.u.gmfn); printf("gnttab dest offset=\t%hu\n", entry->dest.offset); printf("gnttab dest domid=\t%hu\n", entry->dest.domid); if (entry->flags & GNTCOPY_source_gref) printf("gnttab source ref=\t%u\n", entry->source.u.ref); else printf("gnttab source gmfn=\t%"PRI_xen_pfn"\n", entry->source.u.gmfn); printf("gnttab source offset=\t%hu\n", entry->source.offset); printf("gnttab source domid=\t%hu\n", entry->source.domid); printf("gnttab len=\t%hu\n", entry->len); printf("gnttab flags=\t%hu\n", entry->flags); printf("gnttab status=\t%hd\n", entry->status); } static int xnb_dump_rings(SYSCTL_HANDLER_ARGS) { static char results[720]; struct xnb_softc const* xnb = (struct xnb_softc*)arg1; netif_rx_back_ring_t const* rxb = &xnb->ring_configs[XNB_RING_TYPE_RX].back_ring.rx_ring; netif_tx_back_ring_t const* txb = &xnb->ring_configs[XNB_RING_TYPE_TX].back_ring.tx_ring; /* empty the result strings */ results[0] = 0; if ( !txb || !txb->sring || !rxb || !rxb->sring ) return (SYSCTL_OUT(req, results, strnlen(results, 720))); snprintf(results, 720, "\n\t%35s %18s\n" /* TX, RX */ "\t%16s %18d %18d\n" /* req_cons */ "\t%16s %18d %18d\n" /* nr_ents */ "\t%16s %18d %18d\n" /* rsp_prod_pvt */ "\t%16s %18p %18p\n" /* sring */ "\t%16s %18d %18d\n" /* req_prod */ "\t%16s %18d %18d\n" /* req_event */ "\t%16s %18d %18d\n" /* rsp_prod */ "\t%16s %18d %18d\n", /* rsp_event */ "TX", "RX", "req_cons", txb->req_cons, rxb->req_cons, "nr_ents", txb->nr_ents, rxb->nr_ents, "rsp_prod_pvt", txb->rsp_prod_pvt, rxb->rsp_prod_pvt, "sring", txb->sring, rxb->sring, "sring->req_prod", txb->sring->req_prod, rxb->sring->req_prod, "sring->req_event", txb->sring->req_event, rxb->sring->req_event, "sring->rsp_prod", txb->sring->rsp_prod, rxb->sring->rsp_prod, "sring->rsp_event", txb->sring->rsp_event, rxb->sring->rsp_event); return (SYSCTL_OUT(req, results, strnlen(results, 720))); } static void __unused xnb_dump_mbuf(const struct mbuf *m) { int len; uint8_t *d; if (m == NULL) return; printf("xnb_dump_mbuf:\n"); if (m->m_flags & M_PKTHDR) { printf(" flowid=%10d, csum_flags=%#8x, csum_data=%#8x, " "tso_segsz=%5hd\n", m->m_pkthdr.flowid, (int)m->m_pkthdr.csum_flags, m->m_pkthdr.csum_data, m->m_pkthdr.tso_segsz); printf(" rcvif=%16p, len=%19d\n", m->m_pkthdr.rcvif, m->m_pkthdr.len); } printf(" m_next=%16p, m_nextpk=%16p, m_data=%16p\n", m->m_next, m->m_nextpkt, m->m_data); printf(" m_len=%17d, m_flags=%#15x, m_type=%18u\n", m->m_len, m->m_flags, m->m_type); len = m->m_len; d = mtod(m, uint8_t*); while (len > 0) { int i; printf(" "); for (i = 0; (i < 16) && (len > 0); i++, len--) { printf("%02hhx ", *(d++)); } printf("\n"); } } #endif /* XNB_DEBUG */ /*------------------------ Inter-Domain Communication ------------------------*/ /** * Free dynamically allocated KVA or pseudo-physical address allocations. * * \param xnb Per-instance xnb configuration structure. */ static void xnb_free_communication_mem(struct xnb_softc *xnb) { if (xnb->kva != 0) { if (xnb->pseudo_phys_res != NULL) { xenmem_free(xnb->dev, xnb->pseudo_phys_res_id, xnb->pseudo_phys_res); xnb->pseudo_phys_res = NULL; } } xnb->kva = 0; xnb->gnt_base_addr = 0; } /** * Cleanup all inter-domain communication mechanisms. * * \param xnb Per-instance xnb configuration structure. */ static int xnb_disconnect(struct xnb_softc *xnb) { struct gnttab_unmap_grant_ref gnts[XNB_NUM_RING_TYPES]; int error; int i; if (xnb->xen_intr_handle != NULL) xen_intr_unbind(&xnb->xen_intr_handle); /* * We may still have another thread currently processing requests. We * must acquire the rx and tx locks to make sure those threads are done, * but we can release those locks as soon as we acquire them, because no * more interrupts will be arriving. */ mtx_lock(&xnb->tx_lock); mtx_unlock(&xnb->tx_lock); mtx_lock(&xnb->rx_lock); mtx_unlock(&xnb->rx_lock); /* Free malloc'd softc member variables */ if (xnb->bridge != NULL) { free(xnb->bridge, M_XENSTORE); xnb->bridge = NULL; } /* All request processing has stopped, so unmap the rings */ for (i=0; i < XNB_NUM_RING_TYPES; i++) { gnts[i].host_addr = xnb->ring_configs[i].gnt_addr; gnts[i].dev_bus_addr = xnb->ring_configs[i].bus_addr; gnts[i].handle = xnb->ring_configs[i].handle; } error = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, gnts, XNB_NUM_RING_TYPES); KASSERT(error == 0, ("Grant table unmap op failed (%d)", error)); xnb_free_communication_mem(xnb); /* * Zero the ring config structs because the pointers, handles, and * grant refs contained therein are no longer valid. */ bzero(&xnb->ring_configs[XNB_RING_TYPE_TX], sizeof(struct xnb_ring_config)); bzero(&xnb->ring_configs[XNB_RING_TYPE_RX], sizeof(struct xnb_ring_config)); xnb->flags &= ~XNBF_RING_CONNECTED; return (0); } /** * Map a single shared memory ring into domain local address space and * initialize its control structure * * \param xnb Per-instance xnb configuration structure * \param ring_type Array index of this ring in the xnb's array of rings * \return An errno */ static int xnb_connect_ring(struct xnb_softc *xnb, xnb_ring_type_t ring_type) { struct gnttab_map_grant_ref gnt; struct xnb_ring_config *ring = &xnb->ring_configs[ring_type]; int error; /* TX ring type = 0, RX =1 */ ring->va = xnb->kva + ring_type * PAGE_SIZE; ring->gnt_addr = xnb->gnt_base_addr + ring_type * PAGE_SIZE; gnt.host_addr = ring->gnt_addr; gnt.flags = GNTMAP_host_map; gnt.ref = ring->ring_ref; gnt.dom = xnb->otherend_id; error = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, &gnt, 1); if (error != 0) panic("netback: Ring page grant table op failed (%d)", error); if (gnt.status != 0) { ring->va = 0; error = EACCES; xenbus_dev_fatal(xnb->dev, error, "Ring shared page mapping failed. " "Status %d.", gnt.status); } else { ring->handle = gnt.handle; ring->bus_addr = gnt.dev_bus_addr; if (ring_type == XNB_RING_TYPE_TX) { BACK_RING_INIT(&ring->back_ring.tx_ring, (netif_tx_sring_t*)ring->va, ring->ring_pages * PAGE_SIZE); } else if (ring_type == XNB_RING_TYPE_RX) { BACK_RING_INIT(&ring->back_ring.rx_ring, (netif_rx_sring_t*)ring->va, ring->ring_pages * PAGE_SIZE); } else { xenbus_dev_fatal(xnb->dev, error, "Unknown ring type %d", ring_type); } } return error; } /** * Setup the shared memory rings and bind an interrupt to the event channel * used to notify us of ring changes. * * \param xnb Per-instance xnb configuration structure. */ static int xnb_connect_comms(struct xnb_softc *xnb) { int error; xnb_ring_type_t i; if ((xnb->flags & XNBF_RING_CONNECTED) != 0) return (0); /* * Kva for our rings are at the tail of the region of kva allocated * by xnb_alloc_communication_mem(). */ for (i=0; i < XNB_NUM_RING_TYPES; i++) { error = xnb_connect_ring(xnb, i); if (error != 0) return error; } xnb->flags |= XNBF_RING_CONNECTED; error = xen_intr_bind_remote_port(xnb->dev, xnb->otherend_id, xnb->evtchn, /*filter*/NULL, xnb_intr, /*arg*/xnb, INTR_TYPE_BIO | INTR_MPSAFE, &xnb->xen_intr_handle); if (error != 0) { (void)xnb_disconnect(xnb); xenbus_dev_fatal(xnb->dev, error, "binding event channel"); return (error); } DPRINTF("rings connected!\n"); return (0); } /** * Size KVA and pseudo-physical address allocations based on negotiated * values for the size and number of I/O requests, and the size of our * communication ring. * * \param xnb Per-instance xnb configuration structure. * * These address spaces are used to dynamically map pages in the * front-end's domain into our own. */ static int xnb_alloc_communication_mem(struct xnb_softc *xnb) { xnb_ring_type_t i; xnb->kva_size = 0; for (i=0; i < XNB_NUM_RING_TYPES; i++) { xnb->kva_size += xnb->ring_configs[i].ring_pages * PAGE_SIZE; } /* * Reserve a range of pseudo physical memory that we can map * into kva. These pages will only be backed by machine * pages ("real memory") during the lifetime of front-end requests * via grant table operations. We will map the netif tx and rx rings * into this space. */ xnb->pseudo_phys_res_id = 0; xnb->pseudo_phys_res = xenmem_alloc(xnb->dev, &xnb->pseudo_phys_res_id, xnb->kva_size); if (xnb->pseudo_phys_res == NULL) { xnb->kva = 0; return (ENOMEM); } xnb->kva = (vm_offset_t)rman_get_virtual(xnb->pseudo_phys_res); xnb->gnt_base_addr = rman_get_start(xnb->pseudo_phys_res); return (0); } /** * Collect information from the XenStore related to our device and its frontend * * \param xnb Per-instance xnb configuration structure. */ static int xnb_collect_xenstore_info(struct xnb_softc *xnb) { /** * \todo Linux collects the following info. We should collect most * of this, too: * "feature-rx-notify" */ const char *otherend_path; const char *our_path; int err; unsigned int rx_copy, bridge_len; uint8_t no_csum_offload; otherend_path = xenbus_get_otherend_path(xnb->dev); our_path = xenbus_get_node(xnb->dev); /* Collect the critical communication parameters */ err = xs_gather(XST_NIL, otherend_path, "tx-ring-ref", "%l" PRIu32, &xnb->ring_configs[XNB_RING_TYPE_TX].ring_ref, "rx-ring-ref", "%l" PRIu32, &xnb->ring_configs[XNB_RING_TYPE_RX].ring_ref, "event-channel", "%" PRIu32, &xnb->evtchn, NULL); if (err != 0) { xenbus_dev_fatal(xnb->dev, err, "Unable to retrieve ring information from " "frontend %s. Unable to connect.", otherend_path); return (err); } /* Collect the handle from xenstore */ err = xs_scanf(XST_NIL, our_path, "handle", NULL, "%li", &xnb->handle); if (err != 0) { xenbus_dev_fatal(xnb->dev, err, "Error reading handle from frontend %s. " "Unable to connect.", otherend_path); } /* * Collect the bridgename, if any. We do not need bridge_len; we just * throw it away */ err = xs_read(XST_NIL, our_path, "bridge", &bridge_len, (void**)&xnb->bridge); if (err != 0) xnb->bridge = NULL; /* * Does the frontend request that we use rx copy? If not, return an * error because this driver only supports rx copy. */ err = xs_scanf(XST_NIL, otherend_path, "request-rx-copy", NULL, "%" PRIu32, &rx_copy); if (err == ENOENT) { err = 0; rx_copy = 0; } if (err < 0) { xenbus_dev_fatal(xnb->dev, err, "reading %s/request-rx-copy", otherend_path); return err; } /** * \todo: figure out the exact meaning of this feature, and when * the frontend will set it to true. It should be set to true * at some point */ /* if (!rx_copy)*/ /* return EOPNOTSUPP;*/ /** \todo Collect the rx notify feature */ /* Collect the feature-sg. */ if (xs_scanf(XST_NIL, otherend_path, "feature-sg", NULL, "%hhu", &xnb->can_sg) < 0) xnb->can_sg = 0; /* Collect remaining frontend features */ if (xs_scanf(XST_NIL, otherend_path, "feature-gso-tcpv4", NULL, "%hhu", &xnb->gso) < 0) xnb->gso = 0; if (xs_scanf(XST_NIL, otherend_path, "feature-gso-tcpv4-prefix", NULL, "%hhu", &xnb->gso_prefix) < 0) xnb->gso_prefix = 0; if (xs_scanf(XST_NIL, otherend_path, "feature-no-csum-offload", NULL, "%hhu", &no_csum_offload) < 0) no_csum_offload = 0; xnb->ip_csum = (no_csum_offload == 0); return (0); } /** * Supply information about the physical device to the frontend * via XenBus. * * \param xnb Per-instance xnb configuration structure. */ static int xnb_publish_backend_info(struct xnb_softc *xnb) { struct xs_transaction xst; const char *our_path; int error; our_path = xenbus_get_node(xnb->dev); do { error = xs_transaction_start(&xst); if (error != 0) { xenbus_dev_fatal(xnb->dev, error, "Error publishing backend info " "(start transaction)"); break; } error = xs_printf(xst, our_path, "feature-sg", "%d", XNB_SG); if (error != 0) break; error = xs_printf(xst, our_path, "feature-gso-tcpv4", "%d", XNB_GSO_TCPV4); if (error != 0) break; error = xs_printf(xst, our_path, "feature-rx-copy", "%d", XNB_RX_COPY); if (error != 0) break; error = xs_printf(xst, our_path, "feature-rx-flip", "%d", XNB_RX_FLIP); if (error != 0) break; error = xs_transaction_end(xst, 0); if (error != 0 && error != EAGAIN) { xenbus_dev_fatal(xnb->dev, error, "ending transaction"); break; } } while (error == EAGAIN); return (error); } /** * Connect to our netfront peer now that it has completed publishing * its configuration into the XenStore. * * \param xnb Per-instance xnb configuration structure. */ static void xnb_connect(struct xnb_softc *xnb) { int error; if (xenbus_get_state(xnb->dev) == XenbusStateConnected) return; if (xnb_collect_xenstore_info(xnb) != 0) return; xnb->flags &= ~XNBF_SHUTDOWN; /* Read front end configuration. */ /* Allocate resources whose size depends on front-end configuration. */ error = xnb_alloc_communication_mem(xnb); if (error != 0) { xenbus_dev_fatal(xnb->dev, error, "Unable to allocate communication memory"); return; } /* * Connect communication channel. */ error = xnb_connect_comms(xnb); if (error != 0) { /* Specific errors are reported by xnb_connect_comms(). */ return; } xnb->carrier = 1; /* Ready for I/O. */ xenbus_set_state(xnb->dev, XenbusStateConnected); } /*-------------------------- Device Teardown Support -------------------------*/ /** * Perform device shutdown functions. * * \param xnb Per-instance xnb configuration structure. * * Mark this instance as shutting down, wait for any active requests * to drain, disconnect from the front-end, and notify any waiters (e.g. * a thread invoking our detach method) that detach can now proceed. */ static int xnb_shutdown(struct xnb_softc *xnb) { /* * Due to the need to drop our mutex during some * xenbus operations, it is possible for two threads * to attempt to close out shutdown processing at * the same time. Tell the caller that hits this * race to try back later. */ if ((xnb->flags & XNBF_IN_SHUTDOWN) != 0) return (EAGAIN); xnb->flags |= XNBF_SHUTDOWN; xnb->flags |= XNBF_IN_SHUTDOWN; mtx_unlock(&xnb->sc_lock); /* Free the network interface */ xnb->carrier = 0; if (xnb->xnb_ifp != NULL) { ether_ifdetach(xnb->xnb_ifp); if_free(xnb->xnb_ifp); xnb->xnb_ifp = NULL; } mtx_lock(&xnb->sc_lock); xnb_disconnect(xnb); mtx_unlock(&xnb->sc_lock); if (xenbus_get_state(xnb->dev) < XenbusStateClosing) xenbus_set_state(xnb->dev, XenbusStateClosing); mtx_lock(&xnb->sc_lock); xnb->flags &= ~XNBF_IN_SHUTDOWN; /* Indicate to xnb_detach() that is it safe to proceed. */ wakeup(xnb); return (0); } /** * Report an attach time error to the console and Xen, and cleanup * this instance by forcing immediate detach processing. * * \param xnb Per-instance xnb configuration structure. * \param err Errno describing the error. * \param fmt Printf style format and arguments */ static void xnb_attach_failed(struct xnb_softc *xnb, int err, const char *fmt, ...) { va_list ap; va_list ap_hotplug; va_start(ap, fmt); va_copy(ap_hotplug, ap); xs_vprintf(XST_NIL, xenbus_get_node(xnb->dev), "hotplug-error", fmt, ap_hotplug); va_end(ap_hotplug); xs_printf(XST_NIL, xenbus_get_node(xnb->dev), "hotplug-status", "error"); xenbus_dev_vfatal(xnb->dev, err, fmt, ap); va_end(ap); xs_printf(XST_NIL, xenbus_get_node(xnb->dev), "online", "0"); xnb_detach(xnb->dev); } /*---------------------------- NewBus Entrypoints ----------------------------*/ /** * Inspect a XenBus device and claim it if is of the appropriate type. * * \param dev NewBus device object representing a candidate XenBus device. * * \return 0 for success, errno codes for failure. */ static int xnb_probe(device_t dev) { if (!strcmp(xenbus_get_type(dev), "vif")) { DPRINTF("Claiming device %d, %s\n", device_get_unit(dev), devclass_get_name(device_get_devclass(dev))); device_set_desc(dev, "Backend Virtual Network Device"); device_quiet(dev); return (0); } return (ENXIO); } /** * Setup sysctl variables to control various Network Back parameters. * * \param xnb Xen Net Back softc. * */ static void xnb_setup_sysctl(struct xnb_softc *xnb) { struct sysctl_ctx_list *sysctl_ctx = NULL; struct sysctl_oid *sysctl_tree = NULL; sysctl_ctx = device_get_sysctl_ctx(xnb->dev); if (sysctl_ctx == NULL) return; sysctl_tree = device_get_sysctl_tree(xnb->dev); if (sysctl_tree == NULL) return; #ifdef XNB_DEBUG SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "unit_test_results", CTLTYPE_STRING | CTLFLAG_RD, xnb, 0, xnb_unit_test_main, "A", "Results of builtin unit tests"); SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "dump_rings", CTLTYPE_STRING | CTLFLAG_RD, xnb, 0, xnb_dump_rings, "A", "Xennet Back Rings"); #endif /* XNB_DEBUG */ } /** * Create a network device. * @param handle device handle */ int create_netdev(device_t dev) { struct ifnet *ifp; struct xnb_softc *xnb; int err = 0; uint32_t handle; xnb = device_get_softc(dev); mtx_init(&xnb->sc_lock, "xnb_softc", "xen netback softc lock", MTX_DEF); mtx_init(&xnb->tx_lock, "xnb_tx", "xen netback tx lock", MTX_DEF); mtx_init(&xnb->rx_lock, "xnb_rx", "xen netback rx lock", MTX_DEF); xnb->dev = dev; ifmedia_init(&xnb->sc_media, 0, xnb_ifmedia_upd, xnb_ifmedia_sts); ifmedia_add(&xnb->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL); ifmedia_set(&xnb->sc_media, IFM_ETHER|IFM_MANUAL); /* * Set the MAC address to a dummy value (00:00:00:00:00), * if the MAC address of the host-facing interface is set * to the same as the guest-facing one (the value found in * xenstore), the bridge would stop delivering packets to * us because it would see that the destination address of * the packet is the same as the interface, and so the bridge * would expect the packet has already been delivered locally * (and just drop it). */ bzero(&xnb->mac[0], sizeof(xnb->mac)); /* The interface will be named using the following nomenclature: * * xnb. * * Where handle is the oder of the interface referred to the guest. */ err = xs_scanf(XST_NIL, xenbus_get_node(xnb->dev), "handle", NULL, "%" PRIu32, &handle); if (err != 0) return (err); snprintf(xnb->if_name, IFNAMSIZ, "xnb%" PRIu16 ".%" PRIu32, xenbus_get_otherend_id(dev), handle); if (err == 0) { /* Set up ifnet structure */ ifp = xnb->xnb_ifp = if_alloc(IFT_ETHER); ifp->if_softc = xnb; if_initname(ifp, xnb->if_name, IF_DUNIT_NONE); ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_ioctl = xnb_ioctl; ifp->if_output = ether_output; ifp->if_start = xnb_start; #ifdef notyet ifp->if_watchdog = xnb_watchdog; #endif ifp->if_init = xnb_ifinit; ifp->if_mtu = ETHERMTU; ifp->if_snd.ifq_maxlen = NET_RX_RING_SIZE - 1; ifp->if_hwassist = XNB_CSUM_FEATURES; ifp->if_capabilities = IFCAP_HWCSUM; ifp->if_capenable = IFCAP_HWCSUM; ether_ifattach(ifp, xnb->mac); xnb->carrier = 0; } return err; } /** * Attach to a XenBus device that has been claimed by our probe routine. * * \param dev NewBus device object representing this Xen Net Back instance. * * \return 0 for success, errno codes for failure. */ static int xnb_attach(device_t dev) { struct xnb_softc *xnb; int error; xnb_ring_type_t i; error = create_netdev(dev); if (error != 0) { xenbus_dev_fatal(dev, error, "creating netdev"); return (error); } DPRINTF("Attaching to %s\n", xenbus_get_node(dev)); /* * Basic initialization. * After this block it is safe to call xnb_detach() * to clean up any allocated data for this instance. */ xnb = device_get_softc(dev); xnb->otherend_id = xenbus_get_otherend_id(dev); for (i=0; i < XNB_NUM_RING_TYPES; i++) { xnb->ring_configs[i].ring_pages = 1; } /* * Setup sysctl variables. */ xnb_setup_sysctl(xnb); /* Update hot-plug status to satisfy xend. */ error = xs_printf(XST_NIL, xenbus_get_node(xnb->dev), "hotplug-status", "connected"); if (error != 0) { xnb_attach_failed(xnb, error, "writing %s/hotplug-status", xenbus_get_node(xnb->dev)); return (error); } if ((error = xnb_publish_backend_info(xnb)) != 0) { /* * If we can't publish our data, we cannot participate * in this connection, and waiting for a front-end state * change will not help the situation. */ xnb_attach_failed(xnb, error, "Publishing backend status for %s", xenbus_get_node(xnb->dev)); return error; } /* Tell the front end that we are ready to connect. */ xenbus_set_state(dev, XenbusStateInitWait); return (0); } /** * Detach from a net back device instance. * * \param dev NewBus device object representing this Xen Net Back instance. * * \return 0 for success, errno codes for failure. * * \note A net back device may be detached at any time in its life-cycle, * including part way through the attach process. For this reason, * initialization order and the intialization state checks in this * routine must be carefully coupled so that attach time failures * are gracefully handled. */ static int xnb_detach(device_t dev) { struct xnb_softc *xnb; DPRINTF("\n"); xnb = device_get_softc(dev); mtx_lock(&xnb->sc_lock); while (xnb_shutdown(xnb) == EAGAIN) { msleep(xnb, &xnb->sc_lock, /*wakeup prio unchanged*/0, "xnb_shutdown", 0); } mtx_unlock(&xnb->sc_lock); DPRINTF("\n"); mtx_destroy(&xnb->tx_lock); mtx_destroy(&xnb->rx_lock); mtx_destroy(&xnb->sc_lock); return (0); } /** * Prepare this net back device for suspension of this VM. * * \param dev NewBus device object representing this Xen net Back instance. * * \return 0 for success, errno codes for failure. */ static int xnb_suspend(device_t dev) { return (0); } /** * Perform any processing required to recover from a suspended state. * * \param dev NewBus device object representing this Xen Net Back instance. * * \return 0 for success, errno codes for failure. */ static int xnb_resume(device_t dev) { return (0); } /** * Handle state changes expressed via the XenStore by our front-end peer. * * \param dev NewBus device object representing this Xen * Net Back instance. * \param frontend_state The new state of the front-end. * * \return 0 for success, errno codes for failure. */ static void xnb_frontend_changed(device_t dev, XenbusState frontend_state) { struct xnb_softc *xnb; xnb = device_get_softc(dev); DPRINTF("frontend_state=%s, xnb_state=%s\n", xenbus_strstate(frontend_state), xenbus_strstate(xenbus_get_state(xnb->dev))); switch (frontend_state) { case XenbusStateInitialising: break; case XenbusStateInitialised: case XenbusStateConnected: xnb_connect(xnb); break; case XenbusStateClosing: case XenbusStateClosed: mtx_lock(&xnb->sc_lock); xnb_shutdown(xnb); mtx_unlock(&xnb->sc_lock); if (frontend_state == XenbusStateClosed) xenbus_set_state(xnb->dev, XenbusStateClosed); break; default: xenbus_dev_fatal(xnb->dev, EINVAL, "saw state %d at frontend", frontend_state); break; } } /*---------------------------- Request Processing ----------------------------*/ /** * Interrupt handler bound to the shared ring's event channel. * Entry point for the xennet transmit path in netback * Transfers packets from the Xen ring to the host's generic networking stack * * \param arg Callback argument registerd during event channel * binding - the xnb_softc for this instance. */ static void xnb_intr(void *arg) { struct xnb_softc *xnb; struct ifnet *ifp; netif_tx_back_ring_t *txb; RING_IDX req_prod_local; xnb = (struct xnb_softc *)arg; ifp = xnb->xnb_ifp; txb = &xnb->ring_configs[XNB_RING_TYPE_TX].back_ring.tx_ring; mtx_lock(&xnb->tx_lock); do { int notify; req_prod_local = txb->sring->req_prod; xen_rmb(); for (;;) { struct mbuf *mbufc; int err; err = xnb_recv(txb, xnb->otherend_id, &mbufc, ifp, xnb->tx_gnttab); if (err || (mbufc == NULL)) break; /* Send the packet to the generic network stack */ (*xnb->xnb_ifp->if_input)(xnb->xnb_ifp, mbufc); } RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(txb, notify); if (notify != 0) xen_intr_signal(xnb->xen_intr_handle); txb->sring->req_event = txb->req_cons + 1; xen_mb(); } while (txb->sring->req_prod != req_prod_local) ; mtx_unlock(&xnb->tx_lock); xnb_start(ifp); } /** * Build a struct xnb_pkt based on netif_tx_request's from a netif tx ring. * Will read exactly 0 or 1 packets from the ring; never a partial packet. * \param[out] pkt The returned packet. If there is an error building * the packet, pkt.list_len will be set to 0. * \param[in] tx_ring Pointer to the Ring that is the input to this function * \param[in] start The ring index of the first potential request * \return The number of requests consumed to build this packet */ static int xnb_ring2pkt(struct xnb_pkt *pkt, const netif_tx_back_ring_t *tx_ring, RING_IDX start) { /* * Outline: * 1) Initialize pkt * 2) Read the first request of the packet * 3) Read the extras * 4) Set cdr * 5) Loop on the remainder of the packet * 6) Finalize pkt (stuff like car_size and list_len) */ int idx = start; int discard = 0; /* whether to discard the packet */ int more_data = 0; /* there are more request past the last one */ uint16_t cdr_size = 0; /* accumulated size of requests 2 through n */ xnb_pkt_initialize(pkt); /* Read the first request */ if (RING_HAS_UNCONSUMED_REQUESTS_2(tx_ring, idx)) { netif_tx_request_t *tx = RING_GET_REQUEST(tx_ring, idx); pkt->size = tx->size; pkt->flags = tx->flags & ~NETTXF_more_data; more_data = tx->flags & NETTXF_more_data; pkt->list_len++; pkt->car = idx; idx++; } /* Read the extra info */ if ((pkt->flags & NETTXF_extra_info) && RING_HAS_UNCONSUMED_REQUESTS_2(tx_ring, idx)) { netif_extra_info_t *ext = (netif_extra_info_t*) RING_GET_REQUEST(tx_ring, idx); pkt->extra.type = ext->type; switch (pkt->extra.type) { case XEN_NETIF_EXTRA_TYPE_GSO: pkt->extra.u.gso = ext->u.gso; break; default: /* * The reference Linux netfront driver will * never set any other extra.type. So we don't * know what to do with it. Let's print an * error, then consume and discard the packet */ printf("xnb(%s:%d): Unknown extra info type %d." " Discarding packet\n", __func__, __LINE__, pkt->extra.type); xnb_dump_txreq(start, RING_GET_REQUEST(tx_ring, start)); xnb_dump_txreq(idx, RING_GET_REQUEST(tx_ring, idx)); discard = 1; break; } pkt->extra.flags = ext->flags; if (ext->flags & XEN_NETIF_EXTRA_FLAG_MORE) { /* * The reference linux netfront driver never sets this * flag (nor does any other known netfront). So we * will discard the packet. */ printf("xnb(%s:%d): Request sets " "XEN_NETIF_EXTRA_FLAG_MORE, but we can't handle " "that\n", __func__, __LINE__); xnb_dump_txreq(start, RING_GET_REQUEST(tx_ring, start)); xnb_dump_txreq(idx, RING_GET_REQUEST(tx_ring, idx)); discard = 1; } idx++; } /* Set cdr. If there is not more data, cdr is invalid */ pkt->cdr = idx; /* Loop on remainder of packet */ while (more_data && RING_HAS_UNCONSUMED_REQUESTS_2(tx_ring, idx)) { netif_tx_request_t *tx = RING_GET_REQUEST(tx_ring, idx); pkt->list_len++; cdr_size += tx->size; if (tx->flags & ~NETTXF_more_data) { /* There should be no other flags set at this point */ printf("xnb(%s:%d): Request sets unknown flags %d " "after the 1st request in the packet.\n", __func__, __LINE__, tx->flags); xnb_dump_txreq(start, RING_GET_REQUEST(tx_ring, start)); xnb_dump_txreq(idx, RING_GET_REQUEST(tx_ring, idx)); } more_data = tx->flags & NETTXF_more_data; idx++; } /* Finalize packet */ if (more_data != 0) { /* The ring ran out of requests before finishing the packet */ xnb_pkt_invalidate(pkt); idx = start; /* tell caller that we consumed no requests */ } else { /* Calculate car_size */ pkt->car_size = pkt->size - cdr_size; } if (discard != 0) { xnb_pkt_invalidate(pkt); } return idx - start; } /** * Respond to all the requests that constituted pkt. Builds the responses and * writes them to the ring, but doesn't push them to the shared ring. * \param[in] pkt the packet that needs a response * \param[in] error true if there was an error handling the packet, such * as in the hypervisor copy op or mbuf allocation * \param[out] ring Responses go here */ static void xnb_txpkt2rsp(const struct xnb_pkt *pkt, netif_tx_back_ring_t *ring, int error) { /* * Outline: * 1) Respond to the first request * 2) Respond to the extra info reques * Loop through every remaining request in the packet, generating * responses that copy those requests' ids and sets the status * appropriately. */ netif_tx_request_t *tx; netif_tx_response_t *rsp; int i; uint16_t status; status = (xnb_pkt_is_valid(pkt) == 0) || error ? NETIF_RSP_ERROR : NETIF_RSP_OKAY; KASSERT((pkt->list_len == 0) || (ring->rsp_prod_pvt == pkt->car), ("Cannot respond to ring requests out of order")); if (pkt->list_len >= 1) { uint16_t id; tx = RING_GET_REQUEST(ring, ring->rsp_prod_pvt); id = tx->id; rsp = RING_GET_RESPONSE(ring, ring->rsp_prod_pvt); rsp->id = id; rsp->status = status; ring->rsp_prod_pvt++; if (pkt->flags & NETRXF_extra_info) { rsp = RING_GET_RESPONSE(ring, ring->rsp_prod_pvt); rsp->status = NETIF_RSP_NULL; ring->rsp_prod_pvt++; } } for (i=0; i < pkt->list_len - 1; i++) { uint16_t id; tx = RING_GET_REQUEST(ring, ring->rsp_prod_pvt); id = tx->id; rsp = RING_GET_RESPONSE(ring, ring->rsp_prod_pvt); rsp->id = id; rsp->status = status; ring->rsp_prod_pvt++; } } /** * Create an mbuf chain to represent a packet. Initializes all of the headers * in the mbuf chain, but does not copy the data. The returned chain must be * free()'d when no longer needed * \param[in] pkt A packet to model the mbuf chain after * \return A newly allocated mbuf chain, possibly with clusters attached. * NULL on failure */ static struct mbuf* xnb_pkt2mbufc(const struct xnb_pkt *pkt, struct ifnet *ifp) { /** * \todo consider using a memory pool for mbufs instead of * reallocating them for every packet */ /** \todo handle extra data */ struct mbuf *m; m = m_getm(NULL, pkt->size, M_NOWAIT, MT_DATA); if (m != NULL) { m->m_pkthdr.rcvif = ifp; if (pkt->flags & NETTXF_data_validated) { /* * We lie to the host OS and always tell it that the * checksums are ok, because the packet is unlikely to * get corrupted going across domains. */ m->m_pkthdr.csum_flags = ( CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID | CSUM_PSEUDO_HDR ); m->m_pkthdr.csum_data = 0xffff; } } return m; } /** * Build a gnttab_copy table that can be used to copy data from a pkt * to an mbufc. Does not actually perform the copy. Always uses gref's on * the packet side. * \param[in] pkt pkt's associated requests form the src for * the copy operation * \param[in] mbufc mbufc's storage forms the dest for the copy operation * \param[out] gnttab Storage for the returned grant table * \param[in] txb Pointer to the backend ring structure * \param[in] otherend_id The domain ID of the other end of the copy * \return The number of gnttab entries filled */ static int -xnb_txpkt2gnttab(const struct xnb_pkt *pkt, const struct mbuf *mbufc, +xnb_txpkt2gnttab(const struct xnb_pkt *pkt, struct mbuf *mbufc, gnttab_copy_table gnttab, const netif_tx_back_ring_t *txb, domid_t otherend_id) { - const struct mbuf *mbuf = mbufc;/* current mbuf within the chain */ + struct mbuf *mbuf = mbufc;/* current mbuf within the chain */ int gnt_idx = 0; /* index into grant table */ RING_IDX r_idx = pkt->car; /* index into tx ring buffer */ int r_ofs = 0; /* offset of next data within tx request's data area */ int m_ofs = 0; /* offset of next data within mbuf's data area */ /* size in bytes that still needs to be represented in the table */ uint16_t size_remaining = pkt->size; while (size_remaining > 0) { const netif_tx_request_t *txq = RING_GET_REQUEST(txb, r_idx); const size_t mbuf_space = M_TRAILINGSPACE(mbuf) - m_ofs; const size_t req_size = r_idx == pkt->car ? pkt->car_size : txq->size; const size_t pkt_space = req_size - r_ofs; /* * space is the largest amount of data that can be copied in the * grant table's next entry */ const size_t space = MIN(pkt_space, mbuf_space); /* TODO: handle this error condition without panicking */ KASSERT(gnt_idx < GNTTAB_LEN, ("Grant table is too short")); gnttab[gnt_idx].source.u.ref = txq->gref; gnttab[gnt_idx].source.domid = otherend_id; gnttab[gnt_idx].source.offset = txq->offset + r_ofs; gnttab[gnt_idx].dest.u.gmfn = virt_to_mfn( mtod(mbuf, vm_offset_t) + m_ofs); gnttab[gnt_idx].dest.offset = virt_to_offset( mtod(mbuf, vm_offset_t) + m_ofs); gnttab[gnt_idx].dest.domid = DOMID_SELF; gnttab[gnt_idx].len = space; gnttab[gnt_idx].flags = GNTCOPY_source_gref; gnt_idx++; r_ofs += space; m_ofs += space; size_remaining -= space; if (req_size - r_ofs <= 0) { /* Must move to the next tx request */ r_ofs = 0; r_idx = (r_idx == pkt->car) ? pkt->cdr : r_idx + 1; } if (M_TRAILINGSPACE(mbuf) - m_ofs <= 0) { /* Must move to the next mbuf */ m_ofs = 0; mbuf = mbuf->m_next; } } return gnt_idx; } /** * Check the status of the grant copy operations, and update mbufs various * non-data fields to reflect the data present. * \param[in,out] mbufc mbuf chain to update. The chain must be valid and of * the correct length, and data should already be present * \param[in] gnttab A grant table for a just completed copy op * \param[in] n_entries The number of valid entries in the grant table */ static void xnb_update_mbufc(struct mbuf *mbufc, const gnttab_copy_table gnttab, int n_entries) { struct mbuf *mbuf = mbufc; int i; size_t total_size = 0; for (i = 0; i < n_entries; i++) { KASSERT(gnttab[i].status == GNTST_okay, ("Some gnttab_copy entry had error status %hd\n", gnttab[i].status)); mbuf->m_len += gnttab[i].len; total_size += gnttab[i].len; if (M_TRAILINGSPACE(mbuf) <= 0) { mbuf = mbuf->m_next; } } mbufc->m_pkthdr.len = total_size; #if defined(INET) || defined(INET6) xnb_add_mbuf_cksum(mbufc); #endif } /** * Dequeue at most one packet from the shared ring * \param[in,out] txb Netif tx ring. A packet will be removed from it, and * its private indices will be updated. But the indices * will not be pushed to the shared ring. * \param[in] ifnet Interface to which the packet will be sent * \param[in] otherend Domain ID of the other end of the ring * \param[out] mbufc The assembled mbuf chain, ready to send to the generic * networking stack * \param[in,out] gnttab Pointer to enough memory for a grant table. We make * this a function parameter so that we will take less * stack space. * \return An error code */ static int xnb_recv(netif_tx_back_ring_t *txb, domid_t otherend, struct mbuf **mbufc, struct ifnet *ifnet, gnttab_copy_table gnttab) { struct xnb_pkt pkt; /* number of tx requests consumed to build the last packet */ int num_consumed; int nr_ents; *mbufc = NULL; num_consumed = xnb_ring2pkt(&pkt, txb, txb->req_cons); if (num_consumed == 0) return 0; /* Nothing to receive */ /* update statistics independent of errors */ if_inc_counter(ifnet, IFCOUNTER_IPACKETS, 1); /* * if we got here, then 1 or more requests was consumed, but the packet * is not necessarily valid. */ if (xnb_pkt_is_valid(&pkt) == 0) { /* got a garbage packet, respond and drop it */ xnb_txpkt2rsp(&pkt, txb, 1); txb->req_cons += num_consumed; DPRINTF("xnb_intr: garbage packet, num_consumed=%d\n", num_consumed); if_inc_counter(ifnet, IFCOUNTER_IERRORS, 1); return EINVAL; } *mbufc = xnb_pkt2mbufc(&pkt, ifnet); if (*mbufc == NULL) { /* * Couldn't allocate mbufs. Respond and drop the packet. Do * not consume the requests */ xnb_txpkt2rsp(&pkt, txb, 1); DPRINTF("xnb_intr: Couldn't allocate mbufs, num_consumed=%d\n", num_consumed); if_inc_counter(ifnet, IFCOUNTER_IQDROPS, 1); return ENOMEM; } nr_ents = xnb_txpkt2gnttab(&pkt, *mbufc, gnttab, txb, otherend); if (nr_ents > 0) { int __unused hv_ret = HYPERVISOR_grant_table_op(GNTTABOP_copy, gnttab, nr_ents); KASSERT(hv_ret == 0, ("HYPERVISOR_grant_table_op returned %d\n", hv_ret)); xnb_update_mbufc(*mbufc, gnttab, nr_ents); } xnb_txpkt2rsp(&pkt, txb, 0); txb->req_cons += num_consumed; return 0; } /** * Create an xnb_pkt based on the contents of an mbuf chain. * \param[in] mbufc mbuf chain to transform into a packet * \param[out] pkt Storage for the newly generated xnb_pkt * \param[in] start The ring index of the first available slot in the rx * ring * \param[in] space The number of free slots in the rx ring * \retval 0 Success * \retval EINVAL mbufc was corrupt or not convertible into a pkt * \retval EAGAIN There was not enough space in the ring to queue the * packet */ static int xnb_mbufc2pkt(const struct mbuf *mbufc, struct xnb_pkt *pkt, RING_IDX start, int space) { int retval = 0; if ((mbufc == NULL) || ( (mbufc->m_flags & M_PKTHDR) == 0) || (mbufc->m_pkthdr.len == 0)) { xnb_pkt_invalidate(pkt); retval = EINVAL; } else { int slots_required; xnb_pkt_validate(pkt); pkt->flags = 0; pkt->size = mbufc->m_pkthdr.len; pkt->car = start; pkt->car_size = mbufc->m_len; if (mbufc->m_pkthdr.csum_flags & CSUM_TSO) { pkt->flags |= NETRXF_extra_info; pkt->extra.u.gso.size = mbufc->m_pkthdr.tso_segsz; pkt->extra.u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4; pkt->extra.u.gso.pad = 0; pkt->extra.u.gso.features = 0; pkt->extra.type = XEN_NETIF_EXTRA_TYPE_GSO; pkt->extra.flags = 0; pkt->cdr = start + 2; } else { pkt->cdr = start + 1; } if (mbufc->m_pkthdr.csum_flags & (CSUM_TSO | CSUM_DELAY_DATA)) { pkt->flags |= (NETRXF_csum_blank | NETRXF_data_validated); } /* * Each ring response can have up to PAGE_SIZE of data. * Assume that we can defragment the mbuf chain efficiently * into responses so that each response but the last uses all * PAGE_SIZE bytes. */ pkt->list_len = (pkt->size + PAGE_SIZE - 1) / PAGE_SIZE; if (pkt->list_len > 1) { pkt->flags |= NETRXF_more_data; } slots_required = pkt->list_len + (pkt->flags & NETRXF_extra_info ? 1 : 0); if (slots_required > space) { xnb_pkt_invalidate(pkt); retval = EAGAIN; } } return retval; } /** * Build a gnttab_copy table that can be used to copy data from an mbuf chain * to the frontend's shared buffers. Does not actually perform the copy. * Always uses gref's on the other end's side. * \param[in] pkt pkt's associated responses form the dest for the copy * operatoin * \param[in] mbufc The source for the copy operation * \param[out] gnttab Storage for the returned grant table * \param[in] rxb Pointer to the backend ring structure * \param[in] otherend_id The domain ID of the other end of the copy * \return The number of gnttab entries filled */ static int xnb_rxpkt2gnttab(const struct xnb_pkt *pkt, const struct mbuf *mbufc, gnttab_copy_table gnttab, const netif_rx_back_ring_t *rxb, domid_t otherend_id) { const struct mbuf *mbuf = mbufc;/* current mbuf within the chain */ int gnt_idx = 0; /* index into grant table */ RING_IDX r_idx = pkt->car; /* index into rx ring buffer */ int r_ofs = 0; /* offset of next data within rx request's data area */ int m_ofs = 0; /* offset of next data within mbuf's data area */ /* size in bytes that still needs to be represented in the table */ uint16_t size_remaining; size_remaining = (xnb_pkt_is_valid(pkt) != 0) ? pkt->size : 0; while (size_remaining > 0) { const netif_rx_request_t *rxq = RING_GET_REQUEST(rxb, r_idx); const size_t mbuf_space = mbuf->m_len - m_ofs; /* Xen shared pages have an implied size of PAGE_SIZE */ const size_t req_size = PAGE_SIZE; const size_t pkt_space = req_size - r_ofs; /* * space is the largest amount of data that can be copied in the * grant table's next entry */ const size_t space = MIN(pkt_space, mbuf_space); /* TODO: handle this error condition without panicing */ KASSERT(gnt_idx < GNTTAB_LEN, ("Grant table is too short")); gnttab[gnt_idx].dest.u.ref = rxq->gref; gnttab[gnt_idx].dest.domid = otherend_id; gnttab[gnt_idx].dest.offset = r_ofs; gnttab[gnt_idx].source.u.gmfn = virt_to_mfn( mtod(mbuf, vm_offset_t) + m_ofs); gnttab[gnt_idx].source.offset = virt_to_offset( mtod(mbuf, vm_offset_t) + m_ofs); gnttab[gnt_idx].source.domid = DOMID_SELF; gnttab[gnt_idx].len = space; gnttab[gnt_idx].flags = GNTCOPY_dest_gref; gnt_idx++; r_ofs += space; m_ofs += space; size_remaining -= space; if (req_size - r_ofs <= 0) { /* Must move to the next rx request */ r_ofs = 0; r_idx = (r_idx == pkt->car) ? pkt->cdr : r_idx + 1; } if (mbuf->m_len - m_ofs <= 0) { /* Must move to the next mbuf */ m_ofs = 0; mbuf = mbuf->m_next; } } return gnt_idx; } /** * Generates responses for all the requests that constituted pkt. Builds * responses and writes them to the ring, but doesn't push the shared ring * indices. * \param[in] pkt the packet that needs a response * \param[in] gnttab The grant copy table corresponding to this packet. * Used to determine how many rsp->netif_rx_response_t's to * generate. * \param[in] n_entries Number of relevant entries in the grant table * \param[out] ring Responses go here * \return The number of RX requests that were consumed to generate * the responses */ static int xnb_rxpkt2rsp(const struct xnb_pkt *pkt, const gnttab_copy_table gnttab, int n_entries, netif_rx_back_ring_t *ring) { /* * This code makes the following assumptions: * * All entries in gnttab set GNTCOPY_dest_gref * * The entries in gnttab are grouped by their grefs: any two * entries with the same gref must be adjacent */ int error = 0; int gnt_idx, i; int n_responses = 0; grant_ref_t last_gref = GRANT_REF_INVALID; RING_IDX r_idx; KASSERT(gnttab != NULL, ("Received a null granttable copy")); /* * In the event of an error, we only need to send one response to the * netfront. In that case, we musn't write any data to the responses * after the one we send. So we must loop all the way through gnttab * looking for errors before we generate any responses * * Since we're looping through the grant table anyway, we'll count the * number of different gref's in it, which will tell us how many * responses to generate */ for (gnt_idx = 0; gnt_idx < n_entries; gnt_idx++) { int16_t status = gnttab[gnt_idx].status; if (status != GNTST_okay) { DPRINTF( "Got error %d for hypervisor gnttab_copy status\n", status); error = 1; break; } if (gnttab[gnt_idx].dest.u.ref != last_gref) { n_responses++; last_gref = gnttab[gnt_idx].dest.u.ref; } } if (error != 0) { uint16_t id; netif_rx_response_t *rsp; id = RING_GET_REQUEST(ring, ring->rsp_prod_pvt)->id; rsp = RING_GET_RESPONSE(ring, ring->rsp_prod_pvt); rsp->id = id; rsp->status = NETIF_RSP_ERROR; n_responses = 1; } else { gnt_idx = 0; const int has_extra = pkt->flags & NETRXF_extra_info; if (has_extra != 0) n_responses++; for (i = 0; i < n_responses; i++) { netif_rx_request_t rxq; netif_rx_response_t *rsp; r_idx = ring->rsp_prod_pvt + i; /* * We copy the structure of rxq instead of making a * pointer because it shares the same memory as rsp. */ rxq = *(RING_GET_REQUEST(ring, r_idx)); rsp = RING_GET_RESPONSE(ring, r_idx); if (has_extra && (i == 1)) { netif_extra_info_t *ext = (netif_extra_info_t*)rsp; ext->type = XEN_NETIF_EXTRA_TYPE_GSO; ext->flags = 0; ext->u.gso.size = pkt->extra.u.gso.size; ext->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4; ext->u.gso.pad = 0; ext->u.gso.features = 0; } else { rsp->id = rxq.id; rsp->status = GNTST_okay; rsp->offset = 0; rsp->flags = 0; if (i < pkt->list_len - 1) rsp->flags |= NETRXF_more_data; if ((i == 0) && has_extra) rsp->flags |= NETRXF_extra_info; if ((i == 0) && (pkt->flags & NETRXF_data_validated)) { rsp->flags |= NETRXF_data_validated; rsp->flags |= NETRXF_csum_blank; } rsp->status = 0; for (; gnttab[gnt_idx].dest.u.ref == rxq.gref; gnt_idx++) { rsp->status += gnttab[gnt_idx].len; } } } } ring->req_cons += n_responses; ring->rsp_prod_pvt += n_responses; return n_responses; } #if defined(INET) || defined(INET6) /** * Add IP, TCP, and/or UDP checksums to every mbuf in a chain. The first mbuf * in the chain must start with a struct ether_header. * * XXX This function will perform incorrectly on UDP packets that are split up * into multiple ethernet frames. */ static void xnb_add_mbuf_cksum(struct mbuf *mbufc) { struct ether_header *eh; struct ip *iph; uint16_t ether_type; eh = mtod(mbufc, struct ether_header*); ether_type = ntohs(eh->ether_type); if (ether_type != ETHERTYPE_IP) { /* Nothing to calculate */ return; } iph = (struct ip*)(eh + 1); if (mbufc->m_pkthdr.csum_flags & CSUM_IP_VALID) { iph->ip_sum = 0; iph->ip_sum = in_cksum_hdr(iph); } switch (iph->ip_p) { case IPPROTO_TCP: if (mbufc->m_pkthdr.csum_flags & CSUM_IP_VALID) { size_t tcplen = ntohs(iph->ip_len) - sizeof(struct ip); struct tcphdr *th = (struct tcphdr*)(iph + 1); th->th_sum = in_pseudo(iph->ip_src.s_addr, iph->ip_dst.s_addr, htons(IPPROTO_TCP + tcplen)); th->th_sum = in_cksum_skip(mbufc, sizeof(struct ether_header) + ntohs(iph->ip_len), sizeof(struct ether_header) + (iph->ip_hl << 2)); } break; case IPPROTO_UDP: if (mbufc->m_pkthdr.csum_flags & CSUM_IP_VALID) { size_t udplen = ntohs(iph->ip_len) - sizeof(struct ip); struct udphdr *uh = (struct udphdr*)(iph + 1); uh->uh_sum = in_pseudo(iph->ip_src.s_addr, iph->ip_dst.s_addr, htons(IPPROTO_UDP + udplen)); uh->uh_sum = in_cksum_skip(mbufc, sizeof(struct ether_header) + ntohs(iph->ip_len), sizeof(struct ether_header) + (iph->ip_hl << 2)); } break; default: break; } } #endif /* INET || INET6 */ static void xnb_stop(struct xnb_softc *xnb) { struct ifnet *ifp; mtx_assert(&xnb->sc_lock, MA_OWNED); ifp = xnb->xnb_ifp; ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); if_link_state_change(ifp, LINK_STATE_DOWN); } static int xnb_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct xnb_softc *xnb = ifp->if_softc; struct ifreq *ifr = (struct ifreq*) data; #ifdef INET struct ifaddr *ifa = (struct ifaddr*)data; #endif int error = 0; switch (cmd) { case SIOCSIFFLAGS: mtx_lock(&xnb->sc_lock); if (ifp->if_flags & IFF_UP) { xnb_ifinit_locked(xnb); } else { if (ifp->if_drv_flags & IFF_DRV_RUNNING) { xnb_stop(xnb); } } /* * Note: netfront sets a variable named xn_if_flags * here, but that variable is never read */ mtx_unlock(&xnb->sc_lock); break; case SIOCSIFADDR: #ifdef INET mtx_lock(&xnb->sc_lock); if (ifa->ifa_addr->sa_family == AF_INET) { ifp->if_flags |= IFF_UP; if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) { ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); if_link_state_change(ifp, LINK_STATE_DOWN); ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; if_link_state_change(ifp, LINK_STATE_UP); } arp_ifinit(ifp, ifa); mtx_unlock(&xnb->sc_lock); } else { mtx_unlock(&xnb->sc_lock); #endif error = ether_ioctl(ifp, cmd, data); #ifdef INET } #endif break; case SIOCSIFCAP: mtx_lock(&xnb->sc_lock); if (ifr->ifr_reqcap & IFCAP_TXCSUM) { ifp->if_capenable |= IFCAP_TXCSUM; ifp->if_hwassist |= XNB_CSUM_FEATURES; } else { ifp->if_capenable &= ~(IFCAP_TXCSUM); ifp->if_hwassist &= ~(XNB_CSUM_FEATURES); } if ((ifr->ifr_reqcap & IFCAP_RXCSUM)) { ifp->if_capenable |= IFCAP_RXCSUM; } else { ifp->if_capenable &= ~(IFCAP_RXCSUM); } /* * TODO enable TSO4 and LRO once we no longer need * to calculate checksums in software */ #if 0 if (ifr->if_reqcap |= IFCAP_TSO4) { if (IFCAP_TXCSUM & ifp->if_capenable) { printf("xnb: Xen netif requires that " "TXCSUM be enabled in order " "to use TSO4\n"); error = EINVAL; } else { ifp->if_capenable |= IFCAP_TSO4; ifp->if_hwassist |= CSUM_TSO; } } else { ifp->if_capenable &= ~(IFCAP_TSO4); ifp->if_hwassist &= ~(CSUM_TSO); } if (ifr->ifreqcap |= IFCAP_LRO) { ifp->if_capenable |= IFCAP_LRO; } else { ifp->if_capenable &= ~(IFCAP_LRO); } #endif mtx_unlock(&xnb->sc_lock); break; case SIOCSIFMTU: ifp->if_mtu = ifr->ifr_mtu; ifp->if_drv_flags &= ~IFF_DRV_RUNNING; xnb_ifinit(xnb); break; case SIOCADDMULTI: case SIOCDELMULTI: case SIOCSIFMEDIA: case SIOCGIFMEDIA: error = ifmedia_ioctl(ifp, ifr, &xnb->sc_media, cmd); break; default: error = ether_ioctl(ifp, cmd, data); break; } return (error); } static void xnb_start_locked(struct ifnet *ifp) { netif_rx_back_ring_t *rxb; struct xnb_softc *xnb; struct mbuf *mbufc; RING_IDX req_prod_local; xnb = ifp->if_softc; rxb = &xnb->ring_configs[XNB_RING_TYPE_RX].back_ring.rx_ring; if (!xnb->carrier) return; do { int out_of_space = 0; int notify; req_prod_local = rxb->sring->req_prod; xen_rmb(); for (;;) { int error; IF_DEQUEUE(&ifp->if_snd, mbufc); if (mbufc == NULL) break; error = xnb_send(rxb, xnb->otherend_id, mbufc, xnb->rx_gnttab); switch (error) { case EAGAIN: /* * Insufficient space in the ring. * Requeue pkt and send when space is * available. */ IF_PREPEND(&ifp->if_snd, mbufc); /* * Perhaps the frontend missed an IRQ * and went to sleep. Notify it to wake * it up. */ out_of_space = 1; break; case EINVAL: /* OS gave a corrupt packet. Drop it.*/ if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); /* FALLTHROUGH */ default: /* Send succeeded, or packet had error. * Free the packet */ if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); if (mbufc) m_freem(mbufc); break; } if (out_of_space != 0) break; } RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(rxb, notify); if ((notify != 0) || (out_of_space != 0)) xen_intr_signal(xnb->xen_intr_handle); rxb->sring->req_event = req_prod_local + 1; xen_mb(); } while (rxb->sring->req_prod != req_prod_local) ; } /** * Sends one packet to the ring. Blocks until the packet is on the ring * \param[in] mbufc Contains one packet to send. Caller must free * \param[in,out] rxb The packet will be pushed onto this ring, but the * otherend will not be notified. * \param[in] otherend The domain ID of the other end of the connection * \retval EAGAIN The ring did not have enough space for the packet. * The ring has not been modified * \param[in,out] gnttab Pointer to enough memory for a grant table. We make * this a function parameter so that we will take less * stack space. * \retval EINVAL mbufc was corrupt or not convertible into a pkt */ static int xnb_send(netif_rx_back_ring_t *ring, domid_t otherend, const struct mbuf *mbufc, gnttab_copy_table gnttab) { struct xnb_pkt pkt; int error, n_entries, n_reqs; RING_IDX space; space = ring->sring->req_prod - ring->req_cons; error = xnb_mbufc2pkt(mbufc, &pkt, ring->rsp_prod_pvt, space); if (error != 0) return error; n_entries = xnb_rxpkt2gnttab(&pkt, mbufc, gnttab, ring, otherend); if (n_entries != 0) { int __unused hv_ret = HYPERVISOR_grant_table_op(GNTTABOP_copy, gnttab, n_entries); KASSERT(hv_ret == 0, ("HYPERVISOR_grant_table_op returned %d\n", hv_ret)); } n_reqs = xnb_rxpkt2rsp(&pkt, gnttab, n_entries, ring); return 0; } static void xnb_start(struct ifnet *ifp) { struct xnb_softc *xnb; xnb = ifp->if_softc; mtx_lock(&xnb->rx_lock); xnb_start_locked(ifp); mtx_unlock(&xnb->rx_lock); } /* equivalent of network_open() in Linux */ static void xnb_ifinit_locked(struct xnb_softc *xnb) { struct ifnet *ifp; ifp = xnb->xnb_ifp; mtx_assert(&xnb->sc_lock, MA_OWNED); if (ifp->if_drv_flags & IFF_DRV_RUNNING) return; xnb_stop(xnb); ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; if_link_state_change(ifp, LINK_STATE_UP); } static void xnb_ifinit(void *xsc) { struct xnb_softc *xnb = xsc; mtx_lock(&xnb->sc_lock); xnb_ifinit_locked(xnb); mtx_unlock(&xnb->sc_lock); } /** * Callback used by the generic networking code to tell us when our carrier * state has changed. Since we don't have a physical carrier, we don't care */ static int xnb_ifmedia_upd(struct ifnet *ifp) { return (0); } /** * Callback used by the generic networking code to ask us what our carrier * state is. Since we don't have a physical carrier, this is very simple */ static void xnb_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) { ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE; ifmr->ifm_active = IFM_ETHER|IFM_MANUAL; } /*---------------------------- NewBus Registration ---------------------------*/ static device_method_t xnb_methods[] = { /* Device interface */ DEVMETHOD(device_probe, xnb_probe), DEVMETHOD(device_attach, xnb_attach), DEVMETHOD(device_detach, xnb_detach), DEVMETHOD(device_shutdown, bus_generic_shutdown), DEVMETHOD(device_suspend, xnb_suspend), DEVMETHOD(device_resume, xnb_resume), /* Xenbus interface */ DEVMETHOD(xenbus_otherend_changed, xnb_frontend_changed), { 0, 0 } }; static driver_t xnb_driver = { "xnb", xnb_methods, sizeof(struct xnb_softc), }; devclass_t xnb_devclass; DRIVER_MODULE(xnb, xenbusb_back, xnb_driver, xnb_devclass, 0, 0); /*-------------------------- Unit Tests -------------------------------------*/ #ifdef XNB_DEBUG #include "netback_unit_tests.c" #endif Index: head/sys/kern/kern_mbuf.c =================================================================== --- head/sys/kern/kern_mbuf.c (revision 296241) +++ head/sys/kern/kern_mbuf.c (revision 296242) @@ -1,984 +1,935 @@ /*- * Copyright (c) 2004, 2005, * Bosko Milekic . 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 unmodified, 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 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 AUTHOR 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. */ #include __FBSDID("$FreeBSD$"); #include "opt_param.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * In FreeBSD, Mbufs and Mbuf Clusters are allocated from UMA * Zones. * * Mbuf Clusters (2K, contiguous) are allocated from the Cluster * Zone. The Zone can be capped at kern.ipc.nmbclusters, if the * administrator so desires. * * Mbufs are allocated from a UMA Master Zone called the Mbuf * Zone. * * Additionally, FreeBSD provides a Packet Zone, which it * configures as a Secondary Zone to the Mbuf Master Zone, * thus sharing backend Slab kegs with the Mbuf Master Zone. * * Thus common-case allocations and locking are simplified: * * m_clget() m_getcl() * | | * | .------------>[(Packet Cache)] m_get(), m_gethdr() * | | [ Packet ] | * [(Cluster Cache)] [ Secondary ] [ (Mbuf Cache) ] * [ Cluster Zone ] [ Zone ] [ Mbuf Master Zone ] * | \________ | * [ Cluster Keg ] \ / * | [ Mbuf Keg ] * [ Cluster Slabs ] | * | [ Mbuf Slabs ] * \____________(VM)_________________/ * * * Whenever an object is allocated with uma_zalloc() out of * one of the Zones its _ctor_ function is executed. The same * for any deallocation through uma_zfree() the _dtor_ function * is executed. * * Caches are per-CPU and are filled from the Master Zone. * * Whenever an object is allocated from the underlying global * memory pool it gets pre-initialized with the _zinit_ functions. * When the Keg's are overfull objects get decomissioned with * _zfini_ functions and free'd back to the global memory pool. * */ int nmbufs; /* limits number of mbufs */ int nmbclusters; /* limits number of mbuf clusters */ int nmbjumbop; /* limits number of page size jumbo clusters */ int nmbjumbo9; /* limits number of 9k jumbo clusters */ int nmbjumbo16; /* limits number of 16k jumbo clusters */ static quad_t maxmbufmem; /* overall real memory limit for all mbufs */ SYSCTL_QUAD(_kern_ipc, OID_AUTO, maxmbufmem, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &maxmbufmem, 0, "Maximum real memory allocatable to various mbuf types"); /* * tunable_mbinit() has to be run before any mbuf allocations are done. */ static void tunable_mbinit(void *dummy) { quad_t realmem; /* * The default limit for all mbuf related memory is 1/2 of all * available kernel memory (physical or kmem). * At most it can be 3/4 of available kernel memory. */ realmem = qmin((quad_t)physmem * PAGE_SIZE, vm_kmem_size); maxmbufmem = realmem / 2; TUNABLE_QUAD_FETCH("kern.ipc.maxmbufmem", &maxmbufmem); if (maxmbufmem > realmem / 4 * 3) maxmbufmem = realmem / 4 * 3; TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters); if (nmbclusters == 0) nmbclusters = maxmbufmem / MCLBYTES / 4; TUNABLE_INT_FETCH("kern.ipc.nmbjumbop", &nmbjumbop); if (nmbjumbop == 0) nmbjumbop = maxmbufmem / MJUMPAGESIZE / 4; TUNABLE_INT_FETCH("kern.ipc.nmbjumbo9", &nmbjumbo9); if (nmbjumbo9 == 0) nmbjumbo9 = maxmbufmem / MJUM9BYTES / 6; TUNABLE_INT_FETCH("kern.ipc.nmbjumbo16", &nmbjumbo16); if (nmbjumbo16 == 0) nmbjumbo16 = maxmbufmem / MJUM16BYTES / 6; /* * We need at least as many mbufs as we have clusters of * the various types added together. */ TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs); if (nmbufs < nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) nmbufs = lmax(maxmbufmem / MSIZE / 5, nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16); } SYSINIT(tunable_mbinit, SI_SUB_KMEM, SI_ORDER_MIDDLE, tunable_mbinit, NULL); static int sysctl_nmbclusters(SYSCTL_HANDLER_ARGS) { int error, newnmbclusters; newnmbclusters = nmbclusters; error = sysctl_handle_int(oidp, &newnmbclusters, 0, req); if (error == 0 && req->newptr && newnmbclusters != nmbclusters) { if (newnmbclusters > nmbclusters && nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { nmbclusters = newnmbclusters; nmbclusters = uma_zone_set_max(zone_clust, nmbclusters); EVENTHANDLER_INVOKE(nmbclusters_change); } else error = EINVAL; } return (error); } SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters, CTLTYPE_INT|CTLFLAG_RW, &nmbclusters, 0, sysctl_nmbclusters, "IU", "Maximum number of mbuf clusters allowed"); static int sysctl_nmbjumbop(SYSCTL_HANDLER_ARGS) { int error, newnmbjumbop; newnmbjumbop = nmbjumbop; error = sysctl_handle_int(oidp, &newnmbjumbop, 0, req); if (error == 0 && req->newptr && newnmbjumbop != nmbjumbop) { if (newnmbjumbop > nmbjumbop && nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { nmbjumbop = newnmbjumbop; nmbjumbop = uma_zone_set_max(zone_jumbop, nmbjumbop); } else error = EINVAL; } return (error); } SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbop, CTLTYPE_INT|CTLFLAG_RW, &nmbjumbop, 0, sysctl_nmbjumbop, "IU", "Maximum number of mbuf page size jumbo clusters allowed"); static int sysctl_nmbjumbo9(SYSCTL_HANDLER_ARGS) { int error, newnmbjumbo9; newnmbjumbo9 = nmbjumbo9; error = sysctl_handle_int(oidp, &newnmbjumbo9, 0, req); if (error == 0 && req->newptr && newnmbjumbo9 != nmbjumbo9) { if (newnmbjumbo9 > nmbjumbo9 && nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { nmbjumbo9 = newnmbjumbo9; nmbjumbo9 = uma_zone_set_max(zone_jumbo9, nmbjumbo9); } else error = EINVAL; } return (error); } SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo9, CTLTYPE_INT|CTLFLAG_RW, &nmbjumbo9, 0, sysctl_nmbjumbo9, "IU", "Maximum number of mbuf 9k jumbo clusters allowed"); static int sysctl_nmbjumbo16(SYSCTL_HANDLER_ARGS) { int error, newnmbjumbo16; newnmbjumbo16 = nmbjumbo16; error = sysctl_handle_int(oidp, &newnmbjumbo16, 0, req); if (error == 0 && req->newptr && newnmbjumbo16 != nmbjumbo16) { if (newnmbjumbo16 > nmbjumbo16 && nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { nmbjumbo16 = newnmbjumbo16; nmbjumbo16 = uma_zone_set_max(zone_jumbo16, nmbjumbo16); } else error = EINVAL; } return (error); } SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo16, CTLTYPE_INT|CTLFLAG_RW, &nmbjumbo16, 0, sysctl_nmbjumbo16, "IU", "Maximum number of mbuf 16k jumbo clusters allowed"); static int sysctl_nmbufs(SYSCTL_HANDLER_ARGS) { int error, newnmbufs; newnmbufs = nmbufs; error = sysctl_handle_int(oidp, &newnmbufs, 0, req); if (error == 0 && req->newptr && newnmbufs != nmbufs) { if (newnmbufs > nmbufs) { nmbufs = newnmbufs; nmbufs = uma_zone_set_max(zone_mbuf, nmbufs); EVENTHANDLER_INVOKE(nmbufs_change); } else error = EINVAL; } return (error); } SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbufs, CTLTYPE_INT|CTLFLAG_RW, &nmbufs, 0, sysctl_nmbufs, "IU", "Maximum number of mbufs allowed"); /* * Zones from which we allocate. */ uma_zone_t zone_mbuf; uma_zone_t zone_clust; uma_zone_t zone_pack; uma_zone_t zone_jumbop; uma_zone_t zone_jumbo9; uma_zone_t zone_jumbo16; -uma_zone_t zone_ext_refcnt; /* * Local prototypes. */ static int mb_ctor_mbuf(void *, int, void *, int); static int mb_ctor_clust(void *, int, void *, int); static int mb_ctor_pack(void *, int, void *, int); static void mb_dtor_mbuf(void *, int, void *); -static void mb_dtor_clust(void *, int, void *); static void mb_dtor_pack(void *, int, void *); static int mb_zinit_pack(void *, int, int); static void mb_zfini_pack(void *, int); static void mb_reclaim(uma_zone_t, int); static void *mbuf_jumbo_alloc(uma_zone_t, vm_size_t, uint8_t *, int); /* Ensure that MSIZE is a power of 2. */ CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE); /* * Initialize FreeBSD Network buffer allocation. */ static void mbuf_init(void *dummy) { /* * Configure UMA zones for Mbufs, Clusters, and Packets. */ zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE, mb_ctor_mbuf, mb_dtor_mbuf, #ifdef INVARIANTS trash_init, trash_fini, #else NULL, NULL, #endif MSIZE - 1, UMA_ZONE_MAXBUCKET); if (nmbufs > 0) nmbufs = uma_zone_set_max(zone_mbuf, nmbufs); uma_zone_set_warning(zone_mbuf, "kern.ipc.nmbufs limit reached"); uma_zone_set_maxaction(zone_mbuf, mb_reclaim); zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES, - mb_ctor_clust, mb_dtor_clust, + mb_ctor_clust, #ifdef INVARIANTS - trash_init, trash_fini, + trash_dtor, trash_init, trash_fini, #else - NULL, NULL, + NULL, NULL, NULL, #endif - UMA_ALIGN_PTR, UMA_ZONE_REFCNT); + UMA_ALIGN_PTR, 0); if (nmbclusters > 0) nmbclusters = uma_zone_set_max(zone_clust, nmbclusters); uma_zone_set_warning(zone_clust, "kern.ipc.nmbclusters limit reached"); uma_zone_set_maxaction(zone_clust, mb_reclaim); zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack, mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf); /* Make jumbo frame zone too. Page size, 9k and 16k. */ zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE, - mb_ctor_clust, mb_dtor_clust, + mb_ctor_clust, #ifdef INVARIANTS - trash_init, trash_fini, + trash_dtor, trash_init, trash_fini, #else - NULL, NULL, + NULL, NULL, NULL, #endif - UMA_ALIGN_PTR, UMA_ZONE_REFCNT); + UMA_ALIGN_PTR, 0); if (nmbjumbop > 0) nmbjumbop = uma_zone_set_max(zone_jumbop, nmbjumbop); uma_zone_set_warning(zone_jumbop, "kern.ipc.nmbjumbop limit reached"); uma_zone_set_maxaction(zone_jumbop, mb_reclaim); zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES, - mb_ctor_clust, mb_dtor_clust, + mb_ctor_clust, #ifdef INVARIANTS - trash_init, trash_fini, + trash_dtor, trash_init, trash_fini, #else - NULL, NULL, + NULL, NULL, NULL, #endif - UMA_ALIGN_PTR, UMA_ZONE_REFCNT); + UMA_ALIGN_PTR, 0); uma_zone_set_allocf(zone_jumbo9, mbuf_jumbo_alloc); if (nmbjumbo9 > 0) nmbjumbo9 = uma_zone_set_max(zone_jumbo9, nmbjumbo9); uma_zone_set_warning(zone_jumbo9, "kern.ipc.nmbjumbo9 limit reached"); uma_zone_set_maxaction(zone_jumbo9, mb_reclaim); zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES, - mb_ctor_clust, mb_dtor_clust, + mb_ctor_clust, #ifdef INVARIANTS - trash_init, trash_fini, + trash_dtor, trash_init, trash_fini, #else - NULL, NULL, + NULL, NULL, NULL, #endif - UMA_ALIGN_PTR, UMA_ZONE_REFCNT); + UMA_ALIGN_PTR, 0); uma_zone_set_allocf(zone_jumbo16, mbuf_jumbo_alloc); if (nmbjumbo16 > 0) nmbjumbo16 = uma_zone_set_max(zone_jumbo16, nmbjumbo16); uma_zone_set_warning(zone_jumbo16, "kern.ipc.nmbjumbo16 limit reached"); uma_zone_set_maxaction(zone_jumbo16, mb_reclaim); - zone_ext_refcnt = uma_zcreate(MBUF_EXTREFCNT_MEM_NAME, sizeof(u_int), - NULL, NULL, - NULL, NULL, - UMA_ALIGN_PTR, UMA_ZONE_ZINIT); - /* * Hook event handler for low-memory situation, used to * drain protocols and push data back to the caches (UMA * later pushes it back to VM). */ EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL, EVENTHANDLER_PRI_FIRST); } SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL); /* * UMA backend page allocator for the jumbo frame zones. * * Allocates kernel virtual memory that is backed by contiguous physical * pages. */ static void * mbuf_jumbo_alloc(uma_zone_t zone, vm_size_t bytes, uint8_t *flags, int wait) { /* Inform UMA that this allocator uses kernel_map/object. */ *flags = UMA_SLAB_KERNEL; return ((void *)kmem_alloc_contig(kernel_arena, bytes, wait, (vm_paddr_t)0, ~(vm_paddr_t)0, 1, 0, VM_MEMATTR_DEFAULT)); } /* * Constructor for Mbuf master zone. * * The 'arg' pointer points to a mb_args structure which * contains call-specific information required to support the * mbuf allocation API. See mbuf.h. */ static int mb_ctor_mbuf(void *mem, int size, void *arg, int how) { struct mbuf *m; struct mb_args *args; int error; int flags; short type; #ifdef INVARIANTS trash_ctor(mem, size, arg, how); #endif args = (struct mb_args *)arg; type = args->type; /* * The mbuf is initialized later. The caller has the * responsibility to set up any MAC labels too. */ if (type == MT_NOINIT) return (0); m = (struct mbuf *)mem; flags = args->flags; error = m_init(m, how, type, flags); return (error); } /* * The Mbuf master zone destructor. */ static void mb_dtor_mbuf(void *mem, int size, void *arg) { struct mbuf *m; unsigned long flags; m = (struct mbuf *)mem; flags = (unsigned long)arg; KASSERT((m->m_flags & M_NOFREE) == 0, ("%s: M_NOFREE set", __func__)); if ((m->m_flags & M_PKTHDR) && !SLIST_EMPTY(&m->m_pkthdr.tags)) m_tag_delete_chain(m, NULL); #ifdef INVARIANTS trash_dtor(mem, size, arg); #endif } /* * The Mbuf Packet zone destructor. */ static void mb_dtor_pack(void *mem, int size, void *arg) { struct mbuf *m; m = (struct mbuf *)mem; if ((m->m_flags & M_PKTHDR) != 0) m_tag_delete_chain(m, NULL); /* Make sure we've got a clean cluster back. */ KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__)); KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__)); KASSERT(m->m_ext.ext_arg1 == NULL, ("%s: ext_arg1 != NULL", __func__)); KASSERT(m->m_ext.ext_arg2 == NULL, ("%s: ext_arg2 != NULL", __func__)); KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__)); KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__)); - KASSERT(*m->m_ext.ext_cnt == 1, ("%s: ext_cnt != 1", __func__)); #ifdef INVARIANTS trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg); #endif /* * If there are processes blocked on zone_clust, waiting for pages * to be freed up, * cause them to be woken up by draining the * packet zone. We are exposed to a race here * (in the check for * the UMA_ZFLAG_FULL) where we might miss the flag set, but that * is deliberate. We don't want to acquire the zone lock for every * mbuf free. */ if (uma_zone_exhausted_nolock(zone_clust)) zone_drain(zone_pack); } /* * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor. * * Here the 'arg' pointer points to the Mbuf which we * are configuring cluster storage for. If 'arg' is * empty we allocate just the cluster without setting * the mbuf to it. See mbuf.h. */ static int mb_ctor_clust(void *mem, int size, void *arg, int how) { struct mbuf *m; - u_int *refcnt; - int type; - uma_zone_t zone; #ifdef INVARIANTS trash_ctor(mem, size, arg, how); #endif - switch (size) { - case MCLBYTES: - type = EXT_CLUSTER; - zone = zone_clust; - break; -#if MJUMPAGESIZE != MCLBYTES - case MJUMPAGESIZE: - type = EXT_JUMBOP; - zone = zone_jumbop; - break; -#endif - case MJUM9BYTES: - type = EXT_JUMBO9; - zone = zone_jumbo9; - break; - case MJUM16BYTES: - type = EXT_JUMBO16; - zone = zone_jumbo16; - break; - default: - panic("unknown cluster size"); - break; - } - m = (struct mbuf *)arg; - refcnt = uma_find_refcnt(zone, mem); - *refcnt = 1; if (m != NULL) { m->m_ext.ext_buf = (caddr_t)mem; m->m_data = m->m_ext.ext_buf; m->m_flags |= M_EXT; m->m_ext.ext_free = NULL; m->m_ext.ext_arg1 = NULL; m->m_ext.ext_arg2 = NULL; m->m_ext.ext_size = size; - m->m_ext.ext_type = type; - m->m_ext.ext_flags = 0; - m->m_ext.ext_cnt = refcnt; + m->m_ext.ext_type = m_gettype(size); + m->m_ext.ext_flags = EXT_FLAG_EMBREF; + m->m_ext.ext_count = 1; } return (0); } /* - * The Mbuf Cluster zone destructor. - */ -static void -mb_dtor_clust(void *mem, int size, void *arg) -{ -#ifdef INVARIANTS - uma_zone_t zone; - - zone = m_getzone(size); - KASSERT(*(uma_find_refcnt(zone, mem)) <= 1, - ("%s: refcnt incorrect %u", __func__, - *(uma_find_refcnt(zone, mem))) ); - - trash_dtor(mem, size, arg); -#endif -} - -/* * The Packet secondary zone's init routine, executed on the * object's transition from mbuf keg slab to zone cache. */ static int mb_zinit_pack(void *mem, int size, int how) { struct mbuf *m; m = (struct mbuf *)mem; /* m is virgin. */ if (uma_zalloc_arg(zone_clust, m, how) == NULL || m->m_ext.ext_buf == NULL) return (ENOMEM); m->m_ext.ext_type = EXT_PACKET; /* Override. */ #ifdef INVARIANTS trash_init(m->m_ext.ext_buf, MCLBYTES, how); #endif return (0); } /* * The Packet secondary zone's fini routine, executed on the * object's transition from zone cache to keg slab. */ static void mb_zfini_pack(void *mem, int size) { struct mbuf *m; m = (struct mbuf *)mem; #ifdef INVARIANTS trash_fini(m->m_ext.ext_buf, MCLBYTES); #endif uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL); #ifdef INVARIANTS trash_dtor(mem, size, NULL); #endif } /* * The "packet" keg constructor. */ static int mb_ctor_pack(void *mem, int size, void *arg, int how) { struct mbuf *m; struct mb_args *args; int error, flags; short type; m = (struct mbuf *)mem; args = (struct mb_args *)arg; flags = args->flags; type = args->type; #ifdef INVARIANTS trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how); #endif error = m_init(m, how, type, flags); /* m_ext is already initialized. */ m->m_data = m->m_ext.ext_buf; m->m_flags = (flags | M_EXT); return (error); } /* * This is the protocol drain routine. Called by UMA whenever any of the * mbuf zones is closed to its limit. * * No locks should be held when this is called. The drain routines have to * presently acquire some locks which raises the possibility of lock order * reversal. */ static void mb_reclaim(uma_zone_t zone __unused, int pending __unused) { struct domain *dp; struct protosw *pr; WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL, __func__); for (dp = domains; dp != NULL; dp = dp->dom_next) for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) if (pr->pr_drain != NULL) (*pr->pr_drain)(); } /* * Clean up after mbufs with M_EXT storage attached to them if the * reference count hits 1. */ void mb_free_ext(struct mbuf *m) { + volatile u_int *refcnt; + struct mbuf *mref; int freembuf; KASSERT(m->m_flags & M_EXT, ("%s: M_EXT not set on %p", __func__, m)); + /* See if this is the mbuf that holds the embedded refcount. */ + if (m->m_ext.ext_flags & EXT_FLAG_EMBREF) { + refcnt = &m->m_ext.ext_count; + mref = m; + } else { + KASSERT(m->m_ext.ext_cnt != NULL, + ("%s: no refcounting pointer on %p", __func__, m)); + refcnt = m->m_ext.ext_cnt; + mref = __containerof(refcnt, struct mbuf, m_ext.ext_count); + } + /* - * Check if the header is embedded in the cluster. + * Check if the header is embedded in the cluster. It is + * important that we can't touch any of the mbuf fields + * after we have freed the external storage, since mbuf + * could have been embedded in it. */ freembuf = (m->m_flags & M_NOFREE) ? 0 : 1; - switch (m->m_ext.ext_type) { - case EXT_SFBUF: - sf_ext_free(m->m_ext.ext_arg1, m->m_ext.ext_arg2); - break; - case EXT_SFBUF_NOCACHE: - sf_ext_free_nocache(m->m_ext.ext_arg1, m->m_ext.ext_arg2); - break; - default: - KASSERT(m->m_ext.ext_cnt != NULL, - ("%s: no refcounting pointer on %p", __func__, m)); - /* - * Free attached storage if this mbuf is the only - * reference to it. - */ - if (*(m->m_ext.ext_cnt) != 1) { - if (atomic_fetchadd_int(m->m_ext.ext_cnt, -1) != 1) - break; - } - + /* Free attached storage if this mbuf is the only reference to it. */ + if (*refcnt == 1 || atomic_fetchadd_int(refcnt, -1) == 1) { switch (m->m_ext.ext_type) { - case EXT_PACKET: /* The packet zone is special. */ - if (*(m->m_ext.ext_cnt) == 0) - *(m->m_ext.ext_cnt) = 1; - uma_zfree(zone_pack, m); - return; /* Job done. */ + case EXT_PACKET: + /* The packet zone is special. */ + if (*refcnt == 0) + *refcnt = 1; + uma_zfree(zone_pack, mref); + break; case EXT_CLUSTER: uma_zfree(zone_clust, m->m_ext.ext_buf); + uma_zfree(zone_mbuf, mref); break; case EXT_JUMBOP: uma_zfree(zone_jumbop, m->m_ext.ext_buf); + uma_zfree(zone_mbuf, mref); break; case EXT_JUMBO9: uma_zfree(zone_jumbo9, m->m_ext.ext_buf); + uma_zfree(zone_mbuf, mref); break; case EXT_JUMBO16: uma_zfree(zone_jumbo16, m->m_ext.ext_buf); + uma_zfree(zone_mbuf, mref); break; + case EXT_SFBUF: + sf_ext_free(m->m_ext.ext_arg1, m->m_ext.ext_arg2); + uma_zfree(zone_mbuf, mref); + break; + case EXT_SFBUF_NOCACHE: + sf_ext_free_nocache(m->m_ext.ext_arg1, + m->m_ext.ext_arg2); + uma_zfree(zone_mbuf, mref); + break; case EXT_NET_DRV: case EXT_MOD_TYPE: case EXT_DISPOSABLE: - *(m->m_ext.ext_cnt) = 0; - uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *, - m->m_ext.ext_cnt)); + uma_zfree(zone_mbuf, mref); /* FALLTHROUGH */ case EXT_EXTREF: KASSERT(m->m_ext.ext_free != NULL, ("%s: ext_free not set", __func__)); (*(m->m_ext.ext_free))(m, m->m_ext.ext_arg1, m->m_ext.ext_arg2); break; default: KASSERT(m->m_ext.ext_type == 0, ("%s: unknown ext_type", __func__)); } } - if (freembuf) + if (freembuf && m != mref) uma_zfree(zone_mbuf, m); } /* * Official mbuf(9) allocation KPI for stack and drivers: * * m_get() - a single mbuf without any attachments, sys/mbuf.h. * m_gethdr() - a single mbuf initialized as M_PKTHDR, sys/mbuf.h. * m_getcl() - an mbuf + 2k cluster, sys/mbuf.h. * m_clget() - attach cluster to already allocated mbuf. * m_cljget() - attach jumbo cluster to already allocated mbuf. * m_get2() - allocate minimum mbuf that would fit size argument. * m_getm2() - allocate a chain of mbufs/clusters. * m_extadd() - attach external cluster to mbuf. * * m_free() - free single mbuf with its tags and ext, sys/mbuf.h. * m_freem() - free chain of mbufs. */ int m_clget(struct mbuf *m, int how) { KASSERT((m->m_flags & M_EXT) == 0, ("%s: mbuf %p has M_EXT", __func__, m)); m->m_ext.ext_buf = (char *)NULL; uma_zalloc_arg(zone_clust, m, how); /* * On a cluster allocation failure, drain the packet zone and retry, * we might be able to loosen a few clusters up on the drain. */ if ((how & M_NOWAIT) && (m->m_ext.ext_buf == NULL)) { zone_drain(zone_pack); uma_zalloc_arg(zone_clust, m, how); } return (m->m_flags & M_EXT); } /* * m_cljget() is different from m_clget() as it can allocate clusters without * attaching them to an mbuf. In that case the return value is the pointer * to the cluster of the requested size. If an mbuf was specified, it gets * the cluster attached to it and the return value can be safely ignored. * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES. */ void * m_cljget(struct mbuf *m, int how, int size) { uma_zone_t zone; if (m != NULL) { KASSERT((m->m_flags & M_EXT) == 0, ("%s: mbuf %p has M_EXT", __func__, m)); m->m_ext.ext_buf = NULL; } zone = m_getzone(size); return (uma_zalloc_arg(zone, m, how)); } /* * m_get2() allocates minimum mbuf that would fit "size" argument. */ struct mbuf * m_get2(int size, int how, short type, int flags) { struct mb_args args; struct mbuf *m, *n; args.flags = flags; args.type = type; if (size <= MHLEN || (size <= MLEN && (flags & M_PKTHDR) == 0)) return (uma_zalloc_arg(zone_mbuf, &args, how)); if (size <= MCLBYTES) return (uma_zalloc_arg(zone_pack, &args, how)); if (size > MJUMPAGESIZE) return (NULL); m = uma_zalloc_arg(zone_mbuf, &args, how); if (m == NULL) return (NULL); n = uma_zalloc_arg(zone_jumbop, m, how); if (n == NULL) { uma_zfree(zone_mbuf, m); return (NULL); } return (m); } /* * m_getjcl() returns an mbuf with a cluster of the specified size attached. * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES. */ struct mbuf * m_getjcl(int how, short type, int flags, int size) { struct mb_args args; struct mbuf *m, *n; uma_zone_t zone; if (size == MCLBYTES) return m_getcl(how, type, flags); args.flags = flags; args.type = type; m = uma_zalloc_arg(zone_mbuf, &args, how); if (m == NULL) return (NULL); zone = m_getzone(size); n = uma_zalloc_arg(zone, m, how); if (n == NULL) { uma_zfree(zone_mbuf, m); return (NULL); } return (m); } /* * Allocate a given length worth of mbufs and/or clusters (whatever fits * best) and return a pointer to the top of the allocated chain. If an * existing mbuf chain is provided, then we will append the new chain * to the existing one but still return the top of the newly allocated * chain. */ struct mbuf * m_getm2(struct mbuf *m, int len, int how, short type, int flags) { struct mbuf *mb, *nm = NULL, *mtail = NULL; KASSERT(len >= 0, ("%s: len is < 0", __func__)); /* Validate flags. */ flags &= (M_PKTHDR | M_EOR); /* Packet header mbuf must be first in chain. */ if ((flags & M_PKTHDR) && m != NULL) flags &= ~M_PKTHDR; /* Loop and append maximum sized mbufs to the chain tail. */ while (len > 0) { if (len > MCLBYTES) mb = m_getjcl(how, type, (flags & M_PKTHDR), MJUMPAGESIZE); else if (len >= MINCLSIZE) mb = m_getcl(how, type, (flags & M_PKTHDR)); else if (flags & M_PKTHDR) mb = m_gethdr(how, type); else mb = m_get(how, type); /* Fail the whole operation if one mbuf can't be allocated. */ if (mb == NULL) { if (nm != NULL) m_freem(nm); return (NULL); } /* Book keeping. */ len -= M_SIZE(mb); if (mtail != NULL) mtail->m_next = mb; else nm = mb; mtail = mb; flags &= ~M_PKTHDR; /* Only valid on the first mbuf. */ } if (flags & M_EOR) mtail->m_flags |= M_EOR; /* Only valid on the last mbuf. */ /* If mbuf was supplied, append new chain to the end of it. */ if (m != NULL) { for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next) ; mtail->m_next = nm; mtail->m_flags &= ~M_EOR; } else m = nm; return (m); } /*- * Configure a provided mbuf to refer to the provided external storage - * buffer and setup a reference count for said buffer. If the setting - * up of the reference count fails, the M_EXT bit will not be set. If - * successfull, the M_EXT bit is set in the mbuf's flags. + * buffer and setup a reference count for said buffer. * * Arguments: * mb The existing mbuf to which to attach the provided buffer. * buf The address of the provided external storage buffer. * size The size of the provided buffer. * freef A pointer to a routine that is responsible for freeing the * provided external storage buffer. * args A pointer to an argument structure (of any type) to be passed * to the provided freef routine (may be NULL). * flags Any other flags to be passed to the provided mbuf. * type The type that the external storage buffer should be * labeled with. * * Returns: * Nothing. */ -int +void m_extadd(struct mbuf *mb, caddr_t buf, u_int size, void (*freef)(struct mbuf *, void *, void *), void *arg1, void *arg2, - int flags, int type, int wait) + int flags, int type) { + KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__)); - if (type != EXT_EXTREF) - mb->m_ext.ext_cnt = uma_zalloc(zone_ext_refcnt, wait); - - if (mb->m_ext.ext_cnt == NULL) - return (ENOMEM); - - *(mb->m_ext.ext_cnt) = 1; mb->m_flags |= (M_EXT | flags); mb->m_ext.ext_buf = buf; mb->m_data = mb->m_ext.ext_buf; mb->m_ext.ext_size = size; mb->m_ext.ext_free = freef; mb->m_ext.ext_arg1 = arg1; mb->m_ext.ext_arg2 = arg2; mb->m_ext.ext_type = type; - mb->m_ext.ext_flags = 0; - return (0); + if (type != EXT_EXTREF) { + mb->m_ext.ext_count = 1; + mb->m_ext.ext_flags = EXT_FLAG_EMBREF; + } else + mb->m_ext.ext_flags = 0; } /* * Free an entire chain of mbufs and associated external buffers, if * applicable. */ void m_freem(struct mbuf *mb) { while (mb != NULL) mb = m_free(mb); } Index: head/sys/kern/kern_sendfile.c =================================================================== --- head/sys/kern/kern_sendfile.c (revision 296241) +++ head/sys/kern/kern_sendfile.c (revision 296242) @@ -1,1038 +1,1014 @@ /*- * Copyright (c) 2013-2015 Gleb Smirnoff * Copyright (c) 1998, David Greenman. 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. * 4. 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. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #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 /* * Structure describing a single sendfile(2) I/O, which may consist of * several underlying pager I/Os. * * The syscall context allocates the structure and initializes 'nios' * to 1. As sendfile_swapin() runs through pages and starts asynchronous * paging operations, it increments 'nios'. * * Every I/O completion calls sendfile_iodone(), which decrements the 'nios', * and the syscall also calls sendfile_iodone() after allocating all mbufs, * linking them and sending to socket. Whoever reaches zero 'nios' is * responsible to * call pru_ready on the socket, to notify it of readyness * of the data. */ struct sf_io { volatile u_int nios; u_int error; int npages; struct file *sock_fp; struct mbuf *m; vm_page_t pa[]; }; /* * Structure used to track requests with SF_SYNC flag. */ struct sendfile_sync { struct mtx mtx; struct cv cv; unsigned count; }; counter_u64_t sfstat[sizeof(struct sfstat) / sizeof(uint64_t)]; static void sfstat_init(const void *unused) { COUNTER_ARRAY_ALLOC(sfstat, sizeof(struct sfstat) / sizeof(uint64_t), M_WAITOK); } SYSINIT(sfstat, SI_SUB_MBUF, SI_ORDER_FIRST, sfstat_init, NULL); static int sfstat_sysctl(SYSCTL_HANDLER_ARGS) { struct sfstat s; COUNTER_ARRAY_COPY(sfstat, &s, sizeof(s) / sizeof(uint64_t)); if (req->newptr) COUNTER_ARRAY_ZERO(sfstat, sizeof(s) / sizeof(uint64_t)); return (SYSCTL_OUT(req, &s, sizeof(s))); } SYSCTL_PROC(_kern_ipc, OID_AUTO, sfstat, CTLTYPE_OPAQUE | CTLFLAG_RW, NULL, 0, sfstat_sysctl, "I", "sendfile statistics"); /* - * Add more references to a vm_page + sf_buf + sendfile_sync. Called - * by mbuf(9) code to add extra references to a page. - */ -void -sf_ext_ref(void *arg1, void *arg2) -{ - struct sf_buf *sf = arg1; - struct sendfile_sync *sfs = arg2; - vm_page_t pg = sf_buf_page(sf); - - sf_buf_ref(sf); - - vm_page_lock(pg); - vm_page_wire(pg); - vm_page_unlock(pg); - - if (sfs != NULL) { - mtx_lock(&sfs->mtx); - KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0")); - sfs->count++; - mtx_unlock(&sfs->mtx); - } -} - -/* * Detach mapped page and release resources back to the system. Called * by mbuf(9) code when last reference to a page is freed. */ void sf_ext_free(void *arg1, void *arg2) { struct sf_buf *sf = arg1; struct sendfile_sync *sfs = arg2; vm_page_t pg = sf_buf_page(sf); sf_buf_free(sf); vm_page_lock(pg); /* * Check for the object going away on us. This can * happen since we don't hold a reference to it. * If so, we're responsible for freeing the page. */ if (vm_page_unwire(pg, PQ_INACTIVE) && pg->object == NULL) vm_page_free(pg); vm_page_unlock(pg); if (sfs != NULL) { mtx_lock(&sfs->mtx); KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0")); if (--sfs->count == 0) cv_signal(&sfs->cv); mtx_unlock(&sfs->mtx); } } /* * Same as above, but forces the page to be detached from the object * and go into free pool. */ void sf_ext_free_nocache(void *arg1, void *arg2) { struct sf_buf *sf = arg1; struct sendfile_sync *sfs = arg2; vm_page_t pg = sf_buf_page(sf); sf_buf_free(sf); vm_page_lock(pg); if (vm_page_unwire(pg, PQ_NONE)) { vm_object_t obj; /* Try to free the page, but only if it is cheap to. */ if ((obj = pg->object) == NULL) vm_page_free(pg); else if (!vm_page_xbusied(pg) && VM_OBJECT_TRYWLOCK(obj)) { vm_page_free(pg); VM_OBJECT_WUNLOCK(obj); } else vm_page_deactivate(pg); } vm_page_unlock(pg); if (sfs != NULL) { mtx_lock(&sfs->mtx); KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0")); if (--sfs->count == 0) cv_signal(&sfs->cv); mtx_unlock(&sfs->mtx); } } /* * Helper function to calculate how much data to put into page i of n. * Only first and last pages are special. */ static inline off_t xfsize(int i, int n, off_t off, off_t len) { if (i == 0) return (omin(PAGE_SIZE - (off & PAGE_MASK), len)); if (i == n - 1 && ((off + len) & PAGE_MASK) > 0) return ((off + len) & PAGE_MASK); return (PAGE_SIZE); } /* * Helper function to get offset within object for i page. */ static inline vm_offset_t vmoff(int i, off_t off) { if (i == 0) return ((vm_offset_t)off); return (trunc_page(off + i * PAGE_SIZE)); } /* * Helper function used when allocation of a page or sf_buf failed. * Pretend as if we don't have enough space, subtract xfsize() of * all pages that failed. */ static inline void fixspace(int old, int new, off_t off, int *space) { KASSERT(old > new, ("%s: old %d new %d", __func__, old, new)); /* Subtract last one. */ *space -= xfsize(old - 1, old, off, *space); old--; if (new == old) /* There was only one page. */ return; /* Subtract first one. */ if (new == 0) { *space -= xfsize(0, old, off, *space); new++; } /* Rest of pages are full sized. */ *space -= (old - new) * PAGE_SIZE; KASSERT(*space >= 0, ("%s: space went backwards", __func__)); } /* * I/O completion callback. */ static void sendfile_iodone(void *arg, vm_page_t *pg, int count, int error) { struct sf_io *sfio = arg; struct socket *so; for (int i = 0; i < count; i++) vm_page_xunbusy(pg[i]); if (error) sfio->error = error; if (!refcount_release(&sfio->nios)) return; so = sfio->sock_fp->f_data; if (sfio->error) { struct mbuf *m; /* * I/O operation failed. The state of data in the socket * is now inconsistent, and all what we can do is to tear * it down. Protocol abort method would tear down protocol * state, free all ready mbufs and detach not ready ones. * We will free the mbufs corresponding to this I/O manually. * * The socket would be marked with EIO and made available * for read, so that application receives EIO on next * syscall and eventually closes the socket. */ so->so_proto->pr_usrreqs->pru_abort(so); so->so_error = EIO; m = sfio->m; for (int i = 0; i < sfio->npages; i++) m = m_free(m); } else { CURVNET_SET(so->so_vnet); (void )(so->so_proto->pr_usrreqs->pru_ready)(so, sfio->m, sfio->npages); CURVNET_RESTORE(); } /* XXXGL: curthread */ fdrop(sfio->sock_fp, curthread); free(sfio, M_TEMP); } /* * Iterate through pages vector and request paging for non-valid pages. */ static int sendfile_swapin(vm_object_t obj, struct sf_io *sfio, off_t off, off_t len, int npages, int rhpages, int flags) { vm_page_t *pa = sfio->pa; int nios; nios = 0; flags = (flags & SF_NODISKIO) ? VM_ALLOC_NOWAIT : 0; /* * First grab all the pages and wire them. Note that we grab * only required pages. Readahead pages are dealt with later. */ VM_OBJECT_WLOCK(obj); for (int i = 0; i < npages; i++) { pa[i] = vm_page_grab(obj, OFF_TO_IDX(vmoff(i, off)), VM_ALLOC_WIRED | VM_ALLOC_NORMAL | flags); if (pa[i] == NULL) { npages = i; rhpages = 0; break; } } for (int i = 0; i < npages;) { int j, a, count, rv; /* Skip valid pages. */ if (vm_page_is_valid(pa[i], vmoff(i, off) & PAGE_MASK, xfsize(i, npages, off, len))) { vm_page_xunbusy(pa[i]); SFSTAT_INC(sf_pages_valid); i++; continue; } /* * Now 'i' points to first invalid page, iterate further * to make 'j' point at first valid after a bunch of * invalid ones. */ for (j = i + 1; j < npages; j++) if (vm_page_is_valid(pa[j], vmoff(j, off) & PAGE_MASK, xfsize(j, npages, off, len))) { SFSTAT_INC(sf_pages_valid); break; } /* * Now we got region of invalid pages between 'i' and 'j'. * Check that they belong to pager. They may not be there, * which is a regular situation for shmem pager. For vnode * pager this happens only in case of sparse file. * * Important feature of vm_pager_has_page() is the hint * stored in 'a', about how many pages we can pagein after * this page in a single I/O. */ while (!vm_pager_has_page(obj, OFF_TO_IDX(vmoff(i, off)), NULL, &a) && i < j) { pmap_zero_page(pa[i]); pa[i]->valid = VM_PAGE_BITS_ALL; pa[i]->dirty = 0; vm_page_xunbusy(pa[i]); i++; } if (i == j) continue; /* * We want to pagein as many pages as possible, limited only * by the 'a' hint and actual request. * * We should not pagein into already valid page, thus if * 'j' didn't reach last page, trim by that page. * * When the pagein fulfils the request, also specify readahead. */ if (j < npages) a = min(a, j - i - 1); count = min(a + 1, npages - i); refcount_acquire(&sfio->nios); rv = vm_pager_get_pages_async(obj, pa + i, count, NULL, i + count == npages ? &rhpages : NULL, &sendfile_iodone, sfio); KASSERT(rv == VM_PAGER_OK, ("%s: pager fail obj %p page %p", __func__, obj, pa[i])); SFSTAT_INC(sf_iocnt); SFSTAT_ADD(sf_pages_read, count); if (i + count == npages) SFSTAT_ADD(sf_rhpages_read, rhpages); #ifdef INVARIANTS for (j = i; j < i + count && j < npages; j++) KASSERT(pa[j] == vm_page_lookup(obj, OFF_TO_IDX(vmoff(j, off))), ("pa[j] %p lookup %p\n", pa[j], vm_page_lookup(obj, OFF_TO_IDX(vmoff(j, off))))); #endif i += count; nios++; } VM_OBJECT_WUNLOCK(obj); if (nios == 0 && npages != 0) SFSTAT_INC(sf_noiocnt); return (nios); } static int sendfile_getobj(struct thread *td, struct file *fp, vm_object_t *obj_res, struct vnode **vp_res, struct shmfd **shmfd_res, off_t *obj_size, int *bsize) { struct vattr va; vm_object_t obj; struct vnode *vp; struct shmfd *shmfd; int error; vp = *vp_res = NULL; obj = NULL; shmfd = *shmfd_res = NULL; *bsize = 0; /* * The file descriptor must be a regular file and have a * backing VM object. */ if (fp->f_type == DTYPE_VNODE) { vp = fp->f_vnode; vn_lock(vp, LK_SHARED | LK_RETRY); if (vp->v_type != VREG) { error = EINVAL; goto out; } *bsize = vp->v_mount->mnt_stat.f_iosize; error = VOP_GETATTR(vp, &va, td->td_ucred); if (error != 0) goto out; *obj_size = va.va_size; obj = vp->v_object; if (obj == NULL) { error = EINVAL; goto out; } } else if (fp->f_type == DTYPE_SHM) { error = 0; shmfd = fp->f_data; obj = shmfd->shm_object; *obj_size = shmfd->shm_size; } else { error = EINVAL; goto out; } VM_OBJECT_WLOCK(obj); if ((obj->flags & OBJ_DEAD) != 0) { VM_OBJECT_WUNLOCK(obj); error = EBADF; goto out; } /* * Temporarily increase the backing VM object's reference * count so that a forced reclamation of its vnode does not * immediately destroy it. */ vm_object_reference_locked(obj); VM_OBJECT_WUNLOCK(obj); *obj_res = obj; *vp_res = vp; *shmfd_res = shmfd; out: if (vp != NULL) VOP_UNLOCK(vp, 0); return (error); } static int sendfile_getsock(struct thread *td, int s, struct file **sock_fp, struct socket **so) { cap_rights_t rights; int error; *sock_fp = NULL; *so = NULL; /* * The socket must be a stream socket and connected. */ error = getsock_cap(td, s, cap_rights_init(&rights, CAP_SEND), sock_fp, NULL); if (error != 0) return (error); *so = (*sock_fp)->f_data; if ((*so)->so_type != SOCK_STREAM) return (EINVAL); if (((*so)->so_state & SS_ISCONNECTED) == 0) return (ENOTCONN); return (0); } int vn_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, int kflags, struct thread *td) { struct file *sock_fp; struct vnode *vp; struct vm_object *obj; struct socket *so; struct mbuf *m, *mh, *mhtail; struct sf_buf *sf; struct shmfd *shmfd; struct sendfile_sync *sfs; struct vattr va; off_t off, sbytes, rem, obj_size; int error, softerr, bsize, hdrlen; obj = NULL; so = NULL; m = mh = NULL; sfs = NULL; sbytes = 0; softerr = 0; error = sendfile_getobj(td, fp, &obj, &vp, &shmfd, &obj_size, &bsize); if (error != 0) return (error); error = sendfile_getsock(td, sockfd, &sock_fp, &so); if (error != 0) goto out; #ifdef MAC error = mac_socket_check_send(td->td_ucred, so); if (error != 0) goto out; #endif SFSTAT_INC(sf_syscalls); SFSTAT_ADD(sf_rhpages_requested, SF_READAHEAD(flags)); if (flags & SF_SYNC) { sfs = malloc(sizeof *sfs, M_TEMP, M_WAITOK | M_ZERO); mtx_init(&sfs->mtx, "sendfile", NULL, MTX_DEF); cv_init(&sfs->cv, "sendfile"); } /* If headers are specified copy them into mbufs. */ if (hdr_uio != NULL && hdr_uio->uio_resid > 0) { hdr_uio->uio_td = td; hdr_uio->uio_rw = UIO_WRITE; /* * In FBSD < 5.0 the nbytes to send also included * the header. If compat is specified subtract the * header size from nbytes. */ if (kflags & SFK_COMPAT) { if (nbytes > hdr_uio->uio_resid) nbytes -= hdr_uio->uio_resid; else nbytes = 0; } mh = m_uiotombuf(hdr_uio, M_WAITOK, 0, 0, 0); hdrlen = m_length(mh, &mhtail); } else hdrlen = 0; rem = nbytes ? omin(nbytes, obj_size - offset) : obj_size - offset; /* * Protect against multiple writers to the socket. * * XXXRW: Historically this has assumed non-interruptibility, so now * we implement that, but possibly shouldn't. */ (void)sblock(&so->so_snd, SBL_WAIT | SBL_NOINTR); /* * Loop through the pages of the file, starting with the requested * offset. Get a file page (do I/O if necessary), map the file page * into an sf_buf, attach an mbuf header to the sf_buf, and queue * it on the socket. * This is done in two loops. The inner loop turns as many pages * as it can, up to available socket buffer space, without blocking * into mbufs to have it bulk delivered into the socket send buffer. * The outer loop checks the state and available space of the socket * and takes care of the overall progress. */ for (off = offset; rem > 0; ) { struct sf_io *sfio; vm_page_t *pa; struct mbuf *mtail; int nios, space, npages, rhpages; mtail = NULL; /* * Check the socket state for ongoing connection, * no errors and space in socket buffer. * If space is low allow for the remainder of the * file to be processed if it fits the socket buffer. * Otherwise block in waiting for sufficient space * to proceed, or if the socket is nonblocking, return * to userland with EAGAIN while reporting how far * we've come. * We wait until the socket buffer has significant free * space to do bulk sends. This makes good use of file * system read ahead and allows packet segmentation * offloading hardware to take over lots of work. If * we were not careful here we would send off only one * sfbuf at a time. */ SOCKBUF_LOCK(&so->so_snd); if (so->so_snd.sb_lowat < so->so_snd.sb_hiwat / 2) so->so_snd.sb_lowat = so->so_snd.sb_hiwat / 2; retry_space: if (so->so_snd.sb_state & SBS_CANTSENDMORE) { error = EPIPE; SOCKBUF_UNLOCK(&so->so_snd); goto done; } else if (so->so_error) { error = so->so_error; so->so_error = 0; SOCKBUF_UNLOCK(&so->so_snd); goto done; } space = sbspace(&so->so_snd); if (space < rem && (space <= 0 || space < so->so_snd.sb_lowat)) { if (so->so_state & SS_NBIO) { SOCKBUF_UNLOCK(&so->so_snd); error = EAGAIN; goto done; } /* * sbwait drops the lock while sleeping. * When we loop back to retry_space the * state may have changed and we retest * for it. */ error = sbwait(&so->so_snd); /* * An error from sbwait usually indicates that we've * been interrupted by a signal. If we've sent anything * then return bytes sent, otherwise return the error. */ if (error != 0) { SOCKBUF_UNLOCK(&so->so_snd); goto done; } goto retry_space; } SOCKBUF_UNLOCK(&so->so_snd); /* * Reduce space in the socket buffer by the size of * the header mbuf chain. * hdrlen is set to 0 after the first loop. */ space -= hdrlen; if (vp != NULL) { error = vn_lock(vp, LK_SHARED); if (error != 0) goto done; error = VOP_GETATTR(vp, &va, td->td_ucred); if (error != 0 || off >= va.va_size) { VOP_UNLOCK(vp, 0); goto done; } if (va.va_size != obj_size) { if (nbytes == 0) rem += va.va_size - obj_size; else if (offset + nbytes > va.va_size) rem -= (offset + nbytes - va.va_size); obj_size = va.va_size; } } if (space > rem) space = rem; npages = howmany(space + (off & PAGE_MASK), PAGE_SIZE); /* * Calculate maximum allowed number of pages for readahead * at this iteration. First, we allow readahead up to "rem". * If application wants more, let it be, but there is no * reason to go above MAXPHYS. Also check against "obj_size", * since vm_pager_has_page() can hint beyond EOF. */ rhpages = howmany(rem + (off & PAGE_MASK), PAGE_SIZE) - npages; rhpages += SF_READAHEAD(flags); rhpages = min(howmany(MAXPHYS, PAGE_SIZE), rhpages); rhpages = min(howmany(obj_size - trunc_page(off), PAGE_SIZE) - npages, rhpages); sfio = malloc(sizeof(struct sf_io) + npages * sizeof(vm_page_t), M_TEMP, M_WAITOK); refcount_init(&sfio->nios, 1); sfio->error = 0; nios = sendfile_swapin(obj, sfio, off, space, npages, rhpages, flags); /* * Loop and construct maximum sized mbuf chain to be bulk * dumped into socket buffer. */ pa = sfio->pa; for (int i = 0; i < npages; i++) { struct mbuf *m0; /* * If a page wasn't grabbed successfully, then * trim the array. Can happen only with SF_NODISKIO. */ if (pa[i] == NULL) { SFSTAT_INC(sf_busy); fixspace(npages, i, off, &space); npages = i; softerr = EBUSY; break; } /* * Get a sendfile buf. When allocating the * first buffer for mbuf chain, we usually * wait as long as necessary, but this wait * can be interrupted. For consequent * buffers, do not sleep, since several * threads might exhaust the buffers and then * deadlock. */ sf = sf_buf_alloc(pa[i], m != NULL ? SFB_NOWAIT : SFB_CATCH); if (sf == NULL) { SFSTAT_INC(sf_allocfail); for (int j = i; j < npages; j++) { vm_page_lock(pa[j]); vm_page_unwire(pa[j], PQ_INACTIVE); vm_page_unlock(pa[j]); } if (m == NULL) softerr = ENOBUFS; fixspace(npages, i, off, &space); npages = i; break; } m0 = m_get(M_WAITOK, MT_DATA); m0->m_ext.ext_buf = (char *)sf_buf_kva(sf); m0->m_ext.ext_size = PAGE_SIZE; m0->m_ext.ext_arg1 = sf; m0->m_ext.ext_arg2 = sfs; /* * SF_NOCACHE sets the page as being freed upon send. * However, we ignore it for the last page in 'space', * if the page is truncated, and we got more data to * send (rem > space), or if we have readahead * configured (rhpages > 0). */ if ((flags & SF_NOCACHE) == 0 || (i == npages - 1 && ((off + space) & PAGE_MASK) && (rem > space || rhpages > 0))) m0->m_ext.ext_type = EXT_SFBUF; else m0->m_ext.ext_type = EXT_SFBUF_NOCACHE; - m0->m_ext.ext_flags = 0; + m0->m_ext.ext_flags = EXT_FLAG_EMBREF; + m0->m_ext.ext_count = 1; m0->m_flags |= (M_EXT | M_RDONLY); if (nios) m0->m_flags |= M_NOTREADY; m0->m_data = (char *)sf_buf_kva(sf) + (vmoff(i, off) & PAGE_MASK); m0->m_len = xfsize(i, npages, off, space); if (i == 0) sfio->m = m0; /* Append to mbuf chain. */ if (mtail != NULL) mtail->m_next = m0; else m = m0; mtail = m0; if (sfs != NULL) { mtx_lock(&sfs->mtx); sfs->count++; mtx_unlock(&sfs->mtx); } } if (vp != NULL) VOP_UNLOCK(vp, 0); /* Keep track of bytes processed. */ off += space; rem -= space; /* Prepend header, if any. */ if (hdrlen) { mhtail->m_next = m; m = mh; mh = NULL; } if (m == NULL) { KASSERT(softerr, ("%s: m NULL, no error", __func__)); error = softerr; free(sfio, M_TEMP); goto done; } /* Add the buffer chain to the socket buffer. */ KASSERT(m_length(m, NULL) == space + hdrlen, ("%s: mlen %u space %d hdrlen %d", __func__, m_length(m, NULL), space, hdrlen)); CURVNET_SET(so->so_vnet); if (nios == 0) { /* * If sendfile_swapin() didn't initiate any I/Os, * which happens if all data is cached in VM, then * we can send data right now without the * PRUS_NOTREADY flag. */ free(sfio, M_TEMP); error = (*so->so_proto->pr_usrreqs->pru_send) (so, 0, m, NULL, NULL, td); } else { sfio->sock_fp = sock_fp; sfio->npages = npages; fhold(sock_fp); error = (*so->so_proto->pr_usrreqs->pru_send) (so, PRUS_NOTREADY, m, NULL, NULL, td); sendfile_iodone(sfio, NULL, 0, 0); } CURVNET_RESTORE(); m = NULL; /* pru_send always consumes */ if (error) goto done; sbytes += space + hdrlen; if (hdrlen) hdrlen = 0; if (softerr) { error = softerr; goto done; } } /* * Send trailers. Wimp out and use writev(2). */ if (trl_uio != NULL) { sbunlock(&so->so_snd); error = kern_writev(td, sockfd, trl_uio); if (error == 0) sbytes += td->td_retval[0]; goto out; } done: sbunlock(&so->so_snd); out: /* * If there was no error we have to clear td->td_retval[0] * because it may have been set by writev. */ if (error == 0) { td->td_retval[0] = 0; } if (sent != NULL) { (*sent) = sbytes; } if (obj != NULL) vm_object_deallocate(obj); if (so) fdrop(sock_fp, td); if (m) m_freem(m); if (mh) m_freem(mh); if (sfs != NULL) { mtx_lock(&sfs->mtx); if (sfs->count != 0) cv_wait(&sfs->cv, &sfs->mtx); KASSERT(sfs->count == 0, ("sendfile sync still busy")); cv_destroy(&sfs->cv); mtx_destroy(&sfs->mtx); free(sfs, M_TEMP); } if (error == ERESTART) error = EINTR; return (error); } static int sendfile(struct thread *td, struct sendfile_args *uap, int compat) { struct sf_hdtr hdtr; struct uio *hdr_uio, *trl_uio; struct file *fp; cap_rights_t rights; off_t sbytes; int error; /* * File offset must be positive. If it goes beyond EOF * we send only the header/trailer and no payload data. */ if (uap->offset < 0) return (EINVAL); hdr_uio = trl_uio = NULL; if (uap->hdtr != NULL) { error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); if (error != 0) goto out; if (hdtr.headers != NULL) { error = copyinuio(hdtr.headers, hdtr.hdr_cnt, &hdr_uio); if (error != 0) goto out; } if (hdtr.trailers != NULL) { error = copyinuio(hdtr.trailers, hdtr.trl_cnt, &trl_uio); if (error != 0) goto out; } } AUDIT_ARG_FD(uap->fd); /* * sendfile(2) can start at any offset within a file so we require * CAP_READ+CAP_SEEK = CAP_PREAD. */ if ((error = fget_read(td, uap->fd, cap_rights_init(&rights, CAP_PREAD), &fp)) != 0) { goto out; } error = fo_sendfile(fp, uap->s, hdr_uio, trl_uio, uap->offset, uap->nbytes, &sbytes, uap->flags, compat ? SFK_COMPAT : 0, td); fdrop(fp, td); if (uap->sbytes != NULL) copyout(&sbytes, uap->sbytes, sizeof(off_t)); out: free(hdr_uio, M_IOV); free(trl_uio, M_IOV); return (error); } /* * sendfile(2) * * int sendfile(int fd, int s, off_t offset, size_t nbytes, * struct sf_hdtr *hdtr, off_t *sbytes, int flags) * * Send a file specified by 'fd' and starting at 'offset' to a socket * specified by 's'. Send only 'nbytes' of the file or until EOF if nbytes == * 0. Optionally add a header and/or trailer to the socket output. If * specified, write the total number of bytes sent into *sbytes. */ int sys_sendfile(struct thread *td, struct sendfile_args *uap) { return (sendfile(td, uap, 0)); } #ifdef COMPAT_FREEBSD4 int freebsd4_sendfile(struct thread *td, struct freebsd4_sendfile_args *uap) { struct sendfile_args args; args.fd = uap->fd; args.s = uap->s; args.offset = uap->offset; args.nbytes = uap->nbytes; args.hdtr = uap->hdtr; args.sbytes = uap->sbytes; args.flags = uap->flags; return (sendfile(td, &args, 1)); } #endif /* COMPAT_FREEBSD4 */ Index: head/sys/kern/uipc_mbuf.c =================================================================== --- head/sys/kern/uipc_mbuf.c (revision 296241) +++ head/sys/kern/uipc_mbuf.c (revision 296242) @@ -1,1824 +1,1826 @@ /*- * 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. * 4. 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 int max_linkhdr; int max_protohdr; int max_hdr; int max_datalen; #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) == 56); CTASSERT(sizeof(struct m_ext) == 48); #else CTASSERT(offsetof(struct mbuf, m_dat) == 24); CTASSERT(sizeof(struct pkthdr) == 48); CTASSERT(sizeof(struct m_ext) == 28); #endif /* * Assert that the queue(3) macros produce code of the same size as an old * plain pointer does. */ #ifdef INVARIANTS static struct mbuf 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, const struct mbuf *m) +mb_dupcl(struct mbuf *n, struct mbuf *m) { + volatile u_int *refcnt; KASSERT(m->m_flags & M_EXT, ("%s: M_EXT not set on %p", __func__, m)); KASSERT(!(n->m_flags & M_EXT), ("%s: M_EXT set on %p", __func__, n)); - switch (m->m_ext.ext_type) { - case EXT_SFBUF: - case EXT_SFBUF_NOCACHE: - sf_ext_ref(m->m_ext.ext_arg1, m->m_ext.ext_arg2); - break; - default: + n->m_ext = m->m_ext; + n->m_flags |= M_EXT; + n->m_flags |= m->m_flags & M_RDONLY; + + /* 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)); - if (*(m->m_ext.ext_cnt) == 1) - *(m->m_ext.ext_cnt) += 1; - else - atomic_add_int(m->m_ext.ext_cnt, 1); + refcnt = m->m_ext.ext_cnt; } - n->m_ext = m->m_ext; - n->m_flags |= M_EXT; - n->m_flags |= m->m_flags & M_RDONLY; + if (*refcnt == 1) + *refcnt += 1; + else + atomic_add_int(refcnt, 1); } void m_demote_pkthdr(struct mbuf *m) { M_ASSERTPKTHDR(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; 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 = m->m_flags & (M_EXT | M_RDONLY | M_NOFREE | 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 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); 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; } /* * 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); if ((to->m_flags & M_EXT) == 0) to->m_data = to->m_pktdat; to->m_pkthdr = from->m_pkthdr; 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(const struct mbuf *m, int off0, int len, int wait) +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 = 0; 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) { 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) { 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) { 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); } /* * 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); 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); bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 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) || 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; } } } /* * 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; /* * 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; 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); 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) 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) { 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 <= MLEN) { 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); } /* * 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); 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); } /* * 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 = (*f)(arg, mtod(m, caddr_t) + off, count); 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); } /* * 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 it's 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++; /* * 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)) { bcopy(mtod(n, void *), mtod(m, char *) + m->m_len, n->m_len); m->m_len += n->m_len; m->m_next = n->m_next; 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; bcopy(mtod(n, void *), mtod(m, void *), n->m_len); bcopy(mtod(n2, void *), mtod(m, char *) + n->m_len, n2->m_len); m->m_len = n->m_len + n2->m_len; m->m_next = n2->m_next; *prev = m; m_free(n); m_free(n2); if (--curfrags <= maxfrags) /* +1 cl -2 mbufs */ 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_new = NULL, *m_final = NULL; int progress = 0; if (!(m0->m_flags & M_PKTHDR)) return (m0); if ((length == 0) || (length < -2)) return (m0); m_fixhdr(m0); /* Needed sanity check */ m_final = m_getcl(how, MT_DATA, M_PKTHDR); if (m_final == NULL) goto nospace; if (m_dup_pkthdr(m_final, m0, how) == 0) goto nospace; m_new = m_final; if (length == -1) length = 1 + (arc4random() & 255); while (progress < m0->m_pkthdr.len) { int fraglen; if (length > 0) fraglen = length; else fraglen = 1 + (arc4random() & 255); if (fraglen > m0->m_pkthdr.len - progress) fraglen = m0->m_pkthdr.len - progress; if (fraglen > MCLBYTES) fraglen = MCLBYTES; if (m_new == NULL) { m_new = m_getcl(how, MT_DATA, 0); if (m_new == NULL) goto nospace; } m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t)); progress += fraglen; m_new->m_len = fraglen; if (m_new != m_final) m_cat(m_final, m_new); m_new = NULL; } m_freem(m0); m0 = m_final; return (m0); nospace: if (m_final) m_freem(m_final); /* Return the original chain on failure */ return (m0); } #endif /* * 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; /* * 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; /* * 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; } KASSERT(progress == total, ("%s: progress != total", __func__)); return (m); } /* * Copy an mbuf chain into a uio limited by len if set. */ int m_mbuftouio(struct uio *uio, 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); 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 */ #if 0 newipsecstat.ips_mbcoalesced++; #endif } 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 */ #if 0 newipsecstat.ips_clcoalesced++; #endif 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); 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); 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; #define MP_MAXDIGITS 21 /* strlen("16,000,000,000,000,000,000") == 21 */ #define MP_NUMLINES 6 #define MP_NUMSPERLINE 16 #define MP_EXTRABYTES 64 /* > strlen("used:\nwasted:\nsegments:\n") */ /* work out max space needed and add a bit of spare space too */ #define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE) #define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES) char mbprofbuf[MP_BUFSIZE]; 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 void mbprof_textify(void) { int offset; char *c; uint64_t *p; p = &mbprof.wasted[0]; c = mbprofbuf; offset = snprintf(c, MP_MAXLINE + 10, "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]; c += offset; offset = snprintf(c, MP_MAXLINE, "%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]; c += offset; offset = snprintf(c, MP_MAXLINE + 10, "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]; c += offset; offset = snprintf(c, MP_MAXLINE, "%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]; c += offset; offset = snprintf(c, MP_MAXLINE + 10, "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]; c += offset; offset = snprintf(c, MP_MAXLINE, "%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 } static int mbprof_handler(SYSCTL_HANDLER_ARGS) { int error; mbprof_textify(); error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1); 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, NULL, 0, mbprof_handler, "A", "mbuf profiling statistics"); SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr, CTLTYPE_INT|CTLFLAG_RW, NULL, 0, mbprof_clr_handler, "I", "clear mbuf profiling statistics"); #endif Index: head/sys/netinet6/ip6_output.c =================================================================== --- head/sys/netinet6/ip6_output.c (revision 296241) +++ head/sys/netinet6/ip6_output.c (revision 296242) @@ -1,3062 +1,3062 @@ /*- * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * 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 project 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 PROJECT 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 PROJECT 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. * * $KAME: ip6_output.c,v 1.279 2002/01/26 06:12:30 jinmei Exp $ */ /*- * Copyright (c) 1982, 1986, 1988, 1990, 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. * 4. 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. * * @(#)ip_output.c 8.3 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipfw.h" #include "opt_ipsec.h" #include "opt_sctp.h" #include "opt_route.h" #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef IPSEC #include #include #include #include #endif /* IPSEC */ #ifdef SCTP #include #include #endif #include #include #ifdef FLOWTABLE #include #endif extern int in6_mcast_loop; struct ip6_exthdrs { struct mbuf *ip6e_ip6; struct mbuf *ip6e_hbh; struct mbuf *ip6e_dest1; struct mbuf *ip6e_rthdr; struct mbuf *ip6e_dest2; }; static int ip6_pcbopt(int, u_char *, int, struct ip6_pktopts **, struct ucred *, int); static int ip6_pcbopts(struct ip6_pktopts **, struct mbuf *, struct socket *, struct sockopt *); static int ip6_getpcbopt(struct ip6_pktopts *, int, struct sockopt *); static int ip6_setpktopt(int, u_char *, int, struct ip6_pktopts *, struct ucred *, int, int, int); static int ip6_copyexthdr(struct mbuf **, caddr_t, int); static int ip6_insertfraghdr(struct mbuf *, struct mbuf *, int, struct ip6_frag **); static int ip6_insert_jumboopt(struct ip6_exthdrs *, u_int32_t); static int ip6_splithdr(struct mbuf *, struct ip6_exthdrs *); static int ip6_getpmtu(struct route_in6 *, int, struct ifnet *, struct in6_addr *, u_long *, int *, u_int); static int ip6_calcmtu(struct ifnet *, const struct in6_addr *, u_long, u_long *, int *); static int ip6_getpmtu_ctl(u_int, struct in6_addr *, u_long *); static int copypktopts(struct ip6_pktopts *, struct ip6_pktopts *, int); /* * Make an extension header from option data. hp is the source, and * mp is the destination. */ #define MAKE_EXTHDR(hp, mp) \ do { \ if (hp) { \ struct ip6_ext *eh = (struct ip6_ext *)(hp); \ error = ip6_copyexthdr((mp), (caddr_t)(hp), \ ((eh)->ip6e_len + 1) << 3); \ if (error) \ goto freehdrs; \ } \ } while (/*CONSTCOND*/ 0) /* * Form a chain of extension headers. * m is the extension header mbuf * mp is the previous mbuf in the chain * p is the next header * i is the type of option. */ #define MAKE_CHAIN(m, mp, p, i)\ do {\ if (m) {\ if (!hdrsplit) \ panic("assumption failed: hdr not split"); \ *mtod((m), u_char *) = *(p);\ *(p) = (i);\ p = mtod((m), u_char *);\ (m)->m_next = (mp)->m_next;\ (mp)->m_next = (m);\ (mp) = (m);\ }\ } while (/*CONSTCOND*/ 0) void in6_delayed_cksum(struct mbuf *m, uint32_t plen, u_short offset) { u_short csum; csum = in_cksum_skip(m, offset + plen, offset); if (m->m_pkthdr.csum_flags & CSUM_UDP_IPV6 && csum == 0) csum = 0xffff; offset += m->m_pkthdr.csum_data; /* checksum offset */ if (offset + sizeof(u_short) > m->m_len) { printf("%s: delayed m_pullup, m->len: %d plen %u off %u " "csum_flags=%b\n", __func__, m->m_len, plen, offset, (int)m->m_pkthdr.csum_flags, CSUM_BITS); /* * XXX this should not happen, but if it does, the correct * behavior may be to insert the checksum in the appropriate * next mbuf in the chain. */ return; } *(u_short *)(m->m_data + offset) = csum; } int ip6_fragment(struct ifnet *ifp, struct mbuf *m0, int hlen, u_char nextproto, int mtu, uint32_t id) { struct mbuf *m, **mnext, *m_frgpart; struct ip6_hdr *ip6, *mhip6; struct ip6_frag *ip6f; int off; int error; int tlen = m0->m_pkthdr.len; m = m0; ip6 = mtod(m, struct ip6_hdr *); mnext = &m->m_nextpkt; for (off = hlen; off < tlen; off += mtu) { m = m_gethdr(M_NOWAIT, MT_DATA); if (!m) { IP6STAT_INC(ip6s_odropped); return (ENOBUFS); } m->m_flags = m0->m_flags & M_COPYFLAGS; *mnext = m; mnext = &m->m_nextpkt; m->m_data += max_linkhdr; mhip6 = mtod(m, struct ip6_hdr *); *mhip6 = *ip6; m->m_len = sizeof(*mhip6); error = ip6_insertfraghdr(m0, m, hlen, &ip6f); if (error) { IP6STAT_INC(ip6s_odropped); return (error); } ip6f->ip6f_offlg = htons((u_short)((off - hlen) & ~7)); if (off + mtu >= tlen) mtu = tlen - off; else ip6f->ip6f_offlg |= IP6F_MORE_FRAG; mhip6->ip6_plen = htons((u_short)(mtu + hlen + sizeof(*ip6f) - sizeof(struct ip6_hdr))); if ((m_frgpart = m_copy(m0, off, mtu)) == 0) { IP6STAT_INC(ip6s_odropped); return (ENOBUFS); } m_cat(m, m_frgpart); m->m_pkthdr.len = mtu + hlen + sizeof(*ip6f); m->m_pkthdr.fibnum = m0->m_pkthdr.fibnum; m->m_pkthdr.rcvif = NULL; ip6f->ip6f_reserved = 0; ip6f->ip6f_ident = id; ip6f->ip6f_nxt = nextproto; IP6STAT_INC(ip6s_ofragments); in6_ifstat_inc(ifp, ifs6_out_fragcreat); } return (0); } /* * IP6 output. The packet in mbuf chain m contains a skeletal IP6 * header (with pri, len, nxt, hlim, src, dst). * This function may modify ver and hlim only. * The mbuf chain containing the packet will be freed. * The mbuf opt, if present, will not be freed. * If route_in6 ro is present and has ro_rt initialized, route lookup would be * skipped and ro->ro_rt would be used. If ro is present but ro->ro_rt is NULL, * then result of route lookup is stored in ro->ro_rt. * * type of "mtu": rt_mtu is u_long, ifnet.ifr_mtu is int, and * nd_ifinfo.linkmtu is u_int32_t. so we use u_long to hold largest one, * which is rt_mtu. * * ifpp - XXX: just for statistics */ /* * XXX TODO: no flowid is assigned for outbound flows? */ int ip6_output(struct mbuf *m0, struct ip6_pktopts *opt, struct route_in6 *ro, int flags, struct ip6_moptions *im6o, struct ifnet **ifpp, struct inpcb *inp) { struct ip6_hdr *ip6; struct ifnet *ifp, *origifp; struct mbuf *m = m0; struct mbuf *mprev = NULL; int hlen, tlen, len; struct route_in6 ip6route; struct rtentry *rt = NULL; struct sockaddr_in6 *dst, src_sa, dst_sa; struct in6_addr odst; int error = 0; struct in6_ifaddr *ia = NULL; u_long mtu; int alwaysfrag, dontfrag; u_int32_t optlen = 0, plen = 0, unfragpartlen = 0; struct ip6_exthdrs exthdrs; struct in6_addr finaldst, src0, dst0; u_int32_t zone; struct route_in6 *ro_pmtu = NULL; int hdrsplit = 0; int sw_csum, tso; int needfiblookup; uint32_t fibnum; struct m_tag *fwd_tag = NULL; uint32_t id; ip6 = mtod(m, struct ip6_hdr *); if (ip6 == NULL) { printf ("ip6 is NULL"); goto bad; } if (inp != NULL) { M_SETFIB(m, inp->inp_inc.inc_fibnum); if ((flags & IP_NODEFAULTFLOWID) == 0) { /* unconditionally set flowid */ m->m_pkthdr.flowid = inp->inp_flowid; M_HASHTYPE_SET(m, inp->inp_flowtype); } } finaldst = ip6->ip6_dst; bzero(&exthdrs, sizeof(exthdrs)); if (opt) { /* Hop-by-Hop options header */ MAKE_EXTHDR(opt->ip6po_hbh, &exthdrs.ip6e_hbh); /* Destination options header(1st part) */ if (opt->ip6po_rthdr) { /* * Destination options header(1st part) * This only makes sense with a routing header. * See Section 9.2 of RFC 3542. * Disabling this part just for MIP6 convenience is * a bad idea. We need to think carefully about a * way to make the advanced API coexist with MIP6 * options, which might automatically be inserted in * the kernel. */ MAKE_EXTHDR(opt->ip6po_dest1, &exthdrs.ip6e_dest1); } /* Routing header */ MAKE_EXTHDR(opt->ip6po_rthdr, &exthdrs.ip6e_rthdr); /* Destination options header(2nd part) */ MAKE_EXTHDR(opt->ip6po_dest2, &exthdrs.ip6e_dest2); } #ifdef IPSEC /* * IPSec checking which handles several cases. * FAST IPSEC: We re-injected the packet. * XXX: need scope argument. */ switch(ip6_ipsec_output(&m, inp, &error)) { case 1: /* Bad packet */ goto freehdrs; case -1: /* IPSec done */ goto done; case 0: /* No IPSec */ default: break; } #endif /* IPSEC */ /* * Calculate the total length of the extension header chain. * Keep the length of the unfragmentable part for fragmentation. */ optlen = 0; if (exthdrs.ip6e_hbh) optlen += exthdrs.ip6e_hbh->m_len; if (exthdrs.ip6e_dest1) optlen += exthdrs.ip6e_dest1->m_len; if (exthdrs.ip6e_rthdr) optlen += exthdrs.ip6e_rthdr->m_len; unfragpartlen = optlen + sizeof(struct ip6_hdr); /* NOTE: we don't add AH/ESP length here (done in ip6_ipsec_output) */ if (exthdrs.ip6e_dest2) optlen += exthdrs.ip6e_dest2->m_len; /* * If there is at least one extension header, * separate IP6 header from the payload. */ if (optlen && !hdrsplit) { if ((error = ip6_splithdr(m, &exthdrs)) != 0) { m = NULL; goto freehdrs; } m = exthdrs.ip6e_ip6; hdrsplit++; } /* adjust pointer */ ip6 = mtod(m, struct ip6_hdr *); /* adjust mbuf packet header length */ m->m_pkthdr.len += optlen; plen = m->m_pkthdr.len - sizeof(*ip6); /* If this is a jumbo payload, insert a jumbo payload option. */ if (plen > IPV6_MAXPACKET) { if (!hdrsplit) { if ((error = ip6_splithdr(m, &exthdrs)) != 0) { m = NULL; goto freehdrs; } m = exthdrs.ip6e_ip6; hdrsplit++; } /* adjust pointer */ ip6 = mtod(m, struct ip6_hdr *); if ((error = ip6_insert_jumboopt(&exthdrs, plen)) != 0) goto freehdrs; ip6->ip6_plen = 0; } else ip6->ip6_plen = htons(plen); /* * Concatenate headers and fill in next header fields. * Here we have, on "m" * IPv6 payload * and we insert headers accordingly. Finally, we should be getting: * IPv6 hbh dest1 rthdr ah* [esp* dest2 payload] * * during the header composing process, "m" points to IPv6 header. * "mprev" points to an extension header prior to esp. */ u_char *nexthdrp = &ip6->ip6_nxt; mprev = m; /* * we treat dest2 specially. this makes IPsec processing * much easier. the goal here is to make mprev point the * mbuf prior to dest2. * * result: IPv6 dest2 payload * m and mprev will point to IPv6 header. */ if (exthdrs.ip6e_dest2) { if (!hdrsplit) panic("assumption failed: hdr not split"); exthdrs.ip6e_dest2->m_next = m->m_next; m->m_next = exthdrs.ip6e_dest2; *mtod(exthdrs.ip6e_dest2, u_char *) = ip6->ip6_nxt; ip6->ip6_nxt = IPPROTO_DSTOPTS; } /* * result: IPv6 hbh dest1 rthdr dest2 payload * m will point to IPv6 header. mprev will point to the * extension header prior to dest2 (rthdr in the above case). */ MAKE_CHAIN(exthdrs.ip6e_hbh, mprev, nexthdrp, IPPROTO_HOPOPTS); MAKE_CHAIN(exthdrs.ip6e_dest1, mprev, nexthdrp, IPPROTO_DSTOPTS); MAKE_CHAIN(exthdrs.ip6e_rthdr, mprev, nexthdrp, IPPROTO_ROUTING); /* * If there is a routing header, discard the packet. */ if (exthdrs.ip6e_rthdr) { error = EINVAL; goto bad; } /* Source address validation */ if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src) && (flags & IPV6_UNSPECSRC) == 0) { error = EOPNOTSUPP; IP6STAT_INC(ip6s_badscope); goto bad; } if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { error = EOPNOTSUPP; IP6STAT_INC(ip6s_badscope); goto bad; } IP6STAT_INC(ip6s_localout); /* * Route packet. */ if (ro == 0) { ro = &ip6route; bzero((caddr_t)ro, sizeof(*ro)); } ro_pmtu = ro; if (opt && opt->ip6po_rthdr) ro = &opt->ip6po_route; dst = (struct sockaddr_in6 *)&ro->ro_dst; #ifdef FLOWTABLE if (ro->ro_rt == NULL) (void )flowtable_lookup(AF_INET6, m, (struct route *)ro); #endif fibnum = (inp != NULL) ? inp->inp_inc.inc_fibnum : M_GETFIB(m); again: /* * if specified, try to fill in the traffic class field. * do not override if a non-zero value is already set. * we check the diffserv field and the ecn field separately. */ if (opt && opt->ip6po_tclass >= 0) { int mask = 0; if ((ip6->ip6_flow & htonl(0xfc << 20)) == 0) mask |= 0xfc; if ((ip6->ip6_flow & htonl(0x03 << 20)) == 0) mask |= 0x03; if (mask != 0) ip6->ip6_flow |= htonl((opt->ip6po_tclass & mask) << 20); } /* fill in or override the hop limit field, if necessary. */ if (opt && opt->ip6po_hlim != -1) ip6->ip6_hlim = opt->ip6po_hlim & 0xff; else if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { if (im6o != NULL) ip6->ip6_hlim = im6o->im6o_multicast_hlim; else ip6->ip6_hlim = V_ip6_defmcasthlim; } /* adjust pointer */ ip6 = mtod(m, struct ip6_hdr *); if (ro->ro_rt && fwd_tag == NULL) { rt = ro->ro_rt; ifp = ro->ro_rt->rt_ifp; } else { if (fwd_tag == NULL) { bzero(&dst_sa, sizeof(dst_sa)); dst_sa.sin6_family = AF_INET6; dst_sa.sin6_len = sizeof(dst_sa); dst_sa.sin6_addr = ip6->ip6_dst; } error = in6_selectroute_fib(&dst_sa, opt, im6o, ro, &ifp, &rt, fibnum); if (error != 0) { if (ifp != NULL) in6_ifstat_inc(ifp, ifs6_out_discard); goto bad; } } if (rt == NULL) { /* * If in6_selectroute() does not return a route entry, * dst may not have been updated. */ *dst = dst_sa; /* XXX */ } /* * then rt (for unicast) and ifp must be non-NULL valid values. */ if ((flags & IPV6_FORWARDING) == 0) { /* XXX: the FORWARDING flag can be set for mrouting. */ in6_ifstat_inc(ifp, ifs6_out_request); } if (rt != NULL) { ia = (struct in6_ifaddr *)(rt->rt_ifa); counter_u64_add(rt->rt_pksent, 1); } /* * The outgoing interface must be in the zone of source and * destination addresses. */ origifp = ifp; src0 = ip6->ip6_src; if (in6_setscope(&src0, origifp, &zone)) goto badscope; bzero(&src_sa, sizeof(src_sa)); src_sa.sin6_family = AF_INET6; src_sa.sin6_len = sizeof(src_sa); src_sa.sin6_addr = ip6->ip6_src; if (sa6_recoverscope(&src_sa) || zone != src_sa.sin6_scope_id) goto badscope; dst0 = ip6->ip6_dst; if (in6_setscope(&dst0, origifp, &zone)) goto badscope; /* re-initialize to be sure */ bzero(&dst_sa, sizeof(dst_sa)); dst_sa.sin6_family = AF_INET6; dst_sa.sin6_len = sizeof(dst_sa); dst_sa.sin6_addr = ip6->ip6_dst; if (sa6_recoverscope(&dst_sa) || zone != dst_sa.sin6_scope_id) { goto badscope; } /* We should use ia_ifp to support the case of * sending packets to an address of our own. */ if (ia != NULL && ia->ia_ifp) ifp = ia->ia_ifp; /* scope check is done. */ goto routefound; badscope: IP6STAT_INC(ip6s_badscope); in6_ifstat_inc(origifp, ifs6_out_discard); if (error == 0) error = EHOSTUNREACH; /* XXX */ goto bad; routefound: if (rt && !IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { if (opt && opt->ip6po_nextroute.ro_rt) { /* * The nexthop is explicitly specified by the * application. We assume the next hop is an IPv6 * address. */ dst = (struct sockaddr_in6 *)opt->ip6po_nexthop; } else if ((rt->rt_flags & RTF_GATEWAY)) dst = (struct sockaddr_in6 *)rt->rt_gateway; } if (!IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { m->m_flags &= ~(M_BCAST | M_MCAST); /* just in case */ } else { m->m_flags = (m->m_flags & ~M_BCAST) | M_MCAST; in6_ifstat_inc(ifp, ifs6_out_mcast); /* * Confirm that the outgoing interface supports multicast. */ if (!(ifp->if_flags & IFF_MULTICAST)) { IP6STAT_INC(ip6s_noroute); in6_ifstat_inc(ifp, ifs6_out_discard); error = ENETUNREACH; goto bad; } if ((im6o == NULL && in6_mcast_loop) || (im6o && im6o->im6o_multicast_loop)) { /* * Loop back multicast datagram if not expressly * forbidden to do so, even if we have not joined * the address; protocols will filter it later, * thus deferring a hash lookup and lock acquisition * at the expense of an m_copym(). */ ip6_mloopback(ifp, m); } else { /* * If we are acting as a multicast router, perform * multicast forwarding as if the packet had just * arrived on the interface to which we are about * to send. The multicast forwarding function * recursively calls this function, using the * IPV6_FORWARDING flag to prevent infinite recursion. * * Multicasts that are looped back by ip6_mloopback(), * above, will be forwarded by the ip6_input() routine, * if necessary. */ if (V_ip6_mrouter && (flags & IPV6_FORWARDING) == 0) { /* * XXX: ip6_mforward expects that rcvif is NULL * when it is called from the originating path. * However, it may not always be the case. */ m->m_pkthdr.rcvif = NULL; if (ip6_mforward(ip6, ifp, m) != 0) { m_freem(m); goto done; } } } /* * Multicasts with a hoplimit of zero may be looped back, * above, but must not be transmitted on a network. * Also, multicasts addressed to the loopback interface * are not sent -- the above call to ip6_mloopback() will * loop back a copy if this host actually belongs to the * destination group on the loopback interface. */ if (ip6->ip6_hlim == 0 || (ifp->if_flags & IFF_LOOPBACK) || IN6_IS_ADDR_MC_INTFACELOCAL(&ip6->ip6_dst)) { m_freem(m); goto done; } } /* * Fill the outgoing inteface to tell the upper layer * to increment per-interface statistics. */ if (ifpp) *ifpp = ifp; /* Determine path MTU. */ if ((error = ip6_getpmtu(ro_pmtu, ro != ro_pmtu, ifp, &finaldst, &mtu, &alwaysfrag, fibnum)) != 0) goto bad; /* * The caller of this function may specify to use the minimum MTU * in some cases. * An advanced API option (IPV6_USE_MIN_MTU) can also override MTU * setting. The logic is a bit complicated; by default, unicast * packets will follow path MTU while multicast packets will be sent at * the minimum MTU. If IP6PO_MINMTU_ALL is specified, all packets * including unicast ones will be sent at the minimum MTU. Multicast * packets will always be sent at the minimum MTU unless * IP6PO_MINMTU_DISABLE is explicitly specified. * See RFC 3542 for more details. */ if (mtu > IPV6_MMTU) { if ((flags & IPV6_MINMTU)) mtu = IPV6_MMTU; else if (opt && opt->ip6po_minmtu == IP6PO_MINMTU_ALL) mtu = IPV6_MMTU; else if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) && (opt == NULL || opt->ip6po_minmtu != IP6PO_MINMTU_DISABLE)) { mtu = IPV6_MMTU; } } /* * clear embedded scope identifiers if necessary. * in6_clearscope will touch the addresses only when necessary. */ in6_clearscope(&ip6->ip6_src); in6_clearscope(&ip6->ip6_dst); /* * If the outgoing packet contains a hop-by-hop options header, * it must be examined and processed even by the source node. * (RFC 2460, section 4.) */ if (exthdrs.ip6e_hbh) { struct ip6_hbh *hbh = mtod(exthdrs.ip6e_hbh, struct ip6_hbh *); u_int32_t dummy; /* XXX unused */ u_int32_t plen = 0; /* XXX: ip6_process will check the value */ #ifdef DIAGNOSTIC if ((hbh->ip6h_len + 1) << 3 > exthdrs.ip6e_hbh->m_len) panic("ip6e_hbh is not contiguous"); #endif /* * XXX: if we have to send an ICMPv6 error to the sender, * we need the M_LOOP flag since icmp6_error() expects * the IPv6 and the hop-by-hop options header are * contiguous unless the flag is set. */ m->m_flags |= M_LOOP; m->m_pkthdr.rcvif = ifp; if (ip6_process_hopopts(m, (u_int8_t *)(hbh + 1), ((hbh->ip6h_len + 1) << 3) - sizeof(struct ip6_hbh), &dummy, &plen) < 0) { /* m was already freed at this point */ error = EINVAL;/* better error? */ goto done; } m->m_flags &= ~M_LOOP; /* XXX */ m->m_pkthdr.rcvif = NULL; } /* Jump over all PFIL processing if hooks are not active. */ if (!PFIL_HOOKED(&V_inet6_pfil_hook)) goto passout; odst = ip6->ip6_dst; /* Run through list of hooks for output packets. */ error = pfil_run_hooks(&V_inet6_pfil_hook, &m, ifp, PFIL_OUT, inp); if (error != 0 || m == NULL) goto done; ip6 = mtod(m, struct ip6_hdr *); needfiblookup = 0; /* See if destination IP address was changed by packet filter. */ if (!IN6_ARE_ADDR_EQUAL(&odst, &ip6->ip6_dst)) { m->m_flags |= M_SKIP_FIREWALL; /* If destination is now ourself drop to ip6_input(). */ if (in6_localip(&ip6->ip6_dst)) { m->m_flags |= M_FASTFWD_OURS; if (m->m_pkthdr.rcvif == NULL) m->m_pkthdr.rcvif = V_loif; if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) { m->m_pkthdr.csum_flags |= CSUM_DATA_VALID_IPV6 | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xffff; } #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; #endif error = netisr_queue(NETISR_IPV6, m); goto done; } else needfiblookup = 1; /* Redo the routing table lookup. */ } /* See if fib was changed by packet filter. */ if (fibnum != M_GETFIB(m)) { m->m_flags |= M_SKIP_FIREWALL; fibnum = M_GETFIB(m); RO_RTFREE(ro); needfiblookup = 1; } if (needfiblookup) goto again; /* See if local, if yes, send it to netisr. */ if (m->m_flags & M_FASTFWD_OURS) { if (m->m_pkthdr.rcvif == NULL) m->m_pkthdr.rcvif = V_loif; if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) { m->m_pkthdr.csum_flags |= CSUM_DATA_VALID_IPV6 | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xffff; } #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; #endif error = netisr_queue(NETISR_IPV6, m); goto done; } /* Or forward to some other address? */ if ((m->m_flags & M_IP6_NEXTHOP) && (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { dst = (struct sockaddr_in6 *)&ro->ro_dst; bcopy((fwd_tag+1), &dst_sa, sizeof(struct sockaddr_in6)); m->m_flags |= M_SKIP_FIREWALL; m->m_flags &= ~M_IP6_NEXTHOP; m_tag_delete(m, fwd_tag); goto again; } passout: /* * Send the packet to the outgoing interface. * If necessary, do IPv6 fragmentation before sending. * * the logic here is rather complex: * 1: normal case (dontfrag == 0, alwaysfrag == 0) * 1-a: send as is if tlen <= path mtu * 1-b: fragment if tlen > path mtu * * 2: if user asks us not to fragment (dontfrag == 1) * 2-a: send as is if tlen <= interface mtu * 2-b: error if tlen > interface mtu * * 3: if we always need to attach fragment header (alwaysfrag == 1) * always fragment * * 4: if dontfrag == 1 && alwaysfrag == 1 * error, as we cannot handle this conflicting request */ sw_csum = m->m_pkthdr.csum_flags; if (!hdrsplit) { tso = ((sw_csum & ifp->if_hwassist & CSUM_TSO) != 0) ? 1 : 0; sw_csum &= ~ifp->if_hwassist; } else tso = 0; /* * If we added extension headers, we will not do TSO and calculate the * checksums ourselves for now. * XXX-BZ Need a framework to know when the NIC can handle it, even * with ext. hdrs. */ if (sw_csum & CSUM_DELAY_DATA_IPV6) { sw_csum &= ~CSUM_DELAY_DATA_IPV6; in6_delayed_cksum(m, plen, sizeof(struct ip6_hdr)); } #ifdef SCTP if (sw_csum & CSUM_SCTP_IPV6) { sw_csum &= ~CSUM_SCTP_IPV6; sctp_delayed_cksum(m, sizeof(struct ip6_hdr)); } #endif m->m_pkthdr.csum_flags &= ifp->if_hwassist; tlen = m->m_pkthdr.len; if ((opt && (opt->ip6po_flags & IP6PO_DONTFRAG)) || tso) dontfrag = 1; else dontfrag = 0; if (dontfrag && alwaysfrag) { /* case 4 */ /* conflicting request - can't transmit */ error = EMSGSIZE; goto bad; } if (dontfrag && tlen > IN6_LINKMTU(ifp) && !tso) { /* case 2-b */ /* * Even if the DONTFRAG option is specified, we cannot send the * packet when the data length is larger than the MTU of the * outgoing interface. * Notify the error by sending IPV6_PATHMTU ancillary data if * application wanted to know the MTU value. Also return an * error code (this is not described in the API spec). */ if (inp != NULL) ip6_notify_pmtu(inp, &dst_sa, (u_int32_t)mtu); error = EMSGSIZE; goto bad; } /* * transmit packet without fragmentation */ if (dontfrag || (!alwaysfrag && tlen <= mtu)) { /* case 1-a and 2-a */ struct in6_ifaddr *ia6; ip6 = mtod(m, struct ip6_hdr *); ia6 = in6_ifawithifp(ifp, &ip6->ip6_src); if (ia6) { /* Record statistics for this interface address. */ counter_u64_add(ia6->ia_ifa.ifa_opackets, 1); counter_u64_add(ia6->ia_ifa.ifa_obytes, m->m_pkthdr.len); ifa_free(&ia6->ia_ifa); } error = nd6_output_ifp(ifp, origifp, m, dst, NULL); goto done; } /* * try to fragment the packet. case 1-b and 3 */ if (mtu < IPV6_MMTU) { /* path MTU cannot be less than IPV6_MMTU */ error = EMSGSIZE; in6_ifstat_inc(ifp, ifs6_out_fragfail); goto bad; } else if (ip6->ip6_plen == 0) { /* jumbo payload cannot be fragmented */ error = EMSGSIZE; in6_ifstat_inc(ifp, ifs6_out_fragfail); goto bad; } else { u_char nextproto; /* * Too large for the destination or interface; * fragment if possible. * Must be able to put at least 8 bytes per fragment. */ hlen = unfragpartlen; if (mtu > IPV6_MAXPACKET) mtu = IPV6_MAXPACKET; len = (mtu - hlen - sizeof(struct ip6_frag)) & ~7; if (len < 8) { error = EMSGSIZE; in6_ifstat_inc(ifp, ifs6_out_fragfail); goto bad; } /* * If the interface will not calculate checksums on * fragmented packets, then do it here. * XXX-BZ handle the hw offloading case. Need flags. */ if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) { in6_delayed_cksum(m, plen, hlen); m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; } #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) { sctp_delayed_cksum(m, hlen); m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6; } #endif /* * Change the next header field of the last header in the * unfragmentable part. */ if (exthdrs.ip6e_rthdr) { nextproto = *mtod(exthdrs.ip6e_rthdr, u_char *); *mtod(exthdrs.ip6e_rthdr, u_char *) = IPPROTO_FRAGMENT; } else if (exthdrs.ip6e_dest1) { nextproto = *mtod(exthdrs.ip6e_dest1, u_char *); *mtod(exthdrs.ip6e_dest1, u_char *) = IPPROTO_FRAGMENT; } else if (exthdrs.ip6e_hbh) { nextproto = *mtod(exthdrs.ip6e_hbh, u_char *); *mtod(exthdrs.ip6e_hbh, u_char *) = IPPROTO_FRAGMENT; } else { nextproto = ip6->ip6_nxt; ip6->ip6_nxt = IPPROTO_FRAGMENT; } /* * Loop through length of segment after first fragment, * make new header and copy data of each part and link onto * chain. */ m0 = m; id = htonl(ip6_randomid()); if ((error = ip6_fragment(ifp, m, hlen, nextproto, len, id))) goto sendorfree; in6_ifstat_inc(ifp, ifs6_out_fragok); } /* * Remove leading garbages. */ sendorfree: m = m0->m_nextpkt; m0->m_nextpkt = 0; m_freem(m0); for (m0 = m; m; m = m0) { m0 = m->m_nextpkt; m->m_nextpkt = 0; if (error == 0) { /* Record statistics for this interface address. */ if (ia) { counter_u64_add(ia->ia_ifa.ifa_opackets, 1); counter_u64_add(ia->ia_ifa.ifa_obytes, m->m_pkthdr.len); } error = nd6_output_ifp(ifp, origifp, m, dst, NULL); } else m_freem(m); } if (error == 0) IP6STAT_INC(ip6s_fragmented); done: if (ro == &ip6route) RO_RTFREE(ro); return (error); freehdrs: m_freem(exthdrs.ip6e_hbh); /* m_freem will check if mbuf is 0 */ m_freem(exthdrs.ip6e_dest1); m_freem(exthdrs.ip6e_rthdr); m_freem(exthdrs.ip6e_dest2); /* FALLTHROUGH */ bad: if (m) m_freem(m); goto done; } static int ip6_copyexthdr(struct mbuf **mp, caddr_t hdr, int hlen) { struct mbuf *m; if (hlen > MCLBYTES) return (ENOBUFS); /* XXX */ if (hlen > MLEN) m = m_getcl(M_NOWAIT, MT_DATA, 0); else m = m_get(M_NOWAIT, MT_DATA); if (m == NULL) return (ENOBUFS); m->m_len = hlen; if (hdr) bcopy(hdr, mtod(m, caddr_t), hlen); *mp = m; return (0); } /* * Insert jumbo payload option. */ static int ip6_insert_jumboopt(struct ip6_exthdrs *exthdrs, u_int32_t plen) { struct mbuf *mopt; u_char *optbuf; u_int32_t v; #define JUMBOOPTLEN 8 /* length of jumbo payload option and padding */ /* * If there is no hop-by-hop options header, allocate new one. * If there is one but it doesn't have enough space to store the * jumbo payload option, allocate a cluster to store the whole options. * Otherwise, use it to store the options. */ if (exthdrs->ip6e_hbh == 0) { mopt = m_get(M_NOWAIT, MT_DATA); if (mopt == NULL) return (ENOBUFS); mopt->m_len = JUMBOOPTLEN; optbuf = mtod(mopt, u_char *); optbuf[1] = 0; /* = ((JUMBOOPTLEN) >> 3) - 1 */ exthdrs->ip6e_hbh = mopt; } else { struct ip6_hbh *hbh; mopt = exthdrs->ip6e_hbh; if (M_TRAILINGSPACE(mopt) < JUMBOOPTLEN) { /* * XXX assumption: * - exthdrs->ip6e_hbh is not referenced from places * other than exthdrs. * - exthdrs->ip6e_hbh is not an mbuf chain. */ int oldoptlen = mopt->m_len; struct mbuf *n; /* * XXX: give up if the whole (new) hbh header does * not fit even in an mbuf cluster. */ if (oldoptlen + JUMBOOPTLEN > MCLBYTES) return (ENOBUFS); /* * As a consequence, we must always prepare a cluster * at this point. */ n = m_getcl(M_NOWAIT, MT_DATA, 0); if (n == NULL) return (ENOBUFS); n->m_len = oldoptlen + JUMBOOPTLEN; bcopy(mtod(mopt, caddr_t), mtod(n, caddr_t), oldoptlen); optbuf = mtod(n, caddr_t) + oldoptlen; m_freem(mopt); mopt = exthdrs->ip6e_hbh = n; } else { optbuf = mtod(mopt, u_char *) + mopt->m_len; mopt->m_len += JUMBOOPTLEN; } optbuf[0] = IP6OPT_PADN; optbuf[1] = 1; /* * Adjust the header length according to the pad and * the jumbo payload option. */ hbh = mtod(mopt, struct ip6_hbh *); hbh->ip6h_len += (JUMBOOPTLEN >> 3); } /* fill in the option. */ optbuf[2] = IP6OPT_JUMBO; optbuf[3] = 4; v = (u_int32_t)htonl(plen + JUMBOOPTLEN); bcopy(&v, &optbuf[4], sizeof(u_int32_t)); /* finally, adjust the packet header length */ exthdrs->ip6e_ip6->m_pkthdr.len += JUMBOOPTLEN; return (0); #undef JUMBOOPTLEN } /* * Insert fragment header and copy unfragmentable header portions. */ static int ip6_insertfraghdr(struct mbuf *m0, struct mbuf *m, int hlen, struct ip6_frag **frghdrp) { struct mbuf *n, *mlast; if (hlen > sizeof(struct ip6_hdr)) { n = m_copym(m0, sizeof(struct ip6_hdr), hlen - sizeof(struct ip6_hdr), M_NOWAIT); if (n == 0) return (ENOBUFS); m->m_next = n; } else n = m; /* Search for the last mbuf of unfragmentable part. */ for (mlast = n; mlast->m_next; mlast = mlast->m_next) ; if (M_WRITABLE(mlast) && M_TRAILINGSPACE(mlast) >= sizeof(struct ip6_frag)) { /* use the trailing space of the last mbuf for the fragment hdr */ *frghdrp = (struct ip6_frag *)(mtod(mlast, caddr_t) + mlast->m_len); mlast->m_len += sizeof(struct ip6_frag); m->m_pkthdr.len += sizeof(struct ip6_frag); } else { /* allocate a new mbuf for the fragment header */ struct mbuf *mfrg; mfrg = m_get(M_NOWAIT, MT_DATA); if (mfrg == NULL) return (ENOBUFS); mfrg->m_len = sizeof(struct ip6_frag); *frghdrp = mtod(mfrg, struct ip6_frag *); mlast->m_next = mfrg; } return (0); } /* * Calculates IPv6 path mtu for destination @dst. * Resulting MTU is stored in @mtup. * * Returns 0 on success. */ static int ip6_getpmtu_ctl(u_int fibnum, struct in6_addr *dst, u_long *mtup) { struct nhop6_extended nh6; struct in6_addr kdst; uint32_t scopeid; struct ifnet *ifp; u_long mtu; int error; in6_splitscope(dst, &kdst, &scopeid); if (fib6_lookup_nh_ext(fibnum, &kdst, scopeid, NHR_REF, 0, &nh6) != 0) return (EHOSTUNREACH); ifp = nh6.nh_ifp; mtu = nh6.nh_mtu; error = ip6_calcmtu(ifp, dst, mtu, mtup, NULL); fib6_free_nh_ext(fibnum, &nh6); return (error); } /* * Calculates IPv6 path MTU for @dst based on transmit @ifp, * and cached data in @ro_pmtu. * MTU from (successful) route lookup is saved (along with dst) * inside @ro_pmtu to avoid subsequent route lookups after packet * filter processing. * * Stores mtu and always-frag value into @mtup and @alwaysfragp. * Returns 0 on success. */ static int ip6_getpmtu(struct route_in6 *ro_pmtu, int do_lookup, struct ifnet *ifp, struct in6_addr *dst, u_long *mtup, int *alwaysfragp, u_int fibnum) { struct nhop6_basic nh6; struct in6_addr kdst; uint32_t scopeid; struct sockaddr_in6 *sa6_dst; u_long mtu; mtu = 0; if (do_lookup) { /* * Here ro_pmtu has final destination address, while * ro might represent immediate destination. * Use ro_pmtu destination since mtu might differ. */ sa6_dst = (struct sockaddr_in6 *)&ro_pmtu->ro_dst; if (!IN6_ARE_ADDR_EQUAL(&sa6_dst->sin6_addr, dst)) ro_pmtu->ro_mtu = 0; if (ro_pmtu->ro_mtu == 0) { bzero(sa6_dst, sizeof(*sa6_dst)); sa6_dst->sin6_family = AF_INET6; sa6_dst->sin6_len = sizeof(struct sockaddr_in6); sa6_dst->sin6_addr = *dst; in6_splitscope(dst, &kdst, &scopeid); if (fib6_lookup_nh_basic(fibnum, &kdst, scopeid, 0, 0, &nh6) == 0) ro_pmtu->ro_mtu = nh6.nh_mtu; } mtu = ro_pmtu->ro_mtu; } if (ro_pmtu->ro_rt) mtu = ro_pmtu->ro_rt->rt_mtu; return (ip6_calcmtu(ifp, dst, mtu, mtup, alwaysfragp)); } /* * Calculate MTU based on transmit @ifp, route mtu @rt_mtu and * hostcache data for @dst. * Stores mtu and always-frag value into @mtup and @alwaysfragp. * * Returns 0 on success. */ static int ip6_calcmtu(struct ifnet *ifp, const struct in6_addr *dst, u_long rt_mtu, u_long *mtup, int *alwaysfragp) { u_long mtu = 0; int alwaysfrag = 0; int error = 0; if (rt_mtu > 0) { u_int32_t ifmtu; struct in_conninfo inc; bzero(&inc, sizeof(inc)); inc.inc_flags |= INC_ISIPV6; inc.inc6_faddr = *dst; ifmtu = IN6_LINKMTU(ifp); mtu = tcp_hc_getmtu(&inc); if (mtu) mtu = min(mtu, rt_mtu); else mtu = rt_mtu; if (mtu == 0) mtu = ifmtu; else if (mtu < IPV6_MMTU) { /* * RFC2460 section 5, last paragraph: * if we record ICMPv6 too big message with * mtu < IPV6_MMTU, transmit packets sized IPV6_MMTU * or smaller, with framgent header attached. * (fragment header is needed regardless from the * packet size, for translators to identify packets) */ alwaysfrag = 1; mtu = IPV6_MMTU; } } else if (ifp) { mtu = IN6_LINKMTU(ifp); } else error = EHOSTUNREACH; /* XXX */ *mtup = mtu; if (alwaysfragp) *alwaysfragp = alwaysfrag; return (error); } /* * IP6 socket option processing. */ int ip6_ctloutput(struct socket *so, struct sockopt *sopt) { int optdatalen, uproto; void *optdata; struct inpcb *in6p = sotoinpcb(so); int error, optval; int level, op, optname; int optlen; struct thread *td; #ifdef RSS uint32_t rss_bucket; int retval; #endif level = sopt->sopt_level; op = sopt->sopt_dir; optname = sopt->sopt_name; optlen = sopt->sopt_valsize; td = sopt->sopt_td; error = 0; optval = 0; uproto = (int)so->so_proto->pr_protocol; if (level != IPPROTO_IPV6) { error = EINVAL; if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_dir == SOPT_SET) { switch (sopt->sopt_name) { case SO_REUSEADDR: INP_WLOCK(in6p); if ((so->so_options & SO_REUSEADDR) != 0) in6p->inp_flags2 |= INP_REUSEADDR; else in6p->inp_flags2 &= ~INP_REUSEADDR; INP_WUNLOCK(in6p); error = 0; break; case SO_REUSEPORT: INP_WLOCK(in6p); if ((so->so_options & SO_REUSEPORT) != 0) in6p->inp_flags2 |= INP_REUSEPORT; else in6p->inp_flags2 &= ~INP_REUSEPORT; INP_WUNLOCK(in6p); error = 0; break; case SO_SETFIB: INP_WLOCK(in6p); in6p->inp_inc.inc_fibnum = so->so_fibnum; INP_WUNLOCK(in6p); error = 0; break; default: break; } } } else { /* level == IPPROTO_IPV6 */ switch (op) { case SOPT_SET: switch (optname) { case IPV6_2292PKTOPTIONS: #ifdef IPV6_PKTOPTIONS case IPV6_PKTOPTIONS: #endif { struct mbuf *m; error = soopt_getm(sopt, &m); /* XXX */ if (error != 0) break; error = soopt_mcopyin(sopt, m); /* XXX */ if (error != 0) break; error = ip6_pcbopts(&in6p->in6p_outputopts, m, so, sopt); m_freem(m); /* XXX */ break; } /* * Use of some Hop-by-Hop options or some * Destination options, might require special * privilege. That is, normal applications * (without special privilege) might be forbidden * from setting certain options in outgoing packets, * and might never see certain options in received * packets. [RFC 2292 Section 6] * KAME specific note: * KAME prevents non-privileged users from sending or * receiving ANY hbh/dst options in order to avoid * overhead of parsing options in the kernel. */ case IPV6_RECVHOPOPTS: case IPV6_RECVDSTOPTS: case IPV6_RECVRTHDRDSTOPTS: if (td != NULL) { error = priv_check(td, PRIV_NETINET_SETHDROPTS); if (error) break; } /* FALLTHROUGH */ case IPV6_UNICAST_HOPS: case IPV6_HOPLIMIT: case IPV6_RECVPKTINFO: case IPV6_RECVHOPLIMIT: case IPV6_RECVRTHDR: case IPV6_RECVPATHMTU: case IPV6_RECVTCLASS: case IPV6_RECVFLOWID: #ifdef RSS case IPV6_RECVRSSBUCKETID: #endif case IPV6_V6ONLY: case IPV6_AUTOFLOWLABEL: case IPV6_BINDANY: case IPV6_BINDMULTI: #ifdef RSS case IPV6_RSS_LISTEN_BUCKET: #endif if (optname == IPV6_BINDANY && td != NULL) { error = priv_check(td, PRIV_NETINET_BINDANY); if (error) break; } if (optlen != sizeof(int)) { error = EINVAL; break; } error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; switch (optname) { case IPV6_UNICAST_HOPS: if (optval < -1 || optval >= 256) error = EINVAL; else { /* -1 = kernel default */ in6p->in6p_hops = optval; if ((in6p->inp_vflag & INP_IPV4) != 0) in6p->inp_ip_ttl = optval; } break; #define OPTSET(bit) \ do { \ INP_WLOCK(in6p); \ if (optval) \ in6p->inp_flags |= (bit); \ else \ in6p->inp_flags &= ~(bit); \ INP_WUNLOCK(in6p); \ } while (/*CONSTCOND*/ 0) #define OPTSET2292(bit) \ do { \ INP_WLOCK(in6p); \ in6p->inp_flags |= IN6P_RFC2292; \ if (optval) \ in6p->inp_flags |= (bit); \ else \ in6p->inp_flags &= ~(bit); \ INP_WUNLOCK(in6p); \ } while (/*CONSTCOND*/ 0) #define OPTBIT(bit) (in6p->inp_flags & (bit) ? 1 : 0) #define OPTSET2(bit, val) do { \ INP_WLOCK(in6p); \ if (val) \ in6p->inp_flags2 |= bit; \ else \ in6p->inp_flags2 &= ~bit; \ INP_WUNLOCK(in6p); \ } while (0) #define OPTBIT2(bit) (in6p->inp_flags2 & (bit) ? 1 : 0) case IPV6_RECVPKTINFO: /* cannot mix with RFC2292 */ if (OPTBIT(IN6P_RFC2292)) { error = EINVAL; break; } OPTSET(IN6P_PKTINFO); break; case IPV6_HOPLIMIT: { struct ip6_pktopts **optp; /* cannot mix with RFC2292 */ if (OPTBIT(IN6P_RFC2292)) { error = EINVAL; break; } optp = &in6p->in6p_outputopts; error = ip6_pcbopt(IPV6_HOPLIMIT, (u_char *)&optval, sizeof(optval), optp, (td != NULL) ? td->td_ucred : NULL, uproto); break; } case IPV6_RECVHOPLIMIT: /* cannot mix with RFC2292 */ if (OPTBIT(IN6P_RFC2292)) { error = EINVAL; break; } OPTSET(IN6P_HOPLIMIT); break; case IPV6_RECVHOPOPTS: /* cannot mix with RFC2292 */ if (OPTBIT(IN6P_RFC2292)) { error = EINVAL; break; } OPTSET(IN6P_HOPOPTS); break; case IPV6_RECVDSTOPTS: /* cannot mix with RFC2292 */ if (OPTBIT(IN6P_RFC2292)) { error = EINVAL; break; } OPTSET(IN6P_DSTOPTS); break; case IPV6_RECVRTHDRDSTOPTS: /* cannot mix with RFC2292 */ if (OPTBIT(IN6P_RFC2292)) { error = EINVAL; break; } OPTSET(IN6P_RTHDRDSTOPTS); break; case IPV6_RECVRTHDR: /* cannot mix with RFC2292 */ if (OPTBIT(IN6P_RFC2292)) { error = EINVAL; break; } OPTSET(IN6P_RTHDR); break; case IPV6_RECVPATHMTU: /* * We ignore this option for TCP * sockets. * (RFC3542 leaves this case * unspecified.) */ if (uproto != IPPROTO_TCP) OPTSET(IN6P_MTU); break; case IPV6_RECVFLOWID: OPTSET2(INP_RECVFLOWID, optval); break; #ifdef RSS case IPV6_RECVRSSBUCKETID: OPTSET2(INP_RECVRSSBUCKETID, optval); break; #endif case IPV6_V6ONLY: /* * make setsockopt(IPV6_V6ONLY) * available only prior to bind(2). * see ipng mailing list, Jun 22 2001. */ if (in6p->inp_lport || !IN6_IS_ADDR_UNSPECIFIED(&in6p->in6p_laddr)) { error = EINVAL; break; } OPTSET(IN6P_IPV6_V6ONLY); if (optval) in6p->inp_vflag &= ~INP_IPV4; else in6p->inp_vflag |= INP_IPV4; break; case IPV6_RECVTCLASS: /* cannot mix with RFC2292 XXX */ if (OPTBIT(IN6P_RFC2292)) { error = EINVAL; break; } OPTSET(IN6P_TCLASS); break; case IPV6_AUTOFLOWLABEL: OPTSET(IN6P_AUTOFLOWLABEL); break; case IPV6_BINDANY: OPTSET(INP_BINDANY); break; case IPV6_BINDMULTI: OPTSET2(INP_BINDMULTI, optval); break; #ifdef RSS case IPV6_RSS_LISTEN_BUCKET: if ((optval >= 0) && (optval < rss_getnumbuckets())) { in6p->inp_rss_listen_bucket = optval; OPTSET2(INP_RSS_BUCKET_SET, 1); } else { error = EINVAL; } break; #endif } break; case IPV6_TCLASS: case IPV6_DONTFRAG: case IPV6_USE_MIN_MTU: case IPV6_PREFER_TEMPADDR: if (optlen != sizeof(optval)) { error = EINVAL; break; } error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; { struct ip6_pktopts **optp; optp = &in6p->in6p_outputopts; error = ip6_pcbopt(optname, (u_char *)&optval, sizeof(optval), optp, (td != NULL) ? td->td_ucred : NULL, uproto); break; } case IPV6_2292PKTINFO: case IPV6_2292HOPLIMIT: case IPV6_2292HOPOPTS: case IPV6_2292DSTOPTS: case IPV6_2292RTHDR: /* RFC 2292 */ if (optlen != sizeof(int)) { error = EINVAL; break; } error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; switch (optname) { case IPV6_2292PKTINFO: OPTSET2292(IN6P_PKTINFO); break; case IPV6_2292HOPLIMIT: OPTSET2292(IN6P_HOPLIMIT); break; case IPV6_2292HOPOPTS: /* * Check super-user privilege. * See comments for IPV6_RECVHOPOPTS. */ if (td != NULL) { error = priv_check(td, PRIV_NETINET_SETHDROPTS); if (error) return (error); } OPTSET2292(IN6P_HOPOPTS); break; case IPV6_2292DSTOPTS: if (td != NULL) { error = priv_check(td, PRIV_NETINET_SETHDROPTS); if (error) return (error); } OPTSET2292(IN6P_DSTOPTS|IN6P_RTHDRDSTOPTS); /* XXX */ break; case IPV6_2292RTHDR: OPTSET2292(IN6P_RTHDR); break; } break; case IPV6_PKTINFO: case IPV6_HOPOPTS: case IPV6_RTHDR: case IPV6_DSTOPTS: case IPV6_RTHDRDSTOPTS: case IPV6_NEXTHOP: { /* new advanced API (RFC3542) */ u_char *optbuf; u_char optbuf_storage[MCLBYTES]; int optlen; struct ip6_pktopts **optp; /* cannot mix with RFC2292 */ if (OPTBIT(IN6P_RFC2292)) { error = EINVAL; break; } /* * We only ensure valsize is not too large * here. Further validation will be done * later. */ error = sooptcopyin(sopt, optbuf_storage, sizeof(optbuf_storage), 0); if (error) break; optlen = sopt->sopt_valsize; optbuf = optbuf_storage; optp = &in6p->in6p_outputopts; error = ip6_pcbopt(optname, optbuf, optlen, optp, (td != NULL) ? td->td_ucred : NULL, uproto); break; } #undef OPTSET case IPV6_MULTICAST_IF: case IPV6_MULTICAST_HOPS: case IPV6_MULTICAST_LOOP: case IPV6_JOIN_GROUP: case IPV6_LEAVE_GROUP: case IPV6_MSFILTER: case MCAST_BLOCK_SOURCE: case MCAST_UNBLOCK_SOURCE: case MCAST_JOIN_GROUP: case MCAST_LEAVE_GROUP: case MCAST_JOIN_SOURCE_GROUP: case MCAST_LEAVE_SOURCE_GROUP: error = ip6_setmoptions(in6p, sopt); break; case IPV6_PORTRANGE: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; INP_WLOCK(in6p); switch (optval) { case IPV6_PORTRANGE_DEFAULT: in6p->inp_flags &= ~(INP_LOWPORT); in6p->inp_flags &= ~(INP_HIGHPORT); break; case IPV6_PORTRANGE_HIGH: in6p->inp_flags &= ~(INP_LOWPORT); in6p->inp_flags |= INP_HIGHPORT; break; case IPV6_PORTRANGE_LOW: in6p->inp_flags &= ~(INP_HIGHPORT); in6p->inp_flags |= INP_LOWPORT; break; default: error = EINVAL; break; } INP_WUNLOCK(in6p); break; #ifdef IPSEC case IPV6_IPSEC_POLICY: { caddr_t req; struct mbuf *m; if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */ break; if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */ break; req = mtod(m, caddr_t); error = ipsec_set_policy(in6p, optname, req, m->m_len, (sopt->sopt_td != NULL) ? sopt->sopt_td->td_ucred : NULL); m_freem(m); break; } #endif /* IPSEC */ default: error = ENOPROTOOPT; break; } break; case SOPT_GET: switch (optname) { case IPV6_2292PKTOPTIONS: #ifdef IPV6_PKTOPTIONS case IPV6_PKTOPTIONS: #endif /* * RFC3542 (effectively) deprecated the * semantics of the 2292-style pktoptions. * Since it was not reliable in nature (i.e., * applications had to expect the lack of some * information after all), it would make sense * to simplify this part by always returning * empty data. */ sopt->sopt_valsize = 0; break; case IPV6_RECVHOPOPTS: case IPV6_RECVDSTOPTS: case IPV6_RECVRTHDRDSTOPTS: case IPV6_UNICAST_HOPS: case IPV6_RECVPKTINFO: case IPV6_RECVHOPLIMIT: case IPV6_RECVRTHDR: case IPV6_RECVPATHMTU: case IPV6_V6ONLY: case IPV6_PORTRANGE: case IPV6_RECVTCLASS: case IPV6_AUTOFLOWLABEL: case IPV6_BINDANY: case IPV6_FLOWID: case IPV6_FLOWTYPE: case IPV6_RECVFLOWID: #ifdef RSS case IPV6_RSSBUCKETID: case IPV6_RECVRSSBUCKETID: #endif case IPV6_BINDMULTI: switch (optname) { case IPV6_RECVHOPOPTS: optval = OPTBIT(IN6P_HOPOPTS); break; case IPV6_RECVDSTOPTS: optval = OPTBIT(IN6P_DSTOPTS); break; case IPV6_RECVRTHDRDSTOPTS: optval = OPTBIT(IN6P_RTHDRDSTOPTS); break; case IPV6_UNICAST_HOPS: optval = in6p->in6p_hops; break; case IPV6_RECVPKTINFO: optval = OPTBIT(IN6P_PKTINFO); break; case IPV6_RECVHOPLIMIT: optval = OPTBIT(IN6P_HOPLIMIT); break; case IPV6_RECVRTHDR: optval = OPTBIT(IN6P_RTHDR); break; case IPV6_RECVPATHMTU: optval = OPTBIT(IN6P_MTU); break; case IPV6_V6ONLY: optval = OPTBIT(IN6P_IPV6_V6ONLY); break; case IPV6_PORTRANGE: { int flags; flags = in6p->inp_flags; if (flags & INP_HIGHPORT) optval = IPV6_PORTRANGE_HIGH; else if (flags & INP_LOWPORT) optval = IPV6_PORTRANGE_LOW; else optval = 0; break; } case IPV6_RECVTCLASS: optval = OPTBIT(IN6P_TCLASS); break; case IPV6_AUTOFLOWLABEL: optval = OPTBIT(IN6P_AUTOFLOWLABEL); break; case IPV6_BINDANY: optval = OPTBIT(INP_BINDANY); break; case IPV6_FLOWID: optval = in6p->inp_flowid; break; case IPV6_FLOWTYPE: optval = in6p->inp_flowtype; break; case IPV6_RECVFLOWID: optval = OPTBIT2(INP_RECVFLOWID); break; #ifdef RSS case IPV6_RSSBUCKETID: retval = rss_hash2bucket(in6p->inp_flowid, in6p->inp_flowtype, &rss_bucket); if (retval == 0) optval = rss_bucket; else error = EINVAL; break; case IPV6_RECVRSSBUCKETID: optval = OPTBIT2(INP_RECVRSSBUCKETID); break; #endif case IPV6_BINDMULTI: optval = OPTBIT2(INP_BINDMULTI); break; } if (error) break; error = sooptcopyout(sopt, &optval, sizeof optval); break; case IPV6_PATHMTU: { u_long pmtu = 0; struct ip6_mtuinfo mtuinfo; if (!(so->so_state & SS_ISCONNECTED)) return (ENOTCONN); /* * XXX: we dot not consider the case of source * routing, or optional information to specify * the outgoing interface. */ error = ip6_getpmtu_ctl(so->so_fibnum, &in6p->in6p_faddr, &pmtu); if (error) break; if (pmtu > IPV6_MAXPACKET) pmtu = IPV6_MAXPACKET; bzero(&mtuinfo, sizeof(mtuinfo)); mtuinfo.ip6m_mtu = (u_int32_t)pmtu; optdata = (void *)&mtuinfo; optdatalen = sizeof(mtuinfo); error = sooptcopyout(sopt, optdata, optdatalen); break; } case IPV6_2292PKTINFO: case IPV6_2292HOPLIMIT: case IPV6_2292HOPOPTS: case IPV6_2292RTHDR: case IPV6_2292DSTOPTS: switch (optname) { case IPV6_2292PKTINFO: optval = OPTBIT(IN6P_PKTINFO); break; case IPV6_2292HOPLIMIT: optval = OPTBIT(IN6P_HOPLIMIT); break; case IPV6_2292HOPOPTS: optval = OPTBIT(IN6P_HOPOPTS); break; case IPV6_2292RTHDR: optval = OPTBIT(IN6P_RTHDR); break; case IPV6_2292DSTOPTS: optval = OPTBIT(IN6P_DSTOPTS|IN6P_RTHDRDSTOPTS); break; } error = sooptcopyout(sopt, &optval, sizeof optval); break; case IPV6_PKTINFO: case IPV6_HOPOPTS: case IPV6_RTHDR: case IPV6_DSTOPTS: case IPV6_RTHDRDSTOPTS: case IPV6_NEXTHOP: case IPV6_TCLASS: case IPV6_DONTFRAG: case IPV6_USE_MIN_MTU: case IPV6_PREFER_TEMPADDR: error = ip6_getpcbopt(in6p->in6p_outputopts, optname, sopt); break; case IPV6_MULTICAST_IF: case IPV6_MULTICAST_HOPS: case IPV6_MULTICAST_LOOP: case IPV6_MSFILTER: error = ip6_getmoptions(in6p, sopt); break; #ifdef IPSEC case IPV6_IPSEC_POLICY: { caddr_t req = NULL; size_t len = 0; struct mbuf *m = NULL; struct mbuf **mp = &m; size_t ovalsize = sopt->sopt_valsize; caddr_t oval = (caddr_t)sopt->sopt_val; error = soopt_getm(sopt, &m); /* XXX */ if (error != 0) break; error = soopt_mcopyin(sopt, m); /* XXX */ if (error != 0) break; sopt->sopt_valsize = ovalsize; sopt->sopt_val = oval; if (m) { req = mtod(m, caddr_t); len = m->m_len; } error = ipsec_get_policy(in6p, req, len, mp); if (error == 0) error = soopt_mcopyout(sopt, m); /* XXX */ if (error == 0 && m) m_freem(m); break; } #endif /* IPSEC */ default: error = ENOPROTOOPT; break; } break; } } return (error); } int ip6_raw_ctloutput(struct socket *so, struct sockopt *sopt) { int error = 0, optval, optlen; const int icmp6off = offsetof(struct icmp6_hdr, icmp6_cksum); struct inpcb *in6p = sotoinpcb(so); int level, op, optname; level = sopt->sopt_level; op = sopt->sopt_dir; optname = sopt->sopt_name; optlen = sopt->sopt_valsize; if (level != IPPROTO_IPV6) { return (EINVAL); } switch (optname) { case IPV6_CHECKSUM: /* * For ICMPv6 sockets, no modification allowed for checksum * offset, permit "no change" values to help existing apps. * * RFC3542 says: "An attempt to set IPV6_CHECKSUM * for an ICMPv6 socket will fail." * The current behavior does not meet RFC3542. */ switch (op) { case SOPT_SET: if (optlen != sizeof(int)) { error = EINVAL; break; } error = sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (error) break; if ((optval % 2) != 0) { /* the API assumes even offset values */ error = EINVAL; } else if (so->so_proto->pr_protocol == IPPROTO_ICMPV6) { if (optval != icmp6off) error = EINVAL; } else in6p->in6p_cksum = optval; break; case SOPT_GET: if (so->so_proto->pr_protocol == IPPROTO_ICMPV6) optval = icmp6off; else optval = in6p->in6p_cksum; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; default: error = EINVAL; break; } break; default: error = ENOPROTOOPT; break; } return (error); } /* * Set up IP6 options in pcb for insertion in output packets or * specifying behavior of outgoing packets. */ static int ip6_pcbopts(struct ip6_pktopts **pktopt, struct mbuf *m, struct socket *so, struct sockopt *sopt) { struct ip6_pktopts *opt = *pktopt; int error = 0; struct thread *td = sopt->sopt_td; /* turn off any old options. */ if (opt) { #ifdef DIAGNOSTIC if (opt->ip6po_pktinfo || opt->ip6po_nexthop || opt->ip6po_hbh || opt->ip6po_dest1 || opt->ip6po_dest2 || opt->ip6po_rhinfo.ip6po_rhi_rthdr) printf("ip6_pcbopts: all specified options are cleared.\n"); #endif ip6_clearpktopts(opt, -1); } else opt = malloc(sizeof(*opt), M_IP6OPT, M_WAITOK); *pktopt = NULL; if (!m || m->m_len == 0) { /* * Only turning off any previous options, regardless of * whether the opt is just created or given. */ free(opt, M_IP6OPT); return (0); } /* set options specified by user. */ if ((error = ip6_setpktopts(m, opt, NULL, (td != NULL) ? td->td_ucred : NULL, so->so_proto->pr_protocol)) != 0) { ip6_clearpktopts(opt, -1); /* XXX: discard all options */ free(opt, M_IP6OPT); return (error); } *pktopt = opt; return (0); } /* * initialize ip6_pktopts. beware that there are non-zero default values in * the struct. */ void ip6_initpktopts(struct ip6_pktopts *opt) { bzero(opt, sizeof(*opt)); opt->ip6po_hlim = -1; /* -1 means default hop limit */ opt->ip6po_tclass = -1; /* -1 means default traffic class */ opt->ip6po_minmtu = IP6PO_MINMTU_MCASTONLY; opt->ip6po_prefer_tempaddr = IP6PO_TEMPADDR_SYSTEM; } static int ip6_pcbopt(int optname, u_char *buf, int len, struct ip6_pktopts **pktopt, struct ucred *cred, int uproto) { struct ip6_pktopts *opt; if (*pktopt == NULL) { *pktopt = malloc(sizeof(struct ip6_pktopts), M_IP6OPT, M_WAITOK); ip6_initpktopts(*pktopt); } opt = *pktopt; return (ip6_setpktopt(optname, buf, len, opt, cred, 1, 0, uproto)); } static int ip6_getpcbopt(struct ip6_pktopts *pktopt, int optname, struct sockopt *sopt) { void *optdata = NULL; int optdatalen = 0; struct ip6_ext *ip6e; int error = 0; struct in6_pktinfo null_pktinfo; int deftclass = 0, on; int defminmtu = IP6PO_MINMTU_MCASTONLY; int defpreftemp = IP6PO_TEMPADDR_SYSTEM; switch (optname) { case IPV6_PKTINFO: optdata = (void *)&null_pktinfo; if (pktopt && pktopt->ip6po_pktinfo) { bcopy(pktopt->ip6po_pktinfo, &null_pktinfo, sizeof(null_pktinfo)); in6_clearscope(&null_pktinfo.ipi6_addr); } else { /* XXX: we don't have to do this every time... */ bzero(&null_pktinfo, sizeof(null_pktinfo)); } optdatalen = sizeof(struct in6_pktinfo); break; case IPV6_TCLASS: if (pktopt && pktopt->ip6po_tclass >= 0) optdata = (void *)&pktopt->ip6po_tclass; else optdata = (void *)&deftclass; optdatalen = sizeof(int); break; case IPV6_HOPOPTS: if (pktopt && pktopt->ip6po_hbh) { optdata = (void *)pktopt->ip6po_hbh; ip6e = (struct ip6_ext *)pktopt->ip6po_hbh; optdatalen = (ip6e->ip6e_len + 1) << 3; } break; case IPV6_RTHDR: if (pktopt && pktopt->ip6po_rthdr) { optdata = (void *)pktopt->ip6po_rthdr; ip6e = (struct ip6_ext *)pktopt->ip6po_rthdr; optdatalen = (ip6e->ip6e_len + 1) << 3; } break; case IPV6_RTHDRDSTOPTS: if (pktopt && pktopt->ip6po_dest1) { optdata = (void *)pktopt->ip6po_dest1; ip6e = (struct ip6_ext *)pktopt->ip6po_dest1; optdatalen = (ip6e->ip6e_len + 1) << 3; } break; case IPV6_DSTOPTS: if (pktopt && pktopt->ip6po_dest2) { optdata = (void *)pktopt->ip6po_dest2; ip6e = (struct ip6_ext *)pktopt->ip6po_dest2; optdatalen = (ip6e->ip6e_len + 1) << 3; } break; case IPV6_NEXTHOP: if (pktopt && pktopt->ip6po_nexthop) { optdata = (void *)pktopt->ip6po_nexthop; optdatalen = pktopt->ip6po_nexthop->sa_len; } break; case IPV6_USE_MIN_MTU: if (pktopt) optdata = (void *)&pktopt->ip6po_minmtu; else optdata = (void *)&defminmtu; optdatalen = sizeof(int); break; case IPV6_DONTFRAG: if (pktopt && ((pktopt->ip6po_flags) & IP6PO_DONTFRAG)) on = 1; else on = 0; optdata = (void *)&on; optdatalen = sizeof(on); break; case IPV6_PREFER_TEMPADDR: if (pktopt) optdata = (void *)&pktopt->ip6po_prefer_tempaddr; else optdata = (void *)&defpreftemp; optdatalen = sizeof(int); break; default: /* should not happen */ #ifdef DIAGNOSTIC panic("ip6_getpcbopt: unexpected option\n"); #endif return (ENOPROTOOPT); } error = sooptcopyout(sopt, optdata, optdatalen); return (error); } void ip6_clearpktopts(struct ip6_pktopts *pktopt, int optname) { if (pktopt == NULL) return; if (optname == -1 || optname == IPV6_PKTINFO) { if (pktopt->ip6po_pktinfo) free(pktopt->ip6po_pktinfo, M_IP6OPT); pktopt->ip6po_pktinfo = NULL; } if (optname == -1 || optname == IPV6_HOPLIMIT) pktopt->ip6po_hlim = -1; if (optname == -1 || optname == IPV6_TCLASS) pktopt->ip6po_tclass = -1; if (optname == -1 || optname == IPV6_NEXTHOP) { if (pktopt->ip6po_nextroute.ro_rt) { RTFREE(pktopt->ip6po_nextroute.ro_rt); pktopt->ip6po_nextroute.ro_rt = NULL; } if (pktopt->ip6po_nexthop) free(pktopt->ip6po_nexthop, M_IP6OPT); pktopt->ip6po_nexthop = NULL; } if (optname == -1 || optname == IPV6_HOPOPTS) { if (pktopt->ip6po_hbh) free(pktopt->ip6po_hbh, M_IP6OPT); pktopt->ip6po_hbh = NULL; } if (optname == -1 || optname == IPV6_RTHDRDSTOPTS) { if (pktopt->ip6po_dest1) free(pktopt->ip6po_dest1, M_IP6OPT); pktopt->ip6po_dest1 = NULL; } if (optname == -1 || optname == IPV6_RTHDR) { if (pktopt->ip6po_rhinfo.ip6po_rhi_rthdr) free(pktopt->ip6po_rhinfo.ip6po_rhi_rthdr, M_IP6OPT); pktopt->ip6po_rhinfo.ip6po_rhi_rthdr = NULL; if (pktopt->ip6po_route.ro_rt) { RTFREE(pktopt->ip6po_route.ro_rt); pktopt->ip6po_route.ro_rt = NULL; } } if (optname == -1 || optname == IPV6_DSTOPTS) { if (pktopt->ip6po_dest2) free(pktopt->ip6po_dest2, M_IP6OPT); pktopt->ip6po_dest2 = NULL; } } #define PKTOPT_EXTHDRCPY(type) \ do {\ if (src->type) {\ int hlen = (((struct ip6_ext *)src->type)->ip6e_len + 1) << 3;\ dst->type = malloc(hlen, M_IP6OPT, canwait);\ if (dst->type == NULL && canwait == M_NOWAIT)\ goto bad;\ bcopy(src->type, dst->type, hlen);\ }\ } while (/*CONSTCOND*/ 0) static int copypktopts(struct ip6_pktopts *dst, struct ip6_pktopts *src, int canwait) { if (dst == NULL || src == NULL) { printf("ip6_clearpktopts: invalid argument\n"); return (EINVAL); } dst->ip6po_hlim = src->ip6po_hlim; dst->ip6po_tclass = src->ip6po_tclass; dst->ip6po_flags = src->ip6po_flags; dst->ip6po_minmtu = src->ip6po_minmtu; dst->ip6po_prefer_tempaddr = src->ip6po_prefer_tempaddr; if (src->ip6po_pktinfo) { dst->ip6po_pktinfo = malloc(sizeof(*dst->ip6po_pktinfo), M_IP6OPT, canwait); if (dst->ip6po_pktinfo == NULL) goto bad; *dst->ip6po_pktinfo = *src->ip6po_pktinfo; } if (src->ip6po_nexthop) { dst->ip6po_nexthop = malloc(src->ip6po_nexthop->sa_len, M_IP6OPT, canwait); if (dst->ip6po_nexthop == NULL) goto bad; bcopy(src->ip6po_nexthop, dst->ip6po_nexthop, src->ip6po_nexthop->sa_len); } PKTOPT_EXTHDRCPY(ip6po_hbh); PKTOPT_EXTHDRCPY(ip6po_dest1); PKTOPT_EXTHDRCPY(ip6po_dest2); PKTOPT_EXTHDRCPY(ip6po_rthdr); /* not copy the cached route */ return (0); bad: ip6_clearpktopts(dst, -1); return (ENOBUFS); } #undef PKTOPT_EXTHDRCPY struct ip6_pktopts * ip6_copypktopts(struct ip6_pktopts *src, int canwait) { int error; struct ip6_pktopts *dst; dst = malloc(sizeof(*dst), M_IP6OPT, canwait); if (dst == NULL) return (NULL); ip6_initpktopts(dst); if ((error = copypktopts(dst, src, canwait)) != 0) { free(dst, M_IP6OPT); return (NULL); } return (dst); } void ip6_freepcbopts(struct ip6_pktopts *pktopt) { if (pktopt == NULL) return; ip6_clearpktopts(pktopt, -1); free(pktopt, M_IP6OPT); } /* * Set IPv6 outgoing packet options based on advanced API. */ int ip6_setpktopts(struct mbuf *control, struct ip6_pktopts *opt, struct ip6_pktopts *stickyopt, struct ucred *cred, int uproto) { struct cmsghdr *cm = 0; if (control == NULL || opt == NULL) return (EINVAL); ip6_initpktopts(opt); if (stickyopt) { int error; /* * If stickyopt is provided, make a local copy of the options * for this particular packet, then override them by ancillary * objects. * XXX: copypktopts() does not copy the cached route to a next * hop (if any). This is not very good in terms of efficiency, * but we can allow this since this option should be rarely * used. */ if ((error = copypktopts(opt, stickyopt, M_NOWAIT)) != 0) return (error); } /* * XXX: Currently, we assume all the optional information is stored * in a single mbuf. */ if (control->m_next) return (EINVAL); for (; control->m_len > 0; control->m_data += CMSG_ALIGN(cm->cmsg_len), control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { int error; if (control->m_len < CMSG_LEN(0)) return (EINVAL); cm = mtod(control, struct cmsghdr *); if (cm->cmsg_len == 0 || cm->cmsg_len > control->m_len) return (EINVAL); if (cm->cmsg_level != IPPROTO_IPV6) continue; error = ip6_setpktopt(cm->cmsg_type, CMSG_DATA(cm), cm->cmsg_len - CMSG_LEN(0), opt, cred, 0, 1, uproto); if (error) return (error); } return (0); } /* * Set a particular packet option, as a sticky option or an ancillary data * item. "len" can be 0 only when it's a sticky option. * We have 4 cases of combination of "sticky" and "cmsg": * "sticky=0, cmsg=0": impossible * "sticky=0, cmsg=1": RFC2292 or RFC3542 ancillary data * "sticky=1, cmsg=0": RFC3542 socket option * "sticky=1, cmsg=1": RFC2292 socket option */ static int ip6_setpktopt(int optname, u_char *buf, int len, struct ip6_pktopts *opt, struct ucred *cred, int sticky, int cmsg, int uproto) { int minmtupolicy, preftemp; int error; if (!sticky && !cmsg) { #ifdef DIAGNOSTIC printf("ip6_setpktopt: impossible case\n"); #endif return (EINVAL); } /* * IPV6_2292xxx is for backward compatibility to RFC2292, and should * not be specified in the context of RFC3542. Conversely, * RFC3542 types should not be specified in the context of RFC2292. */ if (!cmsg) { switch (optname) { case IPV6_2292PKTINFO: case IPV6_2292HOPLIMIT: case IPV6_2292NEXTHOP: case IPV6_2292HOPOPTS: case IPV6_2292DSTOPTS: case IPV6_2292RTHDR: case IPV6_2292PKTOPTIONS: return (ENOPROTOOPT); } } if (sticky && cmsg) { switch (optname) { case IPV6_PKTINFO: case IPV6_HOPLIMIT: case IPV6_NEXTHOP: case IPV6_HOPOPTS: case IPV6_DSTOPTS: case IPV6_RTHDRDSTOPTS: case IPV6_RTHDR: case IPV6_USE_MIN_MTU: case IPV6_DONTFRAG: case IPV6_TCLASS: case IPV6_PREFER_TEMPADDR: /* XXX: not an RFC3542 option */ return (ENOPROTOOPT); } } switch (optname) { case IPV6_2292PKTINFO: case IPV6_PKTINFO: { struct ifnet *ifp = NULL; struct in6_pktinfo *pktinfo; if (len != sizeof(struct in6_pktinfo)) return (EINVAL); pktinfo = (struct in6_pktinfo *)buf; /* * An application can clear any sticky IPV6_PKTINFO option by * doing a "regular" setsockopt with ipi6_addr being * in6addr_any and ipi6_ifindex being zero. * [RFC 3542, Section 6] */ if (optname == IPV6_PKTINFO && opt->ip6po_pktinfo && pktinfo->ipi6_ifindex == 0 && IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) { ip6_clearpktopts(opt, optname); break; } if (uproto == IPPROTO_TCP && optname == IPV6_PKTINFO && sticky && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) { return (EINVAL); } if (IN6_IS_ADDR_MULTICAST(&pktinfo->ipi6_addr)) return (EINVAL); /* validate the interface index if specified. */ if (pktinfo->ipi6_ifindex > V_if_index) return (ENXIO); if (pktinfo->ipi6_ifindex) { ifp = ifnet_byindex(pktinfo->ipi6_ifindex); if (ifp == NULL) return (ENXIO); } if (ifp != NULL && ( ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) return (ENETDOWN); if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) { struct in6_ifaddr *ia; in6_setscope(&pktinfo->ipi6_addr, ifp, NULL); ia = in6ifa_ifpwithaddr(ifp, &pktinfo->ipi6_addr); if (ia == NULL) return (EADDRNOTAVAIL); ifa_free(&ia->ia_ifa); } /* * We store the address anyway, and let in6_selectsrc() * validate the specified address. This is because ipi6_addr * may not have enough information about its scope zone, and * we may need additional information (such as outgoing * interface or the scope zone of a destination address) to * disambiguate the scope. * XXX: the delay of the validation may confuse the * application when it is used as a sticky option. */ if (opt->ip6po_pktinfo == NULL) { opt->ip6po_pktinfo = malloc(sizeof(*pktinfo), M_IP6OPT, M_NOWAIT); if (opt->ip6po_pktinfo == NULL) return (ENOBUFS); } bcopy(pktinfo, opt->ip6po_pktinfo, sizeof(*pktinfo)); break; } case IPV6_2292HOPLIMIT: case IPV6_HOPLIMIT: { int *hlimp; /* * RFC 3542 deprecated the usage of sticky IPV6_HOPLIMIT * to simplify the ordering among hoplimit options. */ if (optname == IPV6_HOPLIMIT && sticky) return (ENOPROTOOPT); if (len != sizeof(int)) return (EINVAL); hlimp = (int *)buf; if (*hlimp < -1 || *hlimp > 255) return (EINVAL); opt->ip6po_hlim = *hlimp; break; } case IPV6_TCLASS: { int tclass; if (len != sizeof(int)) return (EINVAL); tclass = *(int *)buf; if (tclass < -1 || tclass > 255) return (EINVAL); opt->ip6po_tclass = tclass; break; } case IPV6_2292NEXTHOP: case IPV6_NEXTHOP: if (cred != NULL) { error = priv_check_cred(cred, PRIV_NETINET_SETHDROPTS, 0); if (error) return (error); } if (len == 0) { /* just remove the option */ ip6_clearpktopts(opt, IPV6_NEXTHOP); break; } /* check if cmsg_len is large enough for sa_len */ if (len < sizeof(struct sockaddr) || len < *buf) return (EINVAL); switch (((struct sockaddr *)buf)->sa_family) { case AF_INET6: { struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)buf; int error; if (sa6->sin6_len != sizeof(struct sockaddr_in6)) return (EINVAL); if (IN6_IS_ADDR_UNSPECIFIED(&sa6->sin6_addr) || IN6_IS_ADDR_MULTICAST(&sa6->sin6_addr)) { return (EINVAL); } if ((error = sa6_embedscope(sa6, V_ip6_use_defzone)) != 0) { return (error); } break; } case AF_LINK: /* should eventually be supported */ default: return (EAFNOSUPPORT); } /* turn off the previous option, then set the new option. */ ip6_clearpktopts(opt, IPV6_NEXTHOP); opt->ip6po_nexthop = malloc(*buf, M_IP6OPT, M_NOWAIT); if (opt->ip6po_nexthop == NULL) return (ENOBUFS); bcopy(buf, opt->ip6po_nexthop, *buf); break; case IPV6_2292HOPOPTS: case IPV6_HOPOPTS: { struct ip6_hbh *hbh; int hbhlen; /* * XXX: We don't allow a non-privileged user to set ANY HbH * options, since per-option restriction has too much * overhead. */ if (cred != NULL) { error = priv_check_cred(cred, PRIV_NETINET_SETHDROPTS, 0); if (error) return (error); } if (len == 0) { ip6_clearpktopts(opt, IPV6_HOPOPTS); break; /* just remove the option */ } /* message length validation */ if (len < sizeof(struct ip6_hbh)) return (EINVAL); hbh = (struct ip6_hbh *)buf; hbhlen = (hbh->ip6h_len + 1) << 3; if (len != hbhlen) return (EINVAL); /* turn off the previous option, then set the new option. */ ip6_clearpktopts(opt, IPV6_HOPOPTS); opt->ip6po_hbh = malloc(hbhlen, M_IP6OPT, M_NOWAIT); if (opt->ip6po_hbh == NULL) return (ENOBUFS); bcopy(hbh, opt->ip6po_hbh, hbhlen); break; } case IPV6_2292DSTOPTS: case IPV6_DSTOPTS: case IPV6_RTHDRDSTOPTS: { struct ip6_dest *dest, **newdest = NULL; int destlen; if (cred != NULL) { /* XXX: see the comment for IPV6_HOPOPTS */ error = priv_check_cred(cred, PRIV_NETINET_SETHDROPTS, 0); if (error) return (error); } if (len == 0) { ip6_clearpktopts(opt, optname); break; /* just remove the option */ } /* message length validation */ if (len < sizeof(struct ip6_dest)) return (EINVAL); dest = (struct ip6_dest *)buf; destlen = (dest->ip6d_len + 1) << 3; if (len != destlen) return (EINVAL); /* * Determine the position that the destination options header * should be inserted; before or after the routing header. */ switch (optname) { case IPV6_2292DSTOPTS: /* * The old advacned API is ambiguous on this point. * Our approach is to determine the position based * according to the existence of a routing header. * Note, however, that this depends on the order of the * extension headers in the ancillary data; the 1st * part of the destination options header must appear * before the routing header in the ancillary data, * too. * RFC3542 solved the ambiguity by introducing * separate ancillary data or option types. */ if (opt->ip6po_rthdr == NULL) newdest = &opt->ip6po_dest1; else newdest = &opt->ip6po_dest2; break; case IPV6_RTHDRDSTOPTS: newdest = &opt->ip6po_dest1; break; case IPV6_DSTOPTS: newdest = &opt->ip6po_dest2; break; } /* turn off the previous option, then set the new option. */ ip6_clearpktopts(opt, optname); *newdest = malloc(destlen, M_IP6OPT, M_NOWAIT); if (*newdest == NULL) return (ENOBUFS); bcopy(dest, *newdest, destlen); break; } case IPV6_2292RTHDR: case IPV6_RTHDR: { struct ip6_rthdr *rth; int rthlen; if (len == 0) { ip6_clearpktopts(opt, IPV6_RTHDR); break; /* just remove the option */ } /* message length validation */ if (len < sizeof(struct ip6_rthdr)) return (EINVAL); rth = (struct ip6_rthdr *)buf; rthlen = (rth->ip6r_len + 1) << 3; if (len != rthlen) return (EINVAL); switch (rth->ip6r_type) { case IPV6_RTHDR_TYPE_0: if (rth->ip6r_len == 0) /* must contain one addr */ return (EINVAL); if (rth->ip6r_len % 2) /* length must be even */ return (EINVAL); if (rth->ip6r_len / 2 != rth->ip6r_segleft) return (EINVAL); break; default: return (EINVAL); /* not supported */ } /* turn off the previous option */ ip6_clearpktopts(opt, IPV6_RTHDR); opt->ip6po_rthdr = malloc(rthlen, M_IP6OPT, M_NOWAIT); if (opt->ip6po_rthdr == NULL) return (ENOBUFS); bcopy(rth, opt->ip6po_rthdr, rthlen); break; } case IPV6_USE_MIN_MTU: if (len != sizeof(int)) return (EINVAL); minmtupolicy = *(int *)buf; if (minmtupolicy != IP6PO_MINMTU_MCASTONLY && minmtupolicy != IP6PO_MINMTU_DISABLE && minmtupolicy != IP6PO_MINMTU_ALL) { return (EINVAL); } opt->ip6po_minmtu = minmtupolicy; break; case IPV6_DONTFRAG: if (len != sizeof(int)) return (EINVAL); if (uproto == IPPROTO_TCP || *(int *)buf == 0) { /* * we ignore this option for TCP sockets. * (RFC3542 leaves this case unspecified.) */ opt->ip6po_flags &= ~IP6PO_DONTFRAG; } else opt->ip6po_flags |= IP6PO_DONTFRAG; break; case IPV6_PREFER_TEMPADDR: if (len != sizeof(int)) return (EINVAL); preftemp = *(int *)buf; if (preftemp != IP6PO_TEMPADDR_SYSTEM && preftemp != IP6PO_TEMPADDR_NOTPREFER && preftemp != IP6PO_TEMPADDR_PREFER) { return (EINVAL); } opt->ip6po_prefer_tempaddr = preftemp; break; default: return (ENOPROTOOPT); } /* end of switch */ return (0); } /* * Routine called from ip6_output() to loop back a copy of an IP6 multicast * packet to the input queue of a specified interface. Note that this * calls the output routine of the loopback "driver", but with an interface * pointer that might NOT be &loif -- easier than replicating that code here. */ void -ip6_mloopback(struct ifnet *ifp, const struct mbuf *m) +ip6_mloopback(struct ifnet *ifp, struct mbuf *m) { struct mbuf *copym; struct ip6_hdr *ip6; copym = m_copy(m, 0, M_COPYALL); if (copym == NULL) return; /* * Make sure to deep-copy IPv6 header portion in case the data * is in an mbuf cluster, so that we can safely override the IPv6 * header portion later. */ if (!M_WRITABLE(copym) || copym->m_len < sizeof(struct ip6_hdr)) { copym = m_pullup(copym, sizeof(struct ip6_hdr)); if (copym == NULL) return; } ip6 = mtod(copym, struct ip6_hdr *); /* * clear embedded scope identifiers if necessary. * in6_clearscope will touch the addresses only when necessary. */ in6_clearscope(&ip6->ip6_src); in6_clearscope(&ip6->ip6_dst); if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) { copym->m_pkthdr.csum_flags |= CSUM_DATA_VALID_IPV6 | CSUM_PSEUDO_HDR; copym->m_pkthdr.csum_data = 0xffff; } if_simloop(ifp, copym, AF_INET6, 0); } /* * Chop IPv6 header off from the payload. */ static int ip6_splithdr(struct mbuf *m, struct ip6_exthdrs *exthdrs) { struct mbuf *mh; struct ip6_hdr *ip6; ip6 = mtod(m, struct ip6_hdr *); if (m->m_len > sizeof(*ip6)) { mh = m_gethdr(M_NOWAIT, MT_DATA); if (mh == NULL) { m_freem(m); return ENOBUFS; } m_move_pkthdr(mh, m); M_ALIGN(mh, sizeof(*ip6)); m->m_len -= sizeof(*ip6); m->m_data += sizeof(*ip6); mh->m_next = m; m = mh; m->m_len = sizeof(*ip6); bcopy((caddr_t)ip6, mtod(m, caddr_t), sizeof(*ip6)); } exthdrs->ip6e_ip6 = m; return 0; } /* * Compute IPv6 extension header length. */ int ip6_optlen(struct inpcb *in6p) { int len; if (!in6p->in6p_outputopts) return 0; len = 0; #define elen(x) \ (((struct ip6_ext *)(x)) ? (((struct ip6_ext *)(x))->ip6e_len + 1) << 3 : 0) len += elen(in6p->in6p_outputopts->ip6po_hbh); if (in6p->in6p_outputopts->ip6po_rthdr) /* dest1 is valid with rthdr only */ len += elen(in6p->in6p_outputopts->ip6po_dest1); len += elen(in6p->in6p_outputopts->ip6po_rthdr); len += elen(in6p->in6p_outputopts->ip6po_dest2); return len; #undef elen } Index: head/sys/netinet6/ip6_var.h =================================================================== --- head/sys/netinet6/ip6_var.h (revision 296241) +++ head/sys/netinet6/ip6_var.h (revision 296242) @@ -1,436 +1,436 @@ /*- * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * 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 project 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 PROJECT 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 PROJECT 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. * * $KAME: ip6_var.h,v 1.62 2001/05/03 14:51:48 itojun Exp $ */ /*- * 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. * 4. 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. * * @(#)ip_var.h 8.1 (Berkeley) 6/10/93 * $FreeBSD$ */ #ifndef _NETINET6_IP6_VAR_H_ #define _NETINET6_IP6_VAR_H_ /* * IP6 reassembly queue structure. Each fragment * being reassembled is attached to one of these structures. */ struct ip6q { struct ip6asfrag *ip6q_down; struct ip6asfrag *ip6q_up; u_int32_t ip6q_ident; u_int8_t ip6q_nxt; u_int8_t ip6q_ecn; u_int8_t ip6q_ttl; struct in6_addr ip6q_src, ip6q_dst; struct ip6q *ip6q_next; struct ip6q *ip6q_prev; int ip6q_unfrglen; /* len of unfragmentable part */ #ifdef notyet u_char *ip6q_nxtp; #endif int ip6q_nfrag; /* # of fragments */ struct label *ip6q_label; }; struct ip6asfrag { struct ip6asfrag *ip6af_down; struct ip6asfrag *ip6af_up; struct mbuf *ip6af_m; int ip6af_offset; /* offset in ip6af_m to next header */ int ip6af_frglen; /* fragmentable part length */ int ip6af_off; /* fragment offset */ u_int16_t ip6af_mff; /* more fragment bit in frag off */ }; #define IP6_REASS_MBUF(ip6af) (*(struct mbuf **)&((ip6af)->ip6af_m)) /* * IP6 reinjecting structure. */ struct ip6_direct_ctx { uint32_t ip6dc_nxt; /* next header to process */ uint32_t ip6dc_off; /* offset to next header */ }; /* * Structure attached to inpcb.in6p_moptions and * passed to ip6_output when IPv6 multicast options are in use. * This structure is lazy-allocated. */ struct ip6_moptions { struct ifnet *im6o_multicast_ifp; /* ifp for outgoing multicasts */ u_char im6o_multicast_hlim; /* hoplimit for outgoing multicasts */ u_char im6o_multicast_loop; /* 1 >= hear sends if a member */ u_short im6o_num_memberships; /* no. memberships this socket */ u_short im6o_max_memberships; /* max memberships this socket */ struct in6_multi **im6o_membership; /* group memberships */ struct in6_mfilter *im6o_mfilters; /* source filters */ }; /* * Control options for outgoing packets */ /* Routing header related info */ struct ip6po_rhinfo { struct ip6_rthdr *ip6po_rhi_rthdr; /* Routing header */ struct route_in6 ip6po_rhi_route; /* Route to the 1st hop */ }; #define ip6po_rthdr ip6po_rhinfo.ip6po_rhi_rthdr #define ip6po_route ip6po_rhinfo.ip6po_rhi_route /* Nexthop related info */ struct ip6po_nhinfo { struct sockaddr *ip6po_nhi_nexthop; struct route_in6 ip6po_nhi_route; /* Route to the nexthop */ }; #define ip6po_nexthop ip6po_nhinfo.ip6po_nhi_nexthop #define ip6po_nextroute ip6po_nhinfo.ip6po_nhi_route struct ip6_pktopts { struct mbuf *ip6po_m; /* Pointer to mbuf storing the data */ int ip6po_hlim; /* Hoplimit for outgoing packets */ /* Outgoing IF/address information */ struct in6_pktinfo *ip6po_pktinfo; /* Next-hop address information */ struct ip6po_nhinfo ip6po_nhinfo; struct ip6_hbh *ip6po_hbh; /* Hop-by-Hop options header */ /* Destination options header (before a routing header) */ struct ip6_dest *ip6po_dest1; /* Routing header related info. */ struct ip6po_rhinfo ip6po_rhinfo; /* Destination options header (after a routing header) */ struct ip6_dest *ip6po_dest2; int ip6po_tclass; /* traffic class */ int ip6po_minmtu; /* fragment vs PMTU discovery policy */ #define IP6PO_MINMTU_MCASTONLY -1 /* default; send at min MTU for multicast*/ #define IP6PO_MINMTU_DISABLE 0 /* always perform pmtu disc */ #define IP6PO_MINMTU_ALL 1 /* always send at min MTU */ int ip6po_prefer_tempaddr; /* whether temporary addresses are preferred as source address */ #define IP6PO_TEMPADDR_SYSTEM -1 /* follow the system default */ #define IP6PO_TEMPADDR_NOTPREFER 0 /* not prefer temporary address */ #define IP6PO_TEMPADDR_PREFER 1 /* prefer temporary address */ int ip6po_flags; #if 0 /* parameters in this block is obsolete. do not reuse the values. */ #define IP6PO_REACHCONF 0x01 /* upper-layer reachability confirmation. */ #define IP6PO_MINMTU 0x02 /* use minimum MTU (IPV6_USE_MIN_MTU) */ #endif #define IP6PO_DONTFRAG 0x04 /* disable fragmentation (IPV6_DONTFRAG) */ #define IP6PO_USECOA 0x08 /* use care of address */ }; /* * Control options for incoming packets */ struct ip6stat { uint64_t ip6s_total; /* total packets received */ uint64_t ip6s_tooshort; /* packet too short */ uint64_t ip6s_toosmall; /* not enough data */ uint64_t ip6s_fragments; /* fragments received */ uint64_t ip6s_fragdropped; /* frags dropped(dups, out of space) */ uint64_t ip6s_fragtimeout; /* fragments timed out */ uint64_t ip6s_fragoverflow; /* fragments that exceeded limit */ uint64_t ip6s_forward; /* packets forwarded */ uint64_t ip6s_cantforward; /* packets rcvd for unreachable dest */ uint64_t ip6s_redirectsent; /* packets forwarded on same net */ uint64_t ip6s_delivered; /* datagrams delivered to upper level*/ uint64_t ip6s_localout; /* total ip packets generated here */ uint64_t ip6s_odropped; /* lost packets due to nobufs, etc. */ uint64_t ip6s_reassembled; /* total packets reassembled ok */ uint64_t ip6s_fragmented; /* datagrams successfully fragmented */ uint64_t ip6s_ofragments; /* output fragments created */ uint64_t ip6s_cantfrag; /* don't fragment flag was set, etc. */ uint64_t ip6s_badoptions; /* error in option processing */ uint64_t ip6s_noroute; /* packets discarded due to no route */ uint64_t ip6s_badvers; /* ip6 version != 6 */ uint64_t ip6s_rawout; /* total raw ip packets generated */ uint64_t ip6s_badscope; /* scope error */ uint64_t ip6s_notmember; /* don't join this multicast group */ #define IP6S_HDRCNT 256 /* headers count */ uint64_t ip6s_nxthist[IP6S_HDRCNT]; /* next header history */ uint64_t ip6s_m1; /* one mbuf */ #define IP6S_M2MMAX 32 uint64_t ip6s_m2m[IP6S_M2MMAX]; /* two or more mbuf */ uint64_t ip6s_mext1; /* one ext mbuf */ uint64_t ip6s_mext2m; /* two or more ext mbuf */ uint64_t ip6s_exthdrtoolong; /* ext hdr are not contiguous */ uint64_t ip6s_nogif; /* no match gif found */ uint64_t ip6s_toomanyhdr; /* discarded due to too many headers */ /* * statistics for improvement of the source address selection * algorithm: * XXX: hardcoded 16 = # of ip6 multicast scope types + 1 */ #define IP6S_RULESMAX 16 #define IP6S_SCOPECNT 16 /* number of times that address selection fails */ uint64_t ip6s_sources_none; /* number of times that an address on the outgoing I/F is chosen */ uint64_t ip6s_sources_sameif[IP6S_SCOPECNT]; /* number of times that an address on a non-outgoing I/F is chosen */ uint64_t ip6s_sources_otherif[IP6S_SCOPECNT]; /* * number of times that an address that has the same scope * from the destination is chosen. */ uint64_t ip6s_sources_samescope[IP6S_SCOPECNT]; /* * number of times that an address that has a different scope * from the destination is chosen. */ uint64_t ip6s_sources_otherscope[IP6S_SCOPECNT]; /* number of times that a deprecated address is chosen */ uint64_t ip6s_sources_deprecated[IP6S_SCOPECNT]; /* number of times that each rule of source selection is applied. */ uint64_t ip6s_sources_rule[IP6S_RULESMAX]; }; #ifdef _KERNEL #include VNET_PCPUSTAT_DECLARE(struct ip6stat, ip6stat); #define IP6STAT_ADD(name, val) \ VNET_PCPUSTAT_ADD(struct ip6stat, ip6stat, name, (val)) #define IP6STAT_SUB(name, val) IP6STAT_ADD(name, -(val)) #define IP6STAT_INC(name) IP6STAT_ADD(name, 1) #define IP6STAT_DEC(name) IP6STAT_SUB(name, 1) #endif #ifdef _KERNEL /* flags passed to ip6_output as last parameter */ #define IPV6_UNSPECSRC 0x01 /* allow :: as the source address */ #define IPV6_FORWARDING 0x02 /* most of IPv6 header exists */ #define IPV6_MINMTU 0x04 /* use minimum MTU (IPV6_USE_MIN_MTU) */ #ifdef __NO_STRICT_ALIGNMENT #define IP6_HDR_ALIGNED_P(ip) 1 #else #define IP6_HDR_ALIGNED_P(ip) ((((intptr_t) (ip)) & 3) == 0) #endif VNET_DECLARE(int, ip6_defhlim); /* default hop limit */ VNET_DECLARE(int, ip6_defmcasthlim); /* default multicast hop limit */ VNET_DECLARE(int, ip6_forwarding); /* act as router? */ VNET_DECLARE(int, ip6_use_deprecated); /* allow deprecated addr as source */ VNET_DECLARE(int, ip6_rr_prune); /* router renumbering prefix * walk list every 5 sec. */ VNET_DECLARE(int, ip6_mcast_pmtu); /* enable pMTU discovery for multicast? */ VNET_DECLARE(int, ip6_v6only); #define V_ip6_defhlim VNET(ip6_defhlim) #define V_ip6_defmcasthlim VNET(ip6_defmcasthlim) #define V_ip6_forwarding VNET(ip6_forwarding) #define V_ip6_use_deprecated VNET(ip6_use_deprecated) #define V_ip6_rr_prune VNET(ip6_rr_prune) #define V_ip6_mcast_pmtu VNET(ip6_mcast_pmtu) #define V_ip6_v6only VNET(ip6_v6only) VNET_DECLARE(struct socket *, ip6_mrouter); /* multicast routing daemon */ VNET_DECLARE(int, ip6_sendredirects); /* send IP redirects when forwarding? */ VNET_DECLARE(int, ip6_maxfragpackets); /* Maximum packets in reassembly * queue */ VNET_DECLARE(int, ip6_maxfrags); /* Maximum fragments in reassembly * queue */ VNET_DECLARE(int, ip6_accept_rtadv); /* Acts as a host not a router */ VNET_DECLARE(int, ip6_no_radr); /* No defroute from RA */ VNET_DECLARE(int, ip6_norbit_raif); /* Disable R-bit in NA on RA * receiving IF. */ VNET_DECLARE(int, ip6_rfc6204w3); /* Accept defroute from RA even when forwarding enabled */ VNET_DECLARE(int, ip6_log_interval); VNET_DECLARE(time_t, ip6_log_time); VNET_DECLARE(int, ip6_hdrnestlimit); /* upper limit of # of extension * headers */ VNET_DECLARE(int, ip6_dad_count); /* DupAddrDetectionTransmits */ #define V_ip6_mrouter VNET(ip6_mrouter) #define V_ip6_sendredirects VNET(ip6_sendredirects) #define V_ip6_maxfragpackets VNET(ip6_maxfragpackets) #define V_ip6_maxfrags VNET(ip6_maxfrags) #define V_ip6_accept_rtadv VNET(ip6_accept_rtadv) #define V_ip6_no_radr VNET(ip6_no_radr) #define V_ip6_norbit_raif VNET(ip6_norbit_raif) #define V_ip6_rfc6204w3 VNET(ip6_rfc6204w3) #define V_ip6_log_interval VNET(ip6_log_interval) #define V_ip6_log_time VNET(ip6_log_time) #define V_ip6_hdrnestlimit VNET(ip6_hdrnestlimit) #define V_ip6_dad_count VNET(ip6_dad_count) VNET_DECLARE(int, ip6_auto_flowlabel); VNET_DECLARE(int, ip6_auto_linklocal); #define V_ip6_auto_flowlabel VNET(ip6_auto_flowlabel) #define V_ip6_auto_linklocal VNET(ip6_auto_linklocal) VNET_DECLARE(int, ip6_use_tempaddr); /* Whether to use temporary addresses */ VNET_DECLARE(int, ip6_prefer_tempaddr); /* Whether to prefer temporary * addresses in the source address * selection */ #define V_ip6_use_tempaddr VNET(ip6_use_tempaddr) #define V_ip6_prefer_tempaddr VNET(ip6_prefer_tempaddr) VNET_DECLARE(int, ip6_use_defzone); /* Whether to use the default scope * zone when unspecified */ #define V_ip6_use_defzone VNET(ip6_use_defzone) VNET_DECLARE (struct pfil_head, inet6_pfil_hook); /* packet filter hooks */ #define V_inet6_pfil_hook VNET(inet6_pfil_hook) #ifdef IPSTEALTH VNET_DECLARE(int, ip6stealth); #define V_ip6stealth VNET(ip6stealth) #endif extern struct pr_usrreqs rip6_usrreqs; struct sockopt; struct inpcb; int icmp6_ctloutput(struct socket *, struct sockopt *sopt); struct in6_ifaddr; void ip6_init(void); #ifdef VIMAGE void ip6_destroy(void); #endif int ip6proto_register(short); int ip6proto_unregister(short); void ip6_input(struct mbuf *); void ip6_direct_input(struct mbuf *); void ip6_freepcbopts(struct ip6_pktopts *); int ip6_unknown_opt(u_int8_t *, struct mbuf *, int); char * ip6_get_prevhdr(const struct mbuf *, int); int ip6_nexthdr(const struct mbuf *, int, int, int *); int ip6_lasthdr(const struct mbuf *, int, int, int *); extern int (*ip6_mforward)(struct ip6_hdr *, struct ifnet *, struct mbuf *); int ip6_process_hopopts(struct mbuf *, u_int8_t *, int, u_int32_t *, u_int32_t *); struct mbuf **ip6_savecontrol_v4(struct inpcb *, struct mbuf *, struct mbuf **, int *); void ip6_savecontrol(struct inpcb *, struct mbuf *, struct mbuf **); void ip6_notify_pmtu(struct inpcb *, struct sockaddr_in6 *, u_int32_t); int ip6_sysctl(int *, u_int, void *, size_t *, void *, size_t); void ip6_forward(struct mbuf *, int); -void ip6_mloopback(struct ifnet *, const struct mbuf *); +void ip6_mloopback(struct ifnet *, struct mbuf *); int ip6_output(struct mbuf *, struct ip6_pktopts *, struct route_in6 *, int, struct ip6_moptions *, struct ifnet **, struct inpcb *); int ip6_ctloutput(struct socket *, struct sockopt *); int ip6_raw_ctloutput(struct socket *, struct sockopt *); void ip6_initpktopts(struct ip6_pktopts *); int ip6_setpktopts(struct mbuf *, struct ip6_pktopts *, struct ip6_pktopts *, struct ucred *, int); void ip6_clearpktopts(struct ip6_pktopts *, int); struct ip6_pktopts *ip6_copypktopts(struct ip6_pktopts *, int); int ip6_optlen(struct inpcb *); int ip6_deletefraghdr(struct mbuf *, int, int); int ip6_fragment(struct ifnet *, struct mbuf *, int, u_char, int, uint32_t); int route6_input(struct mbuf **, int *, int); void frag6_init(void); int frag6_input(struct mbuf **, int *, int); void frag6_slowtimo(void); void frag6_drain(void); void rip6_init(void); int rip6_input(struct mbuf **, int *, int); void rip6_ctlinput(int, struct sockaddr *, void *); int rip6_ctloutput(struct socket *, struct sockopt *); int rip6_output(struct mbuf *, struct socket *, ...); int rip6_usrreq(struct socket *, int, struct mbuf *, struct mbuf *, struct mbuf *, struct thread *); int dest6_input(struct mbuf **, int *, int); int none_input(struct mbuf **, int *, int); int in6_selectsrc_socket(struct sockaddr_in6 *, struct ip6_pktopts *, struct inpcb *, struct ucred *, int, struct in6_addr *, int *); int in6_selectsrc_addr(uint32_t, const struct in6_addr *, uint32_t, struct ifnet *, struct in6_addr *, int *); int in6_selectroute(struct sockaddr_in6 *, struct ip6_pktopts *, struct ip6_moptions *, struct route_in6 *, struct ifnet **, struct rtentry **); int in6_selectroute_fib(struct sockaddr_in6 *, struct ip6_pktopts *, struct ip6_moptions *, struct route_in6 *, struct ifnet **, struct rtentry **, u_int); u_int32_t ip6_randomid(void); u_int32_t ip6_randomflowlabel(void); void in6_delayed_cksum(struct mbuf *m, uint32_t plen, u_short offset); #endif /* _KERNEL */ #endif /* !_NETINET6_IP6_VAR_H_ */ Index: head/sys/sys/mbuf.h =================================================================== --- head/sys/sys/mbuf.h (revision 296241) +++ head/sys/sys/mbuf.h (revision 296242) @@ -1,1253 +1,1263 @@ /*- * 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 #ifdef WITNESS #include #endif #endif /* * 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) #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 *); }; /* * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set. * Size ILP32: 48 * LP64: 56 * Compile-time assertions in uipc_mbuf.c test these values to ensure that * they are correct. */ struct pkthdr { struct ifnet *rcvif; /* rcv interface */ 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 */ uint64_t csum_flags; /* checksum and offload features */ uint16_t fibnum; /* this packet should use this fib */ uint8_t cosqos; /* class/quality of service */ uint8_t rsstype; /* hash type */ uint8_t l2hlen; /* layer 2 header length */ uint8_t l3hlen; /* layer 3 header length */ uint8_t l4hlen; /* layer 4 header length */ uint8_t l5hlen; /* layer 5 header length */ 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 { uint8_t eight[8]; uint16_t sixteen[4]; uint32_t thirtytwo[2]; uint64_t sixtyfour[1]; uintptr_t unintptr[1]; void *ptr; } PH_loc; }; #define ether_vtag PH_per.sixteen[0] #define PH_vt PH_per #define vt_nrecs sixteen[0] #define tso_segsz PH_per.sixteen[1] #define csum_phsum PH_per.sixteen[2] #define csum_data PH_per.thirtytwo[1] /* * 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. */ struct m_ext { - volatile u_int *ext_cnt; /* pointer to ref count info */ + union { + volatile u_int ext_count; /* value of ref count info */ + volatile u_int *ext_cnt; /* pointer to ref count info */ + }; caddr_t ext_buf; /* start of buffer */ 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 */ void (*ext_free) /* free routine if not the usual */ (struct mbuf *, void *, void *); void *ext_arg1; /* optional argument pointer */ void *ext_arg2; /* optional argument pointer */ }; /* * 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 { struct pkthdr m_pkthdr; /* M_PKTHDR set */ union { struct m_ext m_ext; /* M_EXT set */ char m_pktdat[0]; }; }; char m_dat[0]; /* !M_PKTHDR, !M_EXT */ }; }; /* * mbuf flags of global significance and layer crossing. * Those of only protocol/layer specific significance are to be mapped * to M_PROTO[1-12] 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_UNUSED_8 0x00000100 /* --available-- */ #define M_NOFREE 0x00000200 /* do not free mbuf, embedded in cluster */ #define M_PROTO1 0x00001000 /* protocol-specific */ #define M_PROTO2 0x00002000 /* protocol-specific */ #define M_PROTO3 0x00004000 /* protocol-specific */ #define M_PROTO4 0x00008000 /* protocol-specific */ #define M_PROTO5 0x00010000 /* protocol-specific */ #define M_PROTO6 0x00020000 /* protocol-specific */ #define M_PROTO7 0x00040000 /* protocol-specific */ #define M_PROTO8 0x00080000 /* protocol-specific */ #define M_PROTO9 0x00100000 /* protocol-specific */ #define M_PROTO10 0x00200000 /* protocol-specific */ #define M_PROTO11 0x00400000 /* protocol-specific */ #define M_PROTO12 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|M_PROTO12) /* * Flags preserved when copying m_pkthdr. */ #define M_COPYFLAGS \ (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG| \ M_PROTOFLAGS) /* * 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" #define M_FLAG_PROTOBITS \ "\15M_PROTO1\16M_PROTO2\17M_PROTO3\20M_PROTO4\21M_PROTO5" \ "\22M_PROTO6\23M_PROTO7\24M_PROTO8\25M_PROTO9\26M_PROTO10" \ "\27M_PROTO11\30M_PROTO12" #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 can be set by non-RSS cards or configurations * that provide an opaque flow identifier, allowing for ordering and * distribution without explicit affinity. */ /* Microsoft RSS standard hash types */ #define M_HASHTYPE_NONE 0 #define M_HASHTYPE_RSS_IPV4 1 /* IPv4 2-tuple */ #define M_HASHTYPE_RSS_TCP_IPV4 2 /* TCPv4 4-tuple */ #define M_HASHTYPE_RSS_IPV6 3 /* IPv6 2-tuple */ #define M_HASHTYPE_RSS_TCP_IPV6 4 /* TCPv6 4-tuple */ #define M_HASHTYPE_RSS_IPV6_EX 5 /* IPv6 2-tuple + ext hdrs */ #define M_HASHTYPE_RSS_TCP_IPV6_EX 6 /* TCPv6 4-tiple + ext hdrs */ /* Non-standard RSS hash types */ #define M_HASHTYPE_RSS_UDP_IPV4 7 /* IPv4 UDP 4-tuple */ #define M_HASHTYPE_RSS_UDP_IPV4_EX 8 /* IPv4 UDP 4-tuple + ext hdrs */ #define M_HASHTYPE_RSS_UDP_IPV6 9 /* IPv6 UDP 4-tuple */ #define M_HASHTYPE_RSS_UDP_IPV6_EX 10 /* IPv6 UDP 4-tuple + ext hdrs */ #define M_HASHTYPE_OPAQUE 255 /* ordering, not affinity */ #define M_HASHTYPE_CLEAR(m) ((m)->m_pkthdr.rsstype = 0) #define M_HASHTYPE_GET(m) ((m)->m_pkthdr.rsstype) #define M_HASHTYPE_SET(m, v) ((m)->m_pkthdr.rsstype = (v)) #define M_HASHTYPE_TEST(m, v) (M_HASHTYPE_GET(m) == (v)) /* * COS/QOS class and quality of service tags. * It uses DSCP code points as base. */ #define QOS_DSCP_CS0 0x00 #define QOS_DSCP_DEF QOS_DSCP_CS0 #define QOS_DSCP_CS1 0x20 #define QOS_DSCP_AF11 0x28 #define QOS_DSCP_AF12 0x30 #define QOS_DSCP_AF13 0x38 #define QOS_DSCP_CS2 0x40 #define QOS_DSCP_AF21 0x48 #define QOS_DSCP_AF22 0x50 #define QOS_DSCP_AF23 0x58 #define QOS_DSCP_CS3 0x60 #define QOS_DSCP_AF31 0x68 #define QOS_DSCP_AF32 0x70 #define QOS_DSCP_AF33 0x78 #define QOS_DSCP_CS4 0x80 #define QOS_DSCP_AF41 0x88 #define QOS_DSCP_AF42 0x90 #define QOS_DSCP_AF43 0x98 #define QOS_DSCP_CS5 0xa0 #define QOS_DSCP_EF 0xb8 #define QOS_DSCP_CS6 0xc0 #define QOS_DSCP_CS7 0xe0 /* * 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 (M_IOVEC) */ #define EXT_SFBUF_NOCACHE 8 /* sendfile(2)'s sf_buf not to be cached */ #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_cnt, notyet */ +#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 /* for vendor-internal use */ #define EXT_FLAG_VENDOR2 0x020000 /* for vendor-internal use */ #define EXT_FLAG_VENDOR3 0x040000 /* for vendor-internal use */ #define EXT_FLAG_VENDOR4 0x080000 /* for vendor-internal use */ #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" /* * External reference/free functions. */ -void sf_ext_ref(void *, void *); void sf_ext_free(void *, void *); void sf_ext_free_nocache(void *, void *); /* * 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. */ #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_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 */ /* Inbound checksum support where the checksum was verified by hardware. */ #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_COALESED 0x40000000 /* contains merged segments */ /* * 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" \ "\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP\15CSUM_IP6_TSO" \ "\16CSUM_IP6_ISCSI" \ "\31CSUM_L3_CALC\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID" \ "\35CSUM_L5_CALC\36CSUM_L5_VALID\37CSUM_COALESED" /* 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_UDP_IPV6 CSUM_IP6_UDP #define CSUM_TCP_IPV6 CSUM_IP6_TCP #define CSUM_SCTP_IPV6 CSUM_IP6_SCTP /* * 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_OOBDATA 15 /* expedited data */ #define MT_NTYPES 16 /* number of mbuf types for mbtypes[] */ #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" #ifdef _KERNEL #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) #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_ext_refcnt; -void mb_dupcl(struct mbuf *, const struct mbuf *); +void mb_dupcl(struct mbuf *, struct mbuf *); void mb_free_ext(struct mbuf *); void m_adj(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(const struct mbuf *, int, int, int); +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)); struct mbuf *m_dup(const struct mbuf *, int); int m_dup_pkthdr(struct mbuf *, const struct mbuf *, int); -int m_extadd(struct mbuf *, caddr_t, u_int, +void m_extadd(struct mbuf *, caddr_t, u_int, void (*)(struct mbuf *, void *, void *), void *, void *, - int, int, int); + int, int); u_int m_fixhdr(struct mbuf *); struct mbuf *m_fragment(struct mbuf *, int, int); void m_freem(struct mbuf *); struct mbuf *m_get2(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 *, 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); struct mbuf *m_unshare(struct mbuf *, int); static __inline int m_gettype(int size) { int type; switch (size) { case MSIZE: type = EXT_MBUF; break; case MCLBYTES: type = EXT_CLUSTER; break; #if MJUMPAGESIZE != MCLBYTES case MJUMPAGESIZE: type = EXT_JUMBOP; break; #endif 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, caddr_t buf, u_int size, u_int *ref_cnt, void (*freef)(struct mbuf *, void *, void *), 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; #if MJUMPAGESIZE != MCLBYTES case MJUMPAGESIZE: zone = zone_jumbop; break; #endif 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) { if ((error = m_pkthdr_init(m, how)) != 0) return (error); } return (0); } static __inline struct mbuf * m_get(int how, short type) { struct mb_args args; args.flags = 0; args.type = type; return (uma_zalloc_arg(zone_mbuf, &args, how)); } static __inline struct mbuf * m_gethdr(int how, short type) { struct mb_args args; args.flags = M_PKTHDR; args.type = type; return (uma_zalloc_arg(zone_mbuf, &args, how)); } static __inline struct mbuf * m_getcl(int how, short type, int flags) { struct mb_args args; args.flags = flags; args.type = type; return (uma_zalloc_arg(zone_pack, &args, how)); } +/* + * 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) { - uma_zone_t zone; int size; switch (type) { case EXT_CLUSTER: size = MCLBYTES; - zone = zone_clust; break; #if MJUMPAGESIZE != MCLBYTES case EXT_JUMBOP: size = MJUMPAGESIZE; - zone = zone_jumbop; break; #endif case EXT_JUMBO9: size = MJUM9BYTES; - zone = zone_jumbo9; break; case EXT_JUMBO16: size = MJUM16BYTES; - zone = zone_jumbo16; 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 = 0; - m->m_ext.ext_cnt = uma_find_refcnt(zone, cl); + m->m_ext.ext_flags = EXT_FLAG_EMBREF; + m->m_ext.ext_count = 1; m->m_flags |= M_EXT; - } 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) \ - (void )m_extadd((m), (caddr_t)(buf), (size), (free), (arg1), (arg2),\ - (flags), (type), M_NOWAIT) + m_extadd((m), (caddr_t)(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)->m_flags & M_EXT)) || \ - (*((m)->m_ext.ext_cnt) == 1)) ) \ + (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__)) /* * 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__)) /* * 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_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)) /* Length to m_copy to copy all. */ #define M_COPYALL 1000000000 /* Compatibility with 4.3. */ #define m_copy(m, o, l) m_copym((m), (o), (l), M_NOWAIT) 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 int nmbclusters; /* Maximum number of clusters */ /*- * 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 | MTAG_PERSISTENT) /* PF/ALTQ information */ #define 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(u_int32_t, int, 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 *, u_int32_t, int, 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, u_int32_t cookie, int 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(int 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, int 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 mbuf * m_free(struct mbuf *m) { struct mbuf *n = m->m_next; if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE)) m_tag_delete_chain(m, NULL); 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_ether_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 *, const 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_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++; } #endif /* _KERNEL */ #endif /* !_SYS_MBUF_H_ */