Index: head/sys/dev/netmap/netmap_freebsd.c =================================================================== --- head/sys/dev/netmap/netmap_freebsd.c (revision 348021) +++ head/sys/dev/netmap/netmap_freebsd.c (revision 348022) @@ -1,1618 +1,1619 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (C) 2013-2014 Universita` di Pisa. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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. */ /* $FreeBSD$ */ #include "opt_inet.h" #include "opt_inet6.h" #include #include #include +#include #include #include /* POLLIN, POLLOUT */ #include /* types used in module initialization */ #include /* DEV_MODULE_ORDERED */ #include #include /* kern_ioctl() */ #include #include /* vtophys */ #include /* vtophys */ #include #include #include #include #include #include #include /* sockaddrs */ #include #include /* kthread_add() */ #include /* PROC_LOCK() */ #include /* RFNOWAIT */ #include /* sched_bind() */ #include /* mp_maxid */ #include /* taskqueue_enqueue(), taskqueue_create(), ... */ #include #include #include /* IFT_ETHER */ #include /* ether_ifdetach */ #include /* LLADDR */ #include /* bus_dmamap_* */ #include /* in6_cksum_pseudo() */ #include /* in_pseudo(), in_cksum_hdr() */ #include #include #include #include /* ======================== FREEBSD-SPECIFIC ROUTINES ================== */ static void nm_kqueue_notify(void *opaque, int pending) { struct nm_selinfo *si = opaque; /* We use a non-zero hint to distinguish this notification call * from the call done in kqueue_scan(), which uses hint=0. */ KNOTE_UNLOCKED(&si->si.si_note, /*hint=*/0x100); } int nm_os_selinfo_init(NM_SELINFO_T *si, const char *name) { int err; TASK_INIT(&si->ntfytask, 0, nm_kqueue_notify, si); si->ntfytq = taskqueue_create(name, M_NOWAIT, taskqueue_thread_enqueue, &si->ntfytq); if (si->ntfytq == NULL) return -ENOMEM; err = taskqueue_start_threads(&si->ntfytq, 1, PI_NET, "tq %s", name); if (err) { taskqueue_free(si->ntfytq); si->ntfytq = NULL; return err; } snprintf(si->mtxname, sizeof(si->mtxname), "nmkl%s", name); mtx_init(&si->m, si->mtxname, NULL, MTX_DEF); knlist_init_mtx(&si->si.si_note, &si->m); si->kqueue_users = 0; return (0); } void nm_os_selinfo_uninit(NM_SELINFO_T *si) { if (si->ntfytq == NULL) { return; /* si was not initialized */ } taskqueue_drain(si->ntfytq, &si->ntfytask); taskqueue_free(si->ntfytq); si->ntfytq = NULL; knlist_delete(&si->si.si_note, curthread, /*islocked=*/0); knlist_destroy(&si->si.si_note); /* now we don't need the mutex anymore */ mtx_destroy(&si->m); } void * nm_os_malloc(size_t size) { return malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO); } void * nm_os_realloc(void *addr, size_t new_size, size_t old_size __unused) { return realloc(addr, new_size, M_DEVBUF, M_NOWAIT | M_ZERO); } void nm_os_free(void *addr) { free(addr, M_DEVBUF); } void nm_os_ifnet_lock(void) { IFNET_RLOCK(); } void nm_os_ifnet_unlock(void) { IFNET_RUNLOCK(); } static int netmap_use_count = 0; void nm_os_get_module(void) { netmap_use_count++; } void nm_os_put_module(void) { netmap_use_count--; } static void netmap_ifnet_arrival_handler(void *arg __unused, struct ifnet *ifp) { netmap_undo_zombie(ifp); } static void netmap_ifnet_departure_handler(void *arg __unused, struct ifnet *ifp) { netmap_make_zombie(ifp); } static eventhandler_tag nm_ifnet_ah_tag; static eventhandler_tag nm_ifnet_dh_tag; int nm_os_ifnet_init(void) { nm_ifnet_ah_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event, netmap_ifnet_arrival_handler, NULL, EVENTHANDLER_PRI_ANY); nm_ifnet_dh_tag = EVENTHANDLER_REGISTER(ifnet_departure_event, netmap_ifnet_departure_handler, NULL, EVENTHANDLER_PRI_ANY); return 0; } void nm_os_ifnet_fini(void) { EVENTHANDLER_DEREGISTER(ifnet_arrival_event, nm_ifnet_ah_tag); EVENTHANDLER_DEREGISTER(ifnet_departure_event, nm_ifnet_dh_tag); } unsigned nm_os_ifnet_mtu(struct ifnet *ifp) { #if __FreeBSD_version < 1100030 return ifp->if_data.ifi_mtu; #else /* __FreeBSD_version >= 1100030 */ return ifp->if_mtu; #endif } rawsum_t nm_os_csum_raw(uint8_t *data, size_t len, rawsum_t cur_sum) { /* TODO XXX please use the FreeBSD implementation for this. */ uint16_t *words = (uint16_t *)data; int nw = len / 2; int i; for (i = 0; i < nw; i++) cur_sum += be16toh(words[i]); if (len & 1) cur_sum += (data[len-1] << 8); return cur_sum; } /* Fold a raw checksum: 'cur_sum' is in host byte order, while the * return value is in network byte order. */ uint16_t nm_os_csum_fold(rawsum_t cur_sum) { /* TODO XXX please use the FreeBSD implementation for this. */ while (cur_sum >> 16) cur_sum = (cur_sum & 0xFFFF) + (cur_sum >> 16); return htobe16((~cur_sum) & 0xFFFF); } uint16_t nm_os_csum_ipv4(struct nm_iphdr *iph) { #if 0 return in_cksum_hdr((void *)iph); #else return nm_os_csum_fold(nm_os_csum_raw((uint8_t*)iph, sizeof(struct nm_iphdr), 0)); #endif } void nm_os_csum_tcpudp_ipv4(struct nm_iphdr *iph, void *data, size_t datalen, uint16_t *check) { #ifdef INET uint16_t pseudolen = datalen + iph->protocol; /* Compute and insert the pseudo-header cheksum. */ *check = in_pseudo(iph->saddr, iph->daddr, htobe16(pseudolen)); /* Compute the checksum on TCP/UDP header + payload * (includes the pseudo-header). */ *check = nm_os_csum_fold(nm_os_csum_raw(data, datalen, 0)); #else static int notsupported = 0; if (!notsupported) { notsupported = 1; nm_prerr("inet4 segmentation not supported"); } #endif } void nm_os_csum_tcpudp_ipv6(struct nm_ipv6hdr *ip6h, void *data, size_t datalen, uint16_t *check) { #ifdef INET6 *check = in6_cksum_pseudo((void*)ip6h, datalen, ip6h->nexthdr, 0); *check = nm_os_csum_fold(nm_os_csum_raw(data, datalen, 0)); #else static int notsupported = 0; if (!notsupported) { notsupported = 1; nm_prerr("inet6 segmentation not supported"); } #endif } /* on FreeBSD we send up one packet at a time */ void * nm_os_send_up(struct ifnet *ifp, struct mbuf *m, struct mbuf *prev) { NA(ifp)->if_input(ifp, m); return NULL; } int nm_os_mbuf_has_csum_offld(struct mbuf *m) { return m->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP | CSUM_SCTP | CSUM_TCP_IPV6 | CSUM_UDP_IPV6 | CSUM_SCTP_IPV6); } int nm_os_mbuf_has_seg_offld(struct mbuf *m) { return m->m_pkthdr.csum_flags & CSUM_TSO; } static void freebsd_generic_rx_handler(struct ifnet *ifp, struct mbuf *m) { int stolen; if (unlikely(!NM_NA_VALID(ifp))) { nm_prlim(1, "Warning: RX packet intercepted, but no" " emulated adapter"); return; } stolen = generic_rx_handler(ifp, m); if (!stolen) { struct netmap_generic_adapter *gna = (struct netmap_generic_adapter *)NA(ifp); gna->save_if_input(ifp, m); } } /* * Intercept the rx routine in the standard device driver. * Second argument is non-zero to intercept, 0 to restore */ int nm_os_catch_rx(struct netmap_generic_adapter *gna, int intercept) { struct netmap_adapter *na = &gna->up.up; struct ifnet *ifp = na->ifp; int ret = 0; nm_os_ifnet_lock(); if (intercept) { if (gna->save_if_input) { nm_prerr("RX on %s already intercepted", na->name); ret = EBUSY; /* already set */ goto out; } gna->save_if_input = ifp->if_input; ifp->if_input = freebsd_generic_rx_handler; } else { if (!gna->save_if_input) { nm_prerr("Failed to undo RX intercept on %s", na->name); ret = EINVAL; /* not saved */ goto out; } ifp->if_input = gna->save_if_input; gna->save_if_input = NULL; } out: nm_os_ifnet_unlock(); return ret; } /* * Intercept the packet steering routine in the tx path, * so that we can decide which queue is used for an mbuf. * Second argument is non-zero to intercept, 0 to restore. * On freebsd we just intercept if_transmit. */ int nm_os_catch_tx(struct netmap_generic_adapter *gna, int intercept) { struct netmap_adapter *na = &gna->up.up; struct ifnet *ifp = netmap_generic_getifp(gna); nm_os_ifnet_lock(); if (intercept) { na->if_transmit = ifp->if_transmit; ifp->if_transmit = netmap_transmit; } else { ifp->if_transmit = na->if_transmit; } nm_os_ifnet_unlock(); return 0; } /* * Transmit routine used by generic_netmap_txsync(). Returns 0 on success * and non-zero on error (which may be packet drops or other errors). * addr and len identify the netmap buffer, m is the (preallocated) * mbuf to use for transmissions. * * We should add a reference to the mbuf so the m_freem() at the end * of the transmission does not consume resources. * * On FreeBSD, and on multiqueue cards, we can force the queue using * if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) * i = m->m_pkthdr.flowid % adapter->num_queues; * else * i = curcpu % adapter->num_queues; * */ int nm_os_generic_xmit_frame(struct nm_os_gen_arg *a) { int ret; u_int len = a->len; struct ifnet *ifp = a->ifp; struct mbuf *m = a->m; #if __FreeBSD_version < 1100000 /* * Old FreeBSD versions. The mbuf has a cluster attached, * we need to copy from the cluster to the netmap buffer. */ if (MBUF_REFCNT(m) != 1) { nm_prerr("invalid refcnt %d for %p", MBUF_REFCNT(m), m); panic("in generic_xmit_frame"); } if (m->m_ext.ext_size < len) { nm_prlim(2, "size %d < len %d", m->m_ext.ext_size, len); len = m->m_ext.ext_size; } bcopy(a->addr, m->m_data, len); #else /* __FreeBSD_version >= 1100000 */ /* New FreeBSD versions. Link the external storage to * the netmap buffer, so that no copy is necessary. */ m->m_ext.ext_buf = m->m_data = a->addr; m->m_ext.ext_size = len; #endif /* __FreeBSD_version >= 1100000 */ m->m_len = m->m_pkthdr.len = len; /* mbuf refcnt is not contended, no need to use atomic * (a memory barrier is enough). */ SET_MBUF_REFCNT(m, 2); M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); m->m_pkthdr.flowid = a->ring_nr; m->m_pkthdr.rcvif = ifp; /* used for tx notification */ ret = NA(ifp)->if_transmit(ifp, m); return ret ? -1 : 0; } #if __FreeBSD_version >= 1100005 struct netmap_adapter * netmap_getna(if_t ifp) { return (NA((struct ifnet *)ifp)); } #endif /* __FreeBSD_version >= 1100005 */ /* * The following two functions are empty until we have a generic * way to extract the info from the ifp */ int nm_os_generic_find_num_desc(struct ifnet *ifp, unsigned int *tx, unsigned int *rx) { return 0; } void nm_os_generic_find_num_queues(struct ifnet *ifp, u_int *txq, u_int *rxq) { unsigned num_rings = netmap_generic_rings ? netmap_generic_rings : 1; *txq = num_rings; *rxq = num_rings; } void nm_os_generic_set_features(struct netmap_generic_adapter *gna) { gna->rxsg = 1; /* Supported through m_copydata. */ gna->txqdisc = 0; /* Not supported. */ } void nm_os_mitigation_init(struct nm_generic_mit *mit, int idx, struct netmap_adapter *na) { mit->mit_pending = 0; mit->mit_ring_idx = idx; mit->mit_na = na; } void nm_os_mitigation_start(struct nm_generic_mit *mit) { } void nm_os_mitigation_restart(struct nm_generic_mit *mit) { } int nm_os_mitigation_active(struct nm_generic_mit *mit) { return 0; } void nm_os_mitigation_cleanup(struct nm_generic_mit *mit) { } static int nm_vi_dummy(struct ifnet *ifp, u_long cmd, caddr_t addr) { return EINVAL; } static void nm_vi_start(struct ifnet *ifp) { panic("nm_vi_start() must not be called"); } /* * Index manager of persistent virtual interfaces. * It is used to decide the lowest byte of the MAC address. * We use the same algorithm with management of bridge port index. */ #define NM_VI_MAX 255 static struct { uint8_t index[NM_VI_MAX]; /* XXX just for a reasonable number */ uint8_t active; struct mtx lock; } nm_vi_indices; void nm_os_vi_init_index(void) { int i; for (i = 0; i < NM_VI_MAX; i++) nm_vi_indices.index[i] = i; nm_vi_indices.active = 0; mtx_init(&nm_vi_indices.lock, "nm_vi_indices_lock", NULL, MTX_DEF); } /* return -1 if no index available */ static int nm_vi_get_index(void) { int ret; mtx_lock(&nm_vi_indices.lock); ret = nm_vi_indices.active == NM_VI_MAX ? -1 : nm_vi_indices.index[nm_vi_indices.active++]; mtx_unlock(&nm_vi_indices.lock); return ret; } static void nm_vi_free_index(uint8_t val) { int i, lim; mtx_lock(&nm_vi_indices.lock); lim = nm_vi_indices.active; for (i = 0; i < lim; i++) { if (nm_vi_indices.index[i] == val) { /* swap index[lim-1] and j */ int tmp = nm_vi_indices.index[lim-1]; nm_vi_indices.index[lim-1] = val; nm_vi_indices.index[i] = tmp; nm_vi_indices.active--; break; } } if (lim == nm_vi_indices.active) nm_prerr("Index %u not found", val); mtx_unlock(&nm_vi_indices.lock); } #undef NM_VI_MAX /* * Implementation of a netmap-capable virtual interface that * registered to the system. * It is based on if_tap.c and ip_fw_log.c in FreeBSD 9. * * Note: Linux sets refcount to 0 on allocation of net_device, * then increments it on registration to the system. * FreeBSD sets refcount to 1 on if_alloc(), and does not * increment this refcount on if_attach(). */ int nm_os_vi_persist(const char *name, struct ifnet **ret) { struct ifnet *ifp; u_short macaddr_hi; uint32_t macaddr_mid; u_char eaddr[6]; int unit = nm_vi_get_index(); /* just to decide MAC address */ if (unit < 0) return EBUSY; /* * We use the same MAC address generation method with tap * except for the highest octet is 00:be instead of 00:bd */ macaddr_hi = htons(0x00be); /* XXX tap + 1 */ macaddr_mid = (uint32_t) ticks; bcopy(&macaddr_hi, eaddr, sizeof(short)); bcopy(&macaddr_mid, &eaddr[2], sizeof(uint32_t)); eaddr[5] = (uint8_t)unit; ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { nm_prerr("if_alloc failed"); return ENOMEM; } if_initname(ifp, name, IF_DUNIT_NONE); ifp->if_mtu = 65536; ifp->if_flags = IFF_UP | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_init = (void *)nm_vi_dummy; ifp->if_ioctl = nm_vi_dummy; ifp->if_start = nm_vi_start; ifp->if_mtu = ETHERMTU; IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); ifp->if_capabilities |= IFCAP_LINKSTATE; ifp->if_capenable |= IFCAP_LINKSTATE; ether_ifattach(ifp, eaddr); *ret = ifp; return 0; } /* unregister from the system and drop the final refcount */ void nm_os_vi_detach(struct ifnet *ifp) { nm_vi_free_index(((char *)IF_LLADDR(ifp))[5]); ether_ifdetach(ifp); if_free(ifp); } #ifdef WITH_EXTMEM #include #include struct nm_os_extmem { vm_object_t obj; vm_offset_t kva; vm_offset_t size; uintptr_t scan; }; void nm_os_extmem_delete(struct nm_os_extmem *e) { nm_prinf("freeing %zx bytes", (size_t)e->size); vm_map_remove(kernel_map, e->kva, e->kva + e->size); nm_os_free(e); } char * nm_os_extmem_nextpage(struct nm_os_extmem *e) { char *rv = NULL; if (e->scan < e->kva + e->size) { rv = (char *)e->scan; e->scan += PAGE_SIZE; } return rv; } int nm_os_extmem_isequal(struct nm_os_extmem *e1, struct nm_os_extmem *e2) { return (e1->obj == e2->obj); } int nm_os_extmem_nr_pages(struct nm_os_extmem *e) { return e->size >> PAGE_SHIFT; } struct nm_os_extmem * nm_os_extmem_create(unsigned long p, struct nmreq_pools_info *pi, int *perror) { vm_map_t map; vm_map_entry_t entry; vm_object_t obj; vm_prot_t prot; vm_pindex_t index; boolean_t wired; struct nm_os_extmem *e = NULL; int rv, error = 0; e = nm_os_malloc(sizeof(*e)); if (e == NULL) { error = ENOMEM; goto out; } map = &curthread->td_proc->p_vmspace->vm_map; rv = vm_map_lookup(&map, p, VM_PROT_RW, &entry, &obj, &index, &prot, &wired); if (rv != KERN_SUCCESS) { nm_prerr("address %lx not found", p); goto out_free; } /* check that we are given the whole vm_object ? */ vm_map_lookup_done(map, entry); // XXX can we really use obj after releasing the map lock? e->obj = obj; vm_object_reference(obj); /* wire the memory and add the vm_object to the kernel map, * to make sure that it is not fred even if the processes that * are mmap()ing it all exit */ e->kva = vm_map_min(kernel_map); e->size = obj->size << PAGE_SHIFT; rv = vm_map_find(kernel_map, obj, 0, &e->kva, e->size, 0, VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE, VM_PROT_READ | VM_PROT_WRITE, 0); if (rv != KERN_SUCCESS) { nm_prerr("vm_map_find(%zx) failed", (size_t)e->size); goto out_rel; } rv = vm_map_wire(kernel_map, e->kva, e->kva + e->size, VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); if (rv != KERN_SUCCESS) { nm_prerr("vm_map_wire failed"); goto out_rem; } e->scan = e->kva; return e; out_rem: vm_map_remove(kernel_map, e->kva, e->kva + e->size); e->obj = NULL; out_rel: vm_object_deallocate(e->obj); out_free: nm_os_free(e); out: if (perror) *perror = error; return NULL; } #endif /* WITH_EXTMEM */ /* ================== PTNETMAP GUEST SUPPORT ==================== */ #ifdef WITH_PTNETMAP #include #include #include /* bus_dmamap_* */ #include #include #include /* * ptnetmap memory device (memdev) for freebsd guest, * ssed to expose host netmap memory to the guest through a PCI BAR. */ /* * ptnetmap memdev private data structure */ struct ptnetmap_memdev { device_t dev; struct resource *pci_io; struct resource *pci_mem; struct netmap_mem_d *nm_mem; }; static int ptn_memdev_probe(device_t); static int ptn_memdev_attach(device_t); static int ptn_memdev_detach(device_t); static int ptn_memdev_shutdown(device_t); static device_method_t ptn_memdev_methods[] = { DEVMETHOD(device_probe, ptn_memdev_probe), DEVMETHOD(device_attach, ptn_memdev_attach), DEVMETHOD(device_detach, ptn_memdev_detach), DEVMETHOD(device_shutdown, ptn_memdev_shutdown), DEVMETHOD_END }; static driver_t ptn_memdev_driver = { PTNETMAP_MEMDEV_NAME, ptn_memdev_methods, sizeof(struct ptnetmap_memdev), }; /* We use (SI_ORDER_MIDDLE+1) here, see DEV_MODULE_ORDERED() invocation * below. */ static devclass_t ptnetmap_devclass; DRIVER_MODULE_ORDERED(ptn_memdev, pci, ptn_memdev_driver, ptnetmap_devclass, NULL, NULL, SI_ORDER_MIDDLE + 1); /* * Map host netmap memory through PCI-BAR in the guest OS, * returning physical (nm_paddr) and virtual (nm_addr) addresses * of the netmap memory mapped in the guest. */ int nm_os_pt_memdev_iomap(struct ptnetmap_memdev *ptn_dev, vm_paddr_t *nm_paddr, void **nm_addr, uint64_t *mem_size) { int rid; nm_prinf("ptn_memdev_driver iomap"); rid = PCIR_BAR(PTNETMAP_MEM_PCI_BAR); *mem_size = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMSIZE_HI); *mem_size = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMSIZE_LO) | (*mem_size << 32); /* map memory allocator */ ptn_dev->pci_mem = bus_alloc_resource(ptn_dev->dev, SYS_RES_MEMORY, &rid, 0, ~0, *mem_size, RF_ACTIVE); if (ptn_dev->pci_mem == NULL) { *nm_paddr = 0; *nm_addr = NULL; return ENOMEM; } *nm_paddr = rman_get_start(ptn_dev->pci_mem); *nm_addr = rman_get_virtual(ptn_dev->pci_mem); nm_prinf("=== BAR %d start %lx len %lx mem_size %lx ===", PTNETMAP_MEM_PCI_BAR, (unsigned long)(*nm_paddr), (unsigned long)rman_get_size(ptn_dev->pci_mem), (unsigned long)*mem_size); return (0); } uint32_t nm_os_pt_memdev_ioread(struct ptnetmap_memdev *ptn_dev, unsigned int reg) { return bus_read_4(ptn_dev->pci_io, reg); } /* Unmap host netmap memory. */ void nm_os_pt_memdev_iounmap(struct ptnetmap_memdev *ptn_dev) { nm_prinf("ptn_memdev_driver iounmap"); if (ptn_dev->pci_mem) { bus_release_resource(ptn_dev->dev, SYS_RES_MEMORY, PCIR_BAR(PTNETMAP_MEM_PCI_BAR), ptn_dev->pci_mem); ptn_dev->pci_mem = NULL; } } /* Device identification routine, return BUS_PROBE_DEFAULT on success, * positive on failure */ static int ptn_memdev_probe(device_t dev) { char desc[256]; if (pci_get_vendor(dev) != PTNETMAP_PCI_VENDOR_ID) return (ENXIO); if (pci_get_device(dev) != PTNETMAP_PCI_DEVICE_ID) return (ENXIO); snprintf(desc, sizeof(desc), "%s PCI adapter", PTNETMAP_MEMDEV_NAME); device_set_desc_copy(dev, desc); return (BUS_PROBE_DEFAULT); } /* Device initialization routine. */ static int ptn_memdev_attach(device_t dev) { struct ptnetmap_memdev *ptn_dev; int rid; uint16_t mem_id; ptn_dev = device_get_softc(dev); ptn_dev->dev = dev; pci_enable_busmaster(dev); rid = PCIR_BAR(PTNETMAP_IO_PCI_BAR); ptn_dev->pci_io = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid, RF_ACTIVE); if (ptn_dev->pci_io == NULL) { device_printf(dev, "cannot map I/O space\n"); return (ENXIO); } mem_id = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMID); /* create guest allocator */ ptn_dev->nm_mem = netmap_mem_pt_guest_attach(ptn_dev, mem_id); if (ptn_dev->nm_mem == NULL) { ptn_memdev_detach(dev); return (ENOMEM); } netmap_mem_get(ptn_dev->nm_mem); nm_prinf("ptnetmap memdev attached, host memid: %u", mem_id); return (0); } /* Device removal routine. */ static int ptn_memdev_detach(device_t dev) { struct ptnetmap_memdev *ptn_dev; ptn_dev = device_get_softc(dev); if (ptn_dev->nm_mem) { nm_prinf("ptnetmap memdev detached, host memid %u", netmap_mem_get_id(ptn_dev->nm_mem)); netmap_mem_put(ptn_dev->nm_mem); ptn_dev->nm_mem = NULL; } if (ptn_dev->pci_mem) { bus_release_resource(dev, SYS_RES_MEMORY, PCIR_BAR(PTNETMAP_MEM_PCI_BAR), ptn_dev->pci_mem); ptn_dev->pci_mem = NULL; } if (ptn_dev->pci_io) { bus_release_resource(dev, SYS_RES_IOPORT, PCIR_BAR(PTNETMAP_IO_PCI_BAR), ptn_dev->pci_io); ptn_dev->pci_io = NULL; } return (0); } static int ptn_memdev_shutdown(device_t dev) { return bus_generic_shutdown(dev); } #endif /* WITH_PTNETMAP */ /* * In order to track whether pages are still mapped, we hook into * the standard cdev_pager and intercept the constructor and * destructor. */ struct netmap_vm_handle_t { struct cdev *dev; struct netmap_priv_d *priv; }; static int netmap_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred, u_short *color) { struct netmap_vm_handle_t *vmh = handle; if (netmap_verbose) nm_prinf("handle %p size %jd prot %d foff %jd", handle, (intmax_t)size, prot, (intmax_t)foff); if (color) *color = 0; dev_ref(vmh->dev); return 0; } static void netmap_dev_pager_dtor(void *handle) { struct netmap_vm_handle_t *vmh = handle; struct cdev *dev = vmh->dev; struct netmap_priv_d *priv = vmh->priv; if (netmap_verbose) nm_prinf("handle %p", handle); netmap_dtor(priv); free(vmh, M_DEVBUF); dev_rel(dev); } static int netmap_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot, vm_page_t *mres) { struct netmap_vm_handle_t *vmh = object->handle; struct netmap_priv_d *priv = vmh->priv; struct netmap_adapter *na = priv->np_na; vm_paddr_t paddr; vm_page_t page; vm_memattr_t memattr; vm_pindex_t pidx; nm_prdis("object %p offset %jd prot %d mres %p", object, (intmax_t)offset, prot, mres); memattr = object->memattr; pidx = OFF_TO_IDX(offset); paddr = netmap_mem_ofstophys(na->nm_mem, offset); if (paddr == 0) return VM_PAGER_FAIL; if (((*mres)->flags & PG_FICTITIOUS) != 0) { /* * If the passed in result page is a fake page, update it with * the new physical address. */ page = *mres; vm_page_updatefake(page, paddr, memattr); } else { /* * Replace the passed in reqpage page with our own fake page and * free up the all of the original pages. */ #ifndef VM_OBJECT_WUNLOCK /* FreeBSD < 10.x */ #define VM_OBJECT_WUNLOCK VM_OBJECT_UNLOCK #define VM_OBJECT_WLOCK VM_OBJECT_LOCK #endif /* VM_OBJECT_WUNLOCK */ VM_OBJECT_WUNLOCK(object); page = vm_page_getfake(paddr, memattr); VM_OBJECT_WLOCK(object); vm_page_lock(*mres); vm_page_free(*mres); vm_page_unlock(*mres); *mres = page; vm_page_insert(page, object, pidx); } page->valid = VM_PAGE_BITS_ALL; return (VM_PAGER_OK); } static struct cdev_pager_ops netmap_cdev_pager_ops = { .cdev_pg_ctor = netmap_dev_pager_ctor, .cdev_pg_dtor = netmap_dev_pager_dtor, .cdev_pg_fault = netmap_dev_pager_fault, }; static int netmap_mmap_single(struct cdev *cdev, vm_ooffset_t *foff, vm_size_t objsize, vm_object_t *objp, int prot) { int error; struct netmap_vm_handle_t *vmh; struct netmap_priv_d *priv; vm_object_t obj; if (netmap_verbose) nm_prinf("cdev %p foff %jd size %jd objp %p prot %d", cdev, (intmax_t )*foff, (intmax_t )objsize, objp, prot); vmh = malloc(sizeof(struct netmap_vm_handle_t), M_DEVBUF, M_NOWAIT | M_ZERO); if (vmh == NULL) return ENOMEM; vmh->dev = cdev; NMG_LOCK(); error = devfs_get_cdevpriv((void**)&priv); if (error) goto err_unlock; if (priv->np_nifp == NULL) { error = EINVAL; goto err_unlock; } vmh->priv = priv; priv->np_refs++; NMG_UNLOCK(); obj = cdev_pager_allocate(vmh, OBJT_DEVICE, &netmap_cdev_pager_ops, objsize, prot, *foff, NULL); if (obj == NULL) { nm_prerr("cdev_pager_allocate failed"); error = EINVAL; goto err_deref; } *objp = obj; return 0; err_deref: NMG_LOCK(); priv->np_refs--; err_unlock: NMG_UNLOCK(); // err: free(vmh, M_DEVBUF); return error; } /* * On FreeBSD the close routine is only called on the last close on * the device (/dev/netmap) so we cannot do anything useful. * To track close() on individual file descriptors we pass netmap_dtor() to * devfs_set_cdevpriv() on open(). The FreeBSD kernel will call the destructor * when the last fd pointing to the device is closed. * * Note that FreeBSD does not even munmap() on close() so we also have * to track mmap() ourselves, and postpone the call to * netmap_dtor() is called when the process has no open fds and no active * memory maps on /dev/netmap, as in linux. */ static int netmap_close(struct cdev *dev, int fflag, int devtype, struct thread *td) { if (netmap_verbose) nm_prinf("dev %p fflag 0x%x devtype %d td %p", dev, fflag, devtype, td); return 0; } static int netmap_open(struct cdev *dev, int oflags, int devtype, struct thread *td) { struct netmap_priv_d *priv; int error; (void)dev; (void)oflags; (void)devtype; (void)td; NMG_LOCK(); priv = netmap_priv_new(); if (priv == NULL) { error = ENOMEM; goto out; } error = devfs_set_cdevpriv(priv, netmap_dtor); if (error) { netmap_priv_delete(priv); } out: NMG_UNLOCK(); return error; } /******************** kthread wrapper ****************/ #include u_int nm_os_ncpus(void) { return mp_maxid + 1; } struct nm_kctx_ctx { /* Userspace thread (kthread creator). */ struct thread *user_td; /* worker function and parameter */ nm_kctx_worker_fn_t worker_fn; void *worker_private; struct nm_kctx *nmk; /* integer to manage multiple worker contexts (e.g., RX or TX on ptnetmap) */ long type; }; struct nm_kctx { struct thread *worker; struct mtx worker_lock; struct nm_kctx_ctx worker_ctx; int run; /* used to stop kthread */ int attach_user; /* kthread attached to user_process */ int affinity; }; static void nm_kctx_worker(void *data) { struct nm_kctx *nmk = data; struct nm_kctx_ctx *ctx = &nmk->worker_ctx; if (nmk->affinity >= 0) { thread_lock(curthread); sched_bind(curthread, nmk->affinity); thread_unlock(curthread); } while (nmk->run) { /* * check if the parent process dies * (when kthread is attached to user process) */ if (ctx->user_td) { PROC_LOCK(curproc); thread_suspend_check(0); PROC_UNLOCK(curproc); } else { kthread_suspend_check(); } /* Continuously execute worker process. */ ctx->worker_fn(ctx->worker_private); /* worker body */ } kthread_exit(); } void nm_os_kctx_worker_setaff(struct nm_kctx *nmk, int affinity) { nmk->affinity = affinity; } struct nm_kctx * nm_os_kctx_create(struct nm_kctx_cfg *cfg, void *opaque) { struct nm_kctx *nmk = NULL; nmk = malloc(sizeof(*nmk), M_DEVBUF, M_NOWAIT | M_ZERO); if (!nmk) return NULL; mtx_init(&nmk->worker_lock, "nm_kthread lock", NULL, MTX_DEF); nmk->worker_ctx.worker_fn = cfg->worker_fn; nmk->worker_ctx.worker_private = cfg->worker_private; nmk->worker_ctx.type = cfg->type; nmk->affinity = -1; /* attach kthread to user process (ptnetmap) */ nmk->attach_user = cfg->attach_user; return nmk; } int nm_os_kctx_worker_start(struct nm_kctx *nmk) { struct proc *p = NULL; int error = 0; /* Temporarily disable this function as it is currently broken * and causes kernel crashes. The failure can be triggered by * the "vale_polling_enable_disable" test in ctrl-api-test.c. */ return EOPNOTSUPP; if (nmk->worker) return EBUSY; /* check if we want to attach kthread to user process */ if (nmk->attach_user) { nmk->worker_ctx.user_td = curthread; p = curthread->td_proc; } /* enable kthread main loop */ nmk->run = 1; /* create kthread */ if((error = kthread_add(nm_kctx_worker, nmk, p, &nmk->worker, RFNOWAIT /* to be checked */, 0, "nm-kthread-%ld", nmk->worker_ctx.type))) { goto err; } nm_prinf("nm_kthread started td %p", nmk->worker); return 0; err: nm_prerr("nm_kthread start failed err %d", error); nmk->worker = NULL; return error; } void nm_os_kctx_worker_stop(struct nm_kctx *nmk) { if (!nmk->worker) return; /* tell to kthread to exit from main loop */ nmk->run = 0; /* wake up kthread if it sleeps */ kthread_resume(nmk->worker); nmk->worker = NULL; } void nm_os_kctx_destroy(struct nm_kctx *nmk) { if (!nmk) return; if (nmk->worker) nm_os_kctx_worker_stop(nmk); free(nmk, M_DEVBUF); } /******************** kqueue support ****************/ /* * In addition to calling selwakeuppri(), nm_os_selwakeup() also * needs to call knote() to wake up kqueue listeners. * This operation is deferred to a taskqueue in order to avoid possible * lock order reversals; these may happen because knote() grabs a * private lock associated to the 'si' (see struct selinfo, * struct nm_selinfo, and nm_os_selinfo_init), and nm_os_selwakeup() * can be called while holding the lock associated to a different * 'si'. * When calling knote() we use a non-zero 'hint' argument to inform * the netmap_knrw() function that it is being called from * 'nm_os_selwakeup'; this is necessary because when netmap_knrw() is * called by the kevent subsystem (i.e. kevent_scan()) we also need to * call netmap_poll(). * * The netmap_kqfilter() function registers one or another f_event * depending on read or write mode. A pointer to the struct * 'netmap_priv_d' is stored into kn->kn_hook, so that it can later * be passed to netmap_poll(). We pass NULL as a third argument to * netmap_poll(), so that the latter only runs the txsync/rxsync * (if necessary), and skips the nm_os_selrecord() calls. */ void nm_os_selwakeup(struct nm_selinfo *si) { selwakeuppri(&si->si, PI_NET); if (si->kqueue_users > 0) { taskqueue_enqueue(si->ntfytq, &si->ntfytask); } } void nm_os_selrecord(struct thread *td, struct nm_selinfo *si) { selrecord(td, &si->si); } static void netmap_knrdetach(struct knote *kn) { struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook; struct nm_selinfo *si = priv->np_si[NR_RX]; knlist_remove(&si->si.si_note, kn, /*islocked=*/0); NMG_LOCK(); KASSERT(si->kqueue_users > 0, ("kqueue_user underflow on %s", si->mtxname)); si->kqueue_users--; nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); NMG_UNLOCK(); } static void netmap_knwdetach(struct knote *kn) { struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook; struct nm_selinfo *si = priv->np_si[NR_TX]; knlist_remove(&si->si.si_note, kn, /*islocked=*/0); NMG_LOCK(); si->kqueue_users--; nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); NMG_UNLOCK(); } /* * Callback triggered by netmap notifications (see netmap_notify()), * and by the application calling kevent(). In the former case we * just return 1 (events ready), since we are not able to do better. * In the latter case we use netmap_poll() to see which events are * ready. */ static int netmap_knrw(struct knote *kn, long hint, int events) { struct netmap_priv_d *priv; int revents; if (hint != 0) { /* Called from netmap_notify(), typically from a * thread different from the one issuing kevent(). * Assume we are ready. */ return 1; } /* Called from kevent(). */ priv = kn->kn_hook; revents = netmap_poll(priv, events, /*thread=*/NULL); return (events & revents) ? 1 : 0; } static int netmap_knread(struct knote *kn, long hint) { return netmap_knrw(kn, hint, POLLIN); } static int netmap_knwrite(struct knote *kn, long hint) { return netmap_knrw(kn, hint, POLLOUT); } static struct filterops netmap_rfiltops = { .f_isfd = 1, .f_detach = netmap_knrdetach, .f_event = netmap_knread, }; static struct filterops netmap_wfiltops = { .f_isfd = 1, .f_detach = netmap_knwdetach, .f_event = netmap_knwrite, }; /* * This is called when a thread invokes kevent() to record * a change in the configuration of the kqueue(). * The 'priv' is the one associated to the open netmap device. */ static int netmap_kqfilter(struct cdev *dev, struct knote *kn) { struct netmap_priv_d *priv; int error; struct netmap_adapter *na; struct nm_selinfo *si; int ev = kn->kn_filter; if (ev != EVFILT_READ && ev != EVFILT_WRITE) { nm_prerr("bad filter request %d", ev); return 1; } error = devfs_get_cdevpriv((void**)&priv); if (error) { nm_prerr("device not yet setup"); return 1; } na = priv->np_na; if (na == NULL) { nm_prerr("no netmap adapter for this file descriptor"); return 1; } /* the si is indicated in the priv */ si = priv->np_si[(ev == EVFILT_WRITE) ? NR_TX : NR_RX]; kn->kn_fop = (ev == EVFILT_WRITE) ? &netmap_wfiltops : &netmap_rfiltops; kn->kn_hook = priv; NMG_LOCK(); si->kqueue_users++; nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); NMG_UNLOCK(); knlist_add(&si->si.si_note, kn, /*islocked=*/0); return 0; } static int freebsd_netmap_poll(struct cdev *cdevi __unused, int events, struct thread *td) { struct netmap_priv_d *priv; if (devfs_get_cdevpriv((void **)&priv)) { return POLLERR; } return netmap_poll(priv, events, td); } static int freebsd_netmap_ioctl(struct cdev *dev __unused, u_long cmd, caddr_t data, int ffla __unused, struct thread *td) { int error; struct netmap_priv_d *priv; CURVNET_SET(TD_TO_VNET(td)); error = devfs_get_cdevpriv((void **)&priv); if (error) { /* XXX ENOENT should be impossible, since the priv * is now created in the open */ if (error == ENOENT) error = ENXIO; goto out; } error = netmap_ioctl(priv, cmd, data, td, /*nr_body_is_user=*/1); out: CURVNET_RESTORE(); return error; } void nm_os_onattach(struct ifnet *ifp) { ifp->if_capabilities |= IFCAP_NETMAP; } void nm_os_onenter(struct ifnet *ifp) { struct netmap_adapter *na = NA(ifp); na->if_transmit = ifp->if_transmit; ifp->if_transmit = netmap_transmit; ifp->if_capenable |= IFCAP_NETMAP; } void nm_os_onexit(struct ifnet *ifp) { struct netmap_adapter *na = NA(ifp); ifp->if_transmit = na->if_transmit; ifp->if_capenable &= ~IFCAP_NETMAP; } extern struct cdevsw netmap_cdevsw; /* XXX used in netmap.c, should go elsewhere */ struct cdevsw netmap_cdevsw = { .d_version = D_VERSION, .d_name = "netmap", .d_open = netmap_open, .d_mmap_single = netmap_mmap_single, .d_ioctl = freebsd_netmap_ioctl, .d_poll = freebsd_netmap_poll, .d_kqfilter = netmap_kqfilter, .d_close = netmap_close, }; /*--- end of kqueue support ----*/ /* * Kernel entry point. * * Initialize/finalize the module and return. * * Return 0 on success, errno on failure. */ static int netmap_loader(__unused struct module *module, int event, __unused void *arg) { int error = 0; switch (event) { case MOD_LOAD: error = netmap_init(); break; case MOD_UNLOAD: /* * if some one is still using netmap, * then the module can not be unloaded. */ if (netmap_use_count) { nm_prerr("netmap module can not be unloaded - netmap_use_count: %d", netmap_use_count); error = EBUSY; break; } netmap_fini(); break; default: error = EOPNOTSUPP; break; } return (error); } #ifdef DEV_MODULE_ORDERED /* * The netmap module contains three drivers: (i) the netmap character device * driver; (ii) the ptnetmap memdev PCI device driver, (iii) the ptnet PCI * device driver. The attach() routines of both (ii) and (iii) need the * lock of the global allocator, and such lock is initialized in netmap_init(), * which is part of (i). * Therefore, we make sure that (i) is loaded before (ii) and (iii), using * the 'order' parameter of driver declaration macros. For (i), we specify * SI_ORDER_MIDDLE, while higher orders are used with the DRIVER_MODULE_ORDERED * macros for (ii) and (iii). */ DEV_MODULE_ORDERED(netmap, netmap_loader, NULL, SI_ORDER_MIDDLE); #else /* !DEV_MODULE_ORDERED */ DEV_MODULE(netmap, netmap_loader, NULL); #endif /* DEV_MODULE_ORDERED */ MODULE_DEPEND(netmap, pci, 1, 1, 1); MODULE_VERSION(netmap, 1); /* reduce conditional code */ // linux API, use for the knlist in FreeBSD /* use a private mutex for the knlist */ Index: head/sys/netinet/netdump/netdump_client.c =================================================================== --- head/sys/netinet/netdump/netdump_client.c (revision 348021) +++ head/sys/netinet/netdump/netdump_client.c (revision 348022) @@ -1,1510 +1,1511 @@ /*- * Copyright (c) 2005-2014 Sandvine Incorporated. All rights reserved. * Copyright (c) 2000 Darrell Anderson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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. */ /* * netdump_client.c * FreeBSD subsystem supporting netdump network dumps. * A dedicated server must be running to accept client dumps. */ #include __FBSDID("$FreeBSD$"); #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 #include #define NETDDEBUG(f, ...) do { \ if (nd_debug > 0) \ printf(("%s: " f), __func__, ## __VA_ARGS__); \ } while (0) #define NETDDEBUG_IF(i, f, ...) do { \ if (nd_debug > 0) \ if_printf((i), ("%s: " f), __func__, ## __VA_ARGS__); \ } while (0) #define NETDDEBUGV(f, ...) do { \ if (nd_debug > 1) \ printf(("%s: " f), __func__, ## __VA_ARGS__); \ } while (0) #define NETDDEBUGV_IF(i, f, ...) do { \ if (nd_debug > 1) \ if_printf((i), ("%s: " f), __func__, ## __VA_ARGS__); \ } while (0) static int netdump_arp_gw(void); static void netdump_cleanup(void); static int netdump_configure(struct diocskerneldump_arg *, struct thread *); static int netdump_dumper(void *priv __unused, void *virtual, vm_offset_t physical __unused, off_t offset, size_t length); static bool netdump_enabled(void); static int netdump_enabled_sysctl(SYSCTL_HANDLER_ARGS); static int netdump_ether_output(struct mbuf *m, struct ifnet *ifp, struct ether_addr dst, u_short etype); static void netdump_handle_arp(struct mbuf **mb); static void netdump_handle_ip(struct mbuf **mb); static int netdump_ioctl(struct cdev *dev __unused, u_long cmd, caddr_t addr, int flags __unused, struct thread *td); static int netdump_modevent(module_t mod, int type, void *priv); static void netdump_network_poll(void); static void netdump_pkt_in(struct ifnet *ifp, struct mbuf *m); static void netdump_reinit_internal(struct ifnet *ifp); static int netdump_send(uint32_t type, off_t offset, unsigned char *data, uint32_t datalen); static int netdump_send_arp(in_addr_t dst); static int netdump_start(struct dumperinfo *di); static int netdump_udp_output(struct mbuf *m); static void netdump_unconfigure(void); /* Must be at least as big as the chunks dumpsys() gives us. */ static unsigned char nd_buf[MAXDUMPPGS * PAGE_SIZE]; static uint32_t nd_seqno; static int dump_failed, have_gw_mac; static void (*drv_if_input)(struct ifnet *, struct mbuf *); static int restore_gw_addr; static uint64_t rcvd_acks; CTASSERT(sizeof(rcvd_acks) * NBBY == NETDUMP_MAX_IN_FLIGHT); /* Configuration parameters. */ static struct { char ndc_iface[IFNAMSIZ]; union kd_ip ndc_server; union kd_ip ndc_client; union kd_ip ndc_gateway; uint8_t ndc_af; } nd_conf; #define nd_server nd_conf.ndc_server.in4 #define nd_client nd_conf.ndc_client.in4 #define nd_gateway nd_conf.ndc_gateway.in4 /* General dynamic settings. */ static struct sx nd_conf_lk; SX_SYSINIT(nd_conf, &nd_conf_lk, "netdump configuration lock"); #define NETDUMP_WLOCK() sx_xlock(&nd_conf_lk) #define NETDUMP_WUNLOCK() sx_xunlock(&nd_conf_lk) #define NETDUMP_RLOCK() sx_slock(&nd_conf_lk) #define NETDUMP_RUNLOCK() sx_sunlock(&nd_conf_lk) #define NETDUMP_ASSERT_WLOCKED() sx_assert(&nd_conf_lk, SA_XLOCKED) #define NETDUMP_ASSERT_LOCKED() sx_assert(&nd_conf_lk, SA_LOCKED) static struct ether_addr nd_gw_mac; static struct ifnet *nd_ifp; static eventhandler_tag nd_detach_cookie; static uint16_t nd_server_port = NETDUMP_PORT; FEATURE(netdump, "Netdump client support"); static SYSCTL_NODE(_net, OID_AUTO, netdump, CTLFLAG_RD, NULL, "netdump parameters"); static int nd_debug; SYSCTL_INT(_net_netdump, OID_AUTO, debug, CTLFLAG_RWTUN, &nd_debug, 0, "Debug message verbosity"); SYSCTL_PROC(_net_netdump, OID_AUTO, enabled, CTLFLAG_RD | CTLTYPE_INT, &nd_ifp, 0, netdump_enabled_sysctl, "I", "netdump configuration status"); static char nd_path[MAXPATHLEN]; SYSCTL_STRING(_net_netdump, OID_AUTO, path, CTLFLAG_RW, nd_path, sizeof(nd_path), "Server path for output files"); static int nd_polls = 2000; SYSCTL_INT(_net_netdump, OID_AUTO, polls, CTLFLAG_RWTUN, &nd_polls, 0, "Number of times to poll before assuming packet loss (0.5ms per poll)"); static int nd_retries = 10; SYSCTL_INT(_net_netdump, OID_AUTO, retries, CTLFLAG_RWTUN, &nd_retries, 0, "Number of retransmit attempts before giving up"); static int nd_arp_retries = 3; SYSCTL_INT(_net_netdump, OID_AUTO, arp_retries, CTLFLAG_RWTUN, &nd_arp_retries, 0, "Number of ARP attempts before giving up"); static bool netdump_enabled(void) { NETDUMP_ASSERT_LOCKED(); return (nd_ifp != NULL); } static int netdump_enabled_sysctl(SYSCTL_HANDLER_ARGS) { int en, error; NETDUMP_RLOCK(); en = netdump_enabled(); NETDUMP_RUNLOCK(); error = SYSCTL_OUT(req, &en, sizeof(en)); if (error != 0 || req->newptr == NULL) return (error); return (EPERM); } /* * Checks for netdump support on a network interface * * Parameters: * ifp The network interface that is being tested for support * * Returns: * int 1 if the interface is supported, 0 if not */ static bool netdump_supported_nic(struct ifnet *ifp) { return (ifp->if_netdump_methods != NULL); } /*- * Network specific primitives. * Following down the code they are divided ordered as: * - Packet buffer primitives * - Output primitives * - Input primitives * - Polling primitives */ /* * Handles creation of the ethernet header, then places outgoing packets into * the tx buffer for the NIC * * Parameters: * m The mbuf containing the packet to be sent (will be freed by * this function or the NIC driver) * ifp The interface to send on * dst The destination ethernet address (source address will be looked * up using ifp) * etype The ETHERTYPE_* value for the protocol that is being sent * * Returns: * int see errno.h, 0 for success */ static int netdump_ether_output(struct mbuf *m, struct ifnet *ifp, struct ether_addr dst, u_short etype) { struct ether_header *eh; if (((ifp->if_flags & (IFF_MONITOR | IFF_UP)) != IFF_UP) || (ifp->if_drv_flags & IFF_DRV_RUNNING) != IFF_DRV_RUNNING) { if_printf(ifp, "netdump_ether_output: interface isn't up\n"); m_freem(m); return (ENETDOWN); } /* Fill in the ethernet header. */ M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT); if (m == NULL) { printf("%s: out of mbufs\n", __func__); return (ENOBUFS); } eh = mtod(m, struct ether_header *); memcpy(eh->ether_shost, IF_LLADDR(ifp), ETHER_ADDR_LEN); memcpy(eh->ether_dhost, dst.octet, ETHER_ADDR_LEN); eh->ether_type = htons(etype); return ((ifp->if_netdump_methods->nd_transmit)(ifp, m)); } /* * Unreliable transmission of an mbuf chain to the netdump server * Note: can't handle fragmentation; fails if the packet is larger than * nd_ifp->if_mtu after adding the UDP/IP headers * * Parameters: * m mbuf chain * * Returns: * int see errno.h, 0 for success */ static int netdump_udp_output(struct mbuf *m) { struct udpiphdr *ui; struct ip *ip; MPASS(netdump_enabled()); M_PREPEND(m, sizeof(struct udpiphdr), M_NOWAIT); if (m == NULL) { printf("%s: out of mbufs\n", __func__); return (ENOBUFS); } if (m->m_pkthdr.len > nd_ifp->if_mtu) { printf("netdump_udp_output: Packet is too big: %d > MTU %u\n", m->m_pkthdr.len, nd_ifp->if_mtu); m_freem(m); return (ENOBUFS); } ui = mtod(m, struct udpiphdr *); bzero(ui->ui_x1, sizeof(ui->ui_x1)); ui->ui_pr = IPPROTO_UDP; ui->ui_len = htons(m->m_pkthdr.len - sizeof(struct ip)); ui->ui_ulen = ui->ui_len; ui->ui_src = nd_client; ui->ui_dst = nd_server; /* Use this src port so that the server can connect() the socket */ ui->ui_sport = htons(NETDUMP_ACKPORT); ui->ui_dport = htons(nd_server_port); ui->ui_sum = 0; if ((ui->ui_sum = in_cksum(m, m->m_pkthdr.len)) == 0) ui->ui_sum = 0xffff; ip = mtod(m, struct ip *); ip->ip_v = IPVERSION; ip->ip_hl = sizeof(struct ip) >> 2; ip->ip_tos = 0; ip->ip_len = htons(m->m_pkthdr.len); ip->ip_id = 0; ip->ip_off = htons(IP_DF); ip->ip_ttl = 255; ip->ip_sum = 0; ip->ip_sum = in_cksum(m, sizeof(struct ip)); return (netdump_ether_output(m, nd_ifp, nd_gw_mac, ETHERTYPE_IP)); } /* * Builds and sends a single ARP request to locate the server * * Return value: * 0 on success * errno on error */ static int netdump_send_arp(in_addr_t dst) { struct ether_addr bcast; struct mbuf *m; struct arphdr *ah; int pktlen; MPASS(netdump_enabled()); /* Fill-up a broadcast address. */ memset(&bcast, 0xFF, ETHER_ADDR_LEN); m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { printf("netdump_send_arp: Out of mbufs\n"); return (ENOBUFS); } pktlen = arphdr_len2(ETHER_ADDR_LEN, sizeof(struct in_addr)); m->m_len = pktlen; m->m_pkthdr.len = pktlen; MH_ALIGN(m, pktlen); ah = mtod(m, struct arphdr *); ah->ar_hrd = htons(ARPHRD_ETHER); ah->ar_pro = htons(ETHERTYPE_IP); ah->ar_hln = ETHER_ADDR_LEN; ah->ar_pln = sizeof(struct in_addr); ah->ar_op = htons(ARPOP_REQUEST); memcpy(ar_sha(ah), IF_LLADDR(nd_ifp), ETHER_ADDR_LEN); ((struct in_addr *)ar_spa(ah))->s_addr = nd_client.s_addr; bzero(ar_tha(ah), ETHER_ADDR_LEN); ((struct in_addr *)ar_tpa(ah))->s_addr = dst; return (netdump_ether_output(m, nd_ifp, bcast, ETHERTYPE_ARP)); } /* * Sends ARP requests to locate the server and waits for a response. * We first try to ARP the server itself, and fall back to the provided * gateway if the server appears to be off-link. * * Return value: * 0 on success * errno on error */ static int netdump_arp_gw(void) { in_addr_t dst; int error, polls, retries; dst = nd_server.s_addr; restart: for (retries = 0; retries < nd_arp_retries && have_gw_mac == 0; retries++) { error = netdump_send_arp(dst); if (error != 0) return (error); for (polls = 0; polls < nd_polls && have_gw_mac == 0; polls++) { netdump_network_poll(); DELAY(500); } if (have_gw_mac == 0) printf("(ARP retry)"); } if (have_gw_mac != 0) return (0); if (dst == nd_server.s_addr && nd_server.s_addr != nd_gateway.s_addr) { printf("Failed to ARP server, trying to reach gateway...\n"); dst = nd_gateway.s_addr; goto restart; } printf("\nARP timed out.\n"); return (ETIMEDOUT); } /* * Dummy free function for netdump clusters. */ static void netdump_mbuf_free(struct mbuf *m __unused) { } /* * Construct and reliably send a netdump packet. May fail from a resource * shortage or extreme number of unacknowledged retransmissions. Wait for * an acknowledgement before returning. Splits packets into chunks small * enough to be sent without fragmentation (looks up the interface MTU) * * Parameters: * type netdump packet type (HERALD, FINISHED, or VMCORE) * offset vmcore data offset (bytes) * data vmcore data * datalen vmcore data size (bytes) * * Returns: * int see errno.h, 0 for success */ static int netdump_send(uint32_t type, off_t offset, unsigned char *data, uint32_t datalen) { struct netdump_msg_hdr *nd_msg_hdr; struct mbuf *m, *m2; uint64_t want_acks; uint32_t i, pktlen, sent_so_far; int retries, polls, error; want_acks = 0; rcvd_acks = 0; retries = 0; MPASS(netdump_enabled()); retransmit: /* Chunks can be too big to fit in packets. */ for (i = sent_so_far = 0; sent_so_far < datalen || (i == 0 && datalen == 0); i++) { pktlen = datalen - sent_so_far; /* First bound: the packet structure. */ pktlen = min(pktlen, NETDUMP_DATASIZE); /* Second bound: the interface MTU (assume no IP options). */ pktlen = min(pktlen, nd_ifp->if_mtu - sizeof(struct udpiphdr) - sizeof(struct netdump_msg_hdr)); /* * Check if it is retransmitting and this has been ACKed * already. */ if ((rcvd_acks & (1 << i)) != 0) { sent_so_far += pktlen; continue; } /* * Get and fill a header mbuf, then chain data as an extended * mbuf. */ m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { printf("netdump_send: Out of mbufs\n"); return (ENOBUFS); } m->m_len = sizeof(struct netdump_msg_hdr); m->m_pkthdr.len = sizeof(struct netdump_msg_hdr); MH_ALIGN(m, sizeof(struct netdump_msg_hdr)); nd_msg_hdr = mtod(m, struct netdump_msg_hdr *); nd_msg_hdr->mh_seqno = htonl(nd_seqno + i); nd_msg_hdr->mh_type = htonl(type); nd_msg_hdr->mh_offset = htobe64(offset + sent_so_far); nd_msg_hdr->mh_len = htonl(pktlen); nd_msg_hdr->mh__pad = 0; if (pktlen != 0) { m2 = m_get(M_NOWAIT, MT_DATA); if (m2 == NULL) { m_freem(m); printf("netdump_send: Out of mbufs\n"); return (ENOBUFS); } MEXTADD(m2, data + sent_so_far, pktlen, netdump_mbuf_free, NULL, NULL, 0, EXT_DISPOSABLE); m2->m_len = pktlen; m_cat(m, m2); m->m_pkthdr.len += pktlen; } error = netdump_udp_output(m); if (error != 0) return (error); /* Note that we're waiting for this packet in the bitfield. */ want_acks |= (1 << i); sent_so_far += pktlen; } if (i >= NETDUMP_MAX_IN_FLIGHT) printf("Warning: Sent more than %d packets (%d). " "Acknowledgements will fail unless the size of " "rcvd_acks/want_acks is increased.\n", NETDUMP_MAX_IN_FLIGHT, i); /* * Wait for acks. A *real* window would speed things up considerably. */ polls = 0; while (rcvd_acks != want_acks) { if (polls++ > nd_polls) { if (retries++ > nd_retries) return (ETIMEDOUT); printf(". "); goto retransmit; } netdump_network_poll(); DELAY(500); } nd_seqno += i; return (0); } /* * Handler for IP packets: checks their sanity and then processes any netdump * ACK packets it finds. * * It needs to replicate partially the behaviour of ip_input() and * udp_input(). * * Parameters: * mb a pointer to an mbuf * containing the packet received * Updates *mb if m_pullup et al change the pointer * Assumes the calling function will take care of freeing the mbuf */ static void netdump_handle_ip(struct mbuf **mb) { struct ip *ip; struct udpiphdr *udp; struct netdump_ack *nd_ack; struct mbuf *m; int rcv_ackno; unsigned short hlen; /* IP processing. */ m = *mb; if (m->m_pkthdr.len < sizeof(struct ip)) { NETDDEBUG("dropping packet too small for IP header\n"); return; } if (m->m_len < sizeof(struct ip)) { m = m_pullup(m, sizeof(struct ip)); *mb = m; if (m == NULL) { NETDDEBUG("m_pullup failed\n"); return; } } ip = mtod(m, struct ip *); /* IP version. */ if (ip->ip_v != IPVERSION) { NETDDEBUG("bad IP version %d\n", ip->ip_v); return; } /* Header length. */ hlen = ip->ip_hl << 2; if (hlen < sizeof(struct ip)) { NETDDEBUG("bad IP header length (%hu)\n", hlen); return; } if (hlen > m->m_len) { m = m_pullup(m, hlen); *mb = m; if (m == NULL) { NETDDEBUG("m_pullup failed\n"); return; } ip = mtod(m, struct ip *); } /* Ignore packets with IP options. */ if (hlen > sizeof(struct ip)) { NETDDEBUG("drop packet with IP options\n"); return; } #ifdef INVARIANTS if ((IN_LOOPBACK(ntohl(ip->ip_dst.s_addr)) || IN_LOOPBACK(ntohl(ip->ip_src.s_addr))) && (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) { NETDDEBUG("Bad IP header (RFC1122)\n"); return; } #endif /* Checksum. */ if ((m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) != 0) { if ((m->m_pkthdr.csum_flags & CSUM_IP_VALID) == 0) { NETDDEBUG("bad IP checksum\n"); return; } } else { /* XXX */ ; } /* Convert fields to host byte order. */ ip->ip_len = ntohs(ip->ip_len); if (ip->ip_len < hlen) { NETDDEBUG("IP packet smaller (%hu) than header (%hu)\n", ip->ip_len, hlen); return; } if (m->m_pkthdr.len < ip->ip_len) { NETDDEBUG("IP packet bigger (%hu) than ethernet packet (%d)\n", ip->ip_len, m->m_pkthdr.len); return; } if (m->m_pkthdr.len > ip->ip_len) { /* Truncate the packet to the IP length. */ if (m->m_len == m->m_pkthdr.len) { m->m_len = ip->ip_len; m->m_pkthdr.len = ip->ip_len; } else m_adj(m, ip->ip_len - m->m_pkthdr.len); } ip->ip_off = ntohs(ip->ip_off); /* Check that the source is the server's IP. */ if (ip->ip_src.s_addr != nd_server.s_addr) { NETDDEBUG("drop packet not from server (from 0x%x)\n", ip->ip_src.s_addr); return; } /* Check if the destination IP is ours. */ if (ip->ip_dst.s_addr != nd_client.s_addr) { NETDDEBUGV("drop packet not to our IP\n"); return; } if (ip->ip_p != IPPROTO_UDP) { NETDDEBUG("drop non-UDP packet\n"); return; } /* Do not deal with fragments. */ if ((ip->ip_off & (IP_MF | IP_OFFMASK)) != 0) { NETDDEBUG("drop fragmented packet\n"); return; } /* UDP custom is to have packet length not include IP header. */ ip->ip_len -= hlen; /* UDP processing. */ /* Get IP and UDP headers together, along with the netdump packet. */ if (m->m_pkthdr.len < sizeof(struct udpiphdr) + sizeof(struct netdump_ack)) { NETDDEBUG("ignoring small packet\n"); return; } if (m->m_len < sizeof(struct udpiphdr) + sizeof(struct netdump_ack)) { m = m_pullup(m, sizeof(struct udpiphdr) + sizeof(struct netdump_ack)); *mb = m; if (m == NULL) { NETDDEBUG("m_pullup failed\n"); return; } } udp = mtod(m, struct udpiphdr *); if (ntohs(udp->ui_u.uh_dport) != NETDUMP_ACKPORT) { NETDDEBUG("not on the netdump port.\n"); return; } /* Netdump processing. */ /* * Packet is meant for us. Extract the ack sequence number and the * port number if necessary. */ nd_ack = (struct netdump_ack *)(mtod(m, caddr_t) + sizeof(struct udpiphdr)); rcv_ackno = ntohl(nd_ack->na_seqno); if (nd_server_port == NETDUMP_PORT) nd_server_port = ntohs(udp->ui_u.uh_sport); if (rcv_ackno >= nd_seqno + NETDUMP_MAX_IN_FLIGHT) printf("%s: ACK %d too far in future!\n", __func__, rcv_ackno); else if (rcv_ackno >= nd_seqno) { /* We're interested in this ack. Record it. */ rcvd_acks |= 1 << (rcv_ackno - nd_seqno); } } /* * Handler for ARP packets: checks their sanity and then * 1. If the ARP is a request for our IP, respond with our MAC address * 2. If the ARP is a response from our server, record its MAC address * * It needs to replicate partially the behaviour of arpintr() and * in_arpinput(). * * Parameters: * mb a pointer to an mbuf * containing the packet received * Updates *mb if m_pullup et al change the pointer * Assumes the calling function will take care of freeing the mbuf */ static void netdump_handle_arp(struct mbuf **mb) { char buf[INET_ADDRSTRLEN]; struct in_addr isaddr, itaddr, myaddr; struct ether_addr dst; struct mbuf *m; struct arphdr *ah; struct ifnet *ifp; uint8_t *enaddr; int req_len, op; m = *mb; ifp = m->m_pkthdr.rcvif; if (m->m_len < sizeof(struct arphdr)) { m = m_pullup(m, sizeof(struct arphdr)); *mb = m; if (m == NULL) { NETDDEBUG("runt packet: m_pullup failed\n"); return; } } ah = mtod(m, struct arphdr *); if (ntohs(ah->ar_hrd) != ARPHRD_ETHER) { NETDDEBUG("unknown hardware address 0x%2D)\n", (unsigned char *)&ah->ar_hrd, ""); return; } if (ntohs(ah->ar_pro) != ETHERTYPE_IP) { NETDDEBUG("drop ARP for unknown protocol %d\n", ntohs(ah->ar_pro)); return; } req_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr)); if (m->m_len < req_len) { m = m_pullup(m, req_len); *mb = m; if (m == NULL) { NETDDEBUG("runt packet: m_pullup failed\n"); return; } } ah = mtod(m, struct arphdr *); op = ntohs(ah->ar_op); memcpy(&isaddr, ar_spa(ah), sizeof(isaddr)); memcpy(&itaddr, ar_tpa(ah), sizeof(itaddr)); enaddr = (uint8_t *)IF_LLADDR(ifp); myaddr = nd_client; if (memcmp(ar_sha(ah), enaddr, ifp->if_addrlen) == 0) { NETDDEBUG("ignoring ARP from myself\n"); return; } if (isaddr.s_addr == nd_client.s_addr) { printf("%s: %*D is using my IP address %s!\n", __func__, ifp->if_addrlen, (u_char *)ar_sha(ah), ":", inet_ntoa_r(isaddr, buf)); return; } if (memcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen) == 0) { NETDDEBUG("ignoring ARP from broadcast address\n"); return; } if (op == ARPOP_REPLY) { if (isaddr.s_addr != nd_gateway.s_addr && isaddr.s_addr != nd_server.s_addr) { inet_ntoa_r(isaddr, buf); NETDDEBUG( "ignoring ARP reply from %s (not netdump server)\n", buf); return; } memcpy(nd_gw_mac.octet, ar_sha(ah), min(ah->ar_hln, ETHER_ADDR_LEN)); have_gw_mac = 1; NETDDEBUG("got server MAC address %6D\n", nd_gw_mac.octet, ":"); return; } if (op != ARPOP_REQUEST) { NETDDEBUG("ignoring ARP non-request/reply\n"); return; } if (itaddr.s_addr != nd_client.s_addr) { NETDDEBUG("ignoring ARP not to our IP\n"); return; } memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln); memcpy(ar_sha(ah), enaddr, ah->ar_hln); memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln); memcpy(ar_spa(ah), &itaddr, ah->ar_pln); ah->ar_op = htons(ARPOP_REPLY); ah->ar_pro = htons(ETHERTYPE_IP); m->m_flags &= ~(M_BCAST|M_MCAST); m->m_len = arphdr_len(ah); m->m_pkthdr.len = m->m_len; memcpy(dst.octet, ar_tha(ah), ETHER_ADDR_LEN); netdump_ether_output(m, ifp, dst, ETHERTYPE_ARP); *mb = NULL; } /* * Handler for incoming packets directly from the network adapter * Identifies the packet type (IP or ARP) and passes it along to one of the * helper functions netdump_handle_ip or netdump_handle_arp. * * It needs to replicate partially the behaviour of ether_input() and * ether_demux(). * * Parameters: * ifp the interface the packet came from (should be nd_ifp) * m an mbuf containing the packet received */ static void netdump_pkt_in(struct ifnet *ifp, struct mbuf *m) { struct ifreq ifr; struct ether_header *eh; u_short etype; /* Ethernet processing. */ if ((m->m_flags & M_PKTHDR) == 0) { NETDDEBUG_IF(ifp, "discard frame without packet header\n"); goto done; } if (m->m_len < ETHER_HDR_LEN) { NETDDEBUG_IF(ifp, "discard frame without leading eth header (len %u pktlen %u)\n", m->m_len, m->m_pkthdr.len); goto done; } if ((m->m_flags & M_HASFCS) != 0) { m_adj(m, -ETHER_CRC_LEN); m->m_flags &= ~M_HASFCS; } eh = mtod(m, struct ether_header *); etype = ntohs(eh->ether_type); if ((m->m_flags & M_VLANTAG) != 0 || etype == ETHERTYPE_VLAN) { NETDDEBUG_IF(ifp, "ignoring vlan packets\n"); goto done; } if (if_gethwaddr(ifp, &ifr) != 0) { NETDDEBUG_IF(ifp, "failed to get hw addr for interface\n"); goto done; } if (memcmp(ifr.ifr_addr.sa_data, eh->ether_dhost, ETHER_ADDR_LEN) != 0) { NETDDEBUG_IF(ifp, "discard frame with incorrect destination addr\n"); goto done; } /* Done ethernet processing. Strip off the ethernet header. */ m_adj(m, ETHER_HDR_LEN); switch (etype) { case ETHERTYPE_ARP: netdump_handle_arp(&m); break; case ETHERTYPE_IP: netdump_handle_ip(&m); break; default: NETDDEBUG_IF(ifp, "dropping unknown ethertype %hu\n", etype); break; } done: if (m != NULL) m_freem(m); } /* * After trapping, instead of assuming that most of the network stack is sane, * we just poll the driver directly for packets. */ static void netdump_network_poll(void) { MPASS(netdump_enabled()); nd_ifp->if_netdump_methods->nd_poll(nd_ifp, 1000); } /*- * Dumping specific primitives. */ /* * Callback from dumpsys() to dump a chunk of memory. * Copies it out to our static buffer then sends it across the network. * Detects the initial KDH and makes sure it is given a special packet type. * * Parameters: * priv Unused. Optional private pointer. * virtual Virtual address (where to read the data from) * physical Unused. Physical memory address. * offset Offset from start of core file * length Data length * * Return value: * 0 on success * errno on error */ static int netdump_dumper(void *priv __unused, void *virtual, vm_offset_t physical __unused, off_t offset, size_t length) { int error; NETDDEBUGV("netdump_dumper(NULL, %p, NULL, %ju, %zu)\n", virtual, (uintmax_t)offset, length); if (virtual == NULL) { if (dump_failed != 0) printf("failed to dump the kernel core\n"); else if (netdump_send(NETDUMP_FINISHED, 0, NULL, 0) != 0) printf("failed to close the transaction\n"); else printf("\nnetdump finished.\n"); netdump_cleanup(); return (0); } if (length > sizeof(nd_buf)) return (ENOSPC); memmove(nd_buf, virtual, length); error = netdump_send(NETDUMP_VMCORE, offset, nd_buf, length); if (error != 0) { dump_failed = 1; return (error); } return (0); } /* * Perform any initalization needed prior to transmitting the kernel core. */ static int netdump_start(struct dumperinfo *di) { char *path; char buf[INET_ADDRSTRLEN]; uint32_t len; int error; error = 0; /* Check if the dumping is allowed to continue. */ if (!netdump_enabled()) return (EINVAL); if (panicstr == NULL) { printf( "netdump_start: netdump may only be used after a panic\n"); return (EINVAL); } if (nd_server.s_addr == INADDR_ANY) { printf("netdump_start: can't netdump; no server IP given\n"); return (EINVAL); } if (nd_client.s_addr == INADDR_ANY) { printf("netdump_start: can't netdump; no client IP given\n"); return (EINVAL); } /* We start dumping at offset 0. */ di->dumpoff = 0; nd_seqno = 1; /* * nd_server_port could have switched after the first ack the * first time it gets called. Adjust it accordingly. */ nd_server_port = NETDUMP_PORT; /* Switch to the netdump mbuf zones. */ netdump_mbuf_dump(); nd_ifp->if_netdump_methods->nd_event(nd_ifp, NETDUMP_START); /* Make the card use *our* receive callback. */ drv_if_input = nd_ifp->if_input; nd_ifp->if_input = netdump_pkt_in; if (nd_gateway.s_addr == INADDR_ANY) { restore_gw_addr = 1; nd_gateway.s_addr = nd_server.s_addr; } printf("netdump in progress. searching for server...\n"); if (netdump_arp_gw()) { printf("failed to locate server MAC address\n"); error = EINVAL; goto trig_abort; } if (nd_path[0] != '\0') { path = nd_path; len = strlen(path) + 1; } else { path = NULL; len = 0; } if (netdump_send(NETDUMP_HERALD, 0, path, len) != 0) { printf("failed to contact netdump server\n"); error = EINVAL; goto trig_abort; } printf("netdumping to %s (%6D)\n", inet_ntoa_r(nd_server, buf), nd_gw_mac.octet, ":"); return (0); trig_abort: netdump_cleanup(); return (error); } static int netdump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh, void *key, uint32_t keysize) { int error; memcpy(nd_buf, kdh, sizeof(*kdh)); error = netdump_send(NETDUMP_KDH, 0, nd_buf, sizeof(*kdh)); if (error == 0 && keysize > 0) { if (keysize > sizeof(nd_buf)) return (EINVAL); memcpy(nd_buf, key, keysize); error = netdump_send(NETDUMP_EKCD_KEY, 0, nd_buf, keysize); } return (error); } /* * Cleanup routine for a possibly failed netdump. */ static void netdump_cleanup(void) { if (restore_gw_addr != 0) { nd_gateway.s_addr = INADDR_ANY; restore_gw_addr = 0; } if (drv_if_input != NULL) { nd_ifp->if_input = drv_if_input; drv_if_input = NULL; } nd_ifp->if_netdump_methods->nd_event(nd_ifp, NETDUMP_END); } /*- * KLD specific code. */ static struct cdevsw netdump_cdevsw = { .d_version = D_VERSION, .d_ioctl = netdump_ioctl, .d_name = "netdump", }; static struct cdev *netdump_cdev; static void netdump_unconfigure(void) { struct diocskerneldump_arg kda; NETDUMP_ASSERT_WLOCKED(); KASSERT(netdump_enabled(), ("%s: nd_ifp NULL", __func__)); bzero(&kda, sizeof(kda)); kda.kda_index = KDA_REMOVE_DEV; (void)dumper_remove(nd_conf.ndc_iface, &kda); netdump_mbuf_drain(); if_rele(nd_ifp); nd_ifp = NULL; bzero(&nd_conf, sizeof(nd_conf)); } static void netdump_ifdetach(void *arg __unused, struct ifnet *ifp) { NETDUMP_WLOCK(); if (ifp == nd_ifp) netdump_unconfigure(); NETDUMP_WUNLOCK(); } static int netdump_configure(struct diocskerneldump_arg *conf, struct thread *td) { struct ifnet *ifp; NETDUMP_ASSERT_WLOCKED(); CURVNET_SET(TD_TO_VNET(td)); if (!IS_DEFAULT_VNET(curvnet)) { CURVNET_RESTORE(); return (EINVAL); } ifp = ifunit_ref(conf->kda_iface); CURVNET_RESTORE(); if (ifp == NULL) return (ENOENT); if ((if_getflags(ifp) & IFF_UP) == 0) { if_rele(ifp); return (ENXIO); } if (!netdump_supported_nic(ifp) || ifp->if_type != IFT_ETHER) { if_rele(ifp); return (ENODEV); } if (netdump_enabled()) if_rele(nd_ifp); nd_ifp = ifp; netdump_reinit_internal(ifp); #define COPY_SIZED(elm) do { \ _Static_assert(sizeof(nd_conf.ndc_ ## elm) == \ sizeof(conf->kda_ ## elm), "elm " __XSTRING(elm) " mismatch"); \ memcpy(&nd_conf.ndc_ ## elm, &conf->kda_ ## elm, \ sizeof(nd_conf.ndc_ ## elm)); \ } while (0) COPY_SIZED(iface); COPY_SIZED(server); COPY_SIZED(client); COPY_SIZED(gateway); COPY_SIZED(af); #undef COPY_SIZED return (0); } /* * Reinitialize the mbuf pool used by drivers while dumping. This is called * from the generic ioctl handler for SIOCSIFMTU after any NIC driver has * reconfigured itself. (I.e., it may not be a configured netdump interface.) */ void netdump_reinit(struct ifnet *ifp) { NETDUMP_WLOCK(); if (ifp != nd_ifp) { NETDUMP_WUNLOCK(); return; } netdump_reinit_internal(ifp); NETDUMP_WUNLOCK(); } static void netdump_reinit_internal(struct ifnet *ifp) { int clsize, nmbuf, ncl, nrxr; NETDUMP_ASSERT_WLOCKED(); ifp->if_netdump_methods->nd_init(ifp, &nrxr, &ncl, &clsize); KASSERT(nrxr > 0, ("invalid receive ring count %d", nrxr)); /* * We need two headers per message on the transmit side. Multiply by * four to give us some breathing room. */ nmbuf = ncl * (4 + nrxr); ncl *= nrxr; netdump_mbuf_reinit(nmbuf, ncl, clsize); } /* * ioctl(2) handler for the netdump device. This is currently only used to * register netdump as a dump device. * * Parameters: * dev, Unused. * cmd, The ioctl to be handled. * addr, The parameter for the ioctl. * flags, Unused. * td, The thread invoking this ioctl. * * Returns: * 0 on success, and an errno value on failure. */ static int netdump_ioctl(struct cdev *dev __unused, u_long cmd, caddr_t addr, int flags __unused, struct thread *td) { struct diocskerneldump_arg kda_copy, *conf; struct dumperinfo dumper; uint8_t *encryptedkey; int error; #ifdef COMPAT_FREEBSD11 u_int u; #endif #ifdef COMPAT_FREEBSD12 struct diocskerneldump_arg_freebsd12 *kda12; struct netdump_conf_freebsd12 *conf12; #endif conf = NULL; error = 0; NETDUMP_WLOCK(); switch (cmd) { #ifdef COMPAT_FREEBSD11 case DIOCSKERNELDUMP_FREEBSD11: gone_in(13, "11.x ABI compatibility"); u = *(u_int *)addr; if (u != 0) { error = ENXIO; break; } if (netdump_enabled()) netdump_unconfigure(); break; #endif #ifdef COMPAT_FREEBSD12 /* * Used by dumpon(8) in 12.x for clearing previous * configuration -- then NETDUMPSCONF_FREEBSD12 is used to * actually configure netdump. */ case DIOCSKERNELDUMP_FREEBSD12: gone_in(14, "12.x ABI compatibility"); kda12 = (void *)addr; if (kda12->kda12_enable) { error = ENXIO; break; } if (netdump_enabled()) netdump_unconfigure(); break; case NETDUMPGCONF_FREEBSD12: gone_in(14, "FreeBSD 12.x ABI compat"); conf12 = (void *)addr; if (!netdump_enabled()) { error = ENXIO; break; } if (nd_conf.ndc_af != AF_INET) { error = EOPNOTSUPP; break; } strlcpy(conf12->ndc12_iface, nd_ifp->if_xname, sizeof(conf12->ndc12_iface)); memcpy(&conf12->ndc12_server, &nd_server, sizeof(conf12->ndc12_server)); memcpy(&conf12->ndc12_client, &nd_client, sizeof(conf12->ndc12_client)); memcpy(&conf12->ndc12_gateway, &nd_gateway, sizeof(conf12->ndc12_gateway)); break; #endif case DIOCGKERNELDUMP: conf = (void *)addr; /* * For now, index is ignored; netdump doesn't support multiple * configurations (yet). */ if (!netdump_enabled()) { error = ENXIO; conf = NULL; break; } strlcpy(conf->kda_iface, nd_ifp->if_xname, sizeof(conf->kda_iface)); memcpy(&conf->kda_server, &nd_server, sizeof(nd_server)); memcpy(&conf->kda_client, &nd_client, sizeof(nd_client)); memcpy(&conf->kda_gateway, &nd_gateway, sizeof(nd_gateway)); conf->kda_af = nd_conf.ndc_af; conf = NULL; break; #ifdef COMPAT_FREEBSD12 case NETDUMPSCONF_FREEBSD12: gone_in(14, "FreeBSD 12.x ABI compat"); conf12 = (struct netdump_conf_freebsd12 *)addr; _Static_assert(offsetof(struct diocskerneldump_arg, kda_server) == offsetof(struct netdump_conf_freebsd12, ndc12_server), "simplifying assumption"); memset(&kda_copy, 0, sizeof(kda_copy)); memcpy(&kda_copy, conf12, offsetof(struct diocskerneldump_arg, kda_server)); /* 12.x ABI could only configure IPv4 (INET) netdump. */ kda_copy.kda_af = AF_INET; memcpy(&kda_copy.kda_server.in4, &conf12->ndc12_server, sizeof(struct in_addr)); memcpy(&kda_copy.kda_client.in4, &conf12->ndc12_client, sizeof(struct in_addr)); memcpy(&kda_copy.kda_gateway.in4, &conf12->ndc12_gateway, sizeof(struct in_addr)); kda_copy.kda_index = (conf12->ndc12_kda.kda12_enable ? 0 : KDA_REMOVE_ALL); conf = &kda_copy; explicit_bzero(conf12, sizeof(*conf12)); /* FALLTHROUGH */ #endif case DIOCSKERNELDUMP: encryptedkey = NULL; if (cmd == DIOCSKERNELDUMP) { conf = (void *)addr; memcpy(&kda_copy, conf, sizeof(kda_copy)); } /* Netdump only supports IP4 at this time. */ if (conf->kda_af != AF_INET) { error = EPROTONOSUPPORT; break; } conf->kda_iface[sizeof(conf->kda_iface) - 1] = '\0'; if (conf->kda_index == KDA_REMOVE || conf->kda_index == KDA_REMOVE_DEV || conf->kda_index == KDA_REMOVE_ALL) { if (netdump_enabled()) netdump_unconfigure(); if (conf->kda_index == KDA_REMOVE_ALL) error = dumper_remove(NULL, conf); break; } error = netdump_configure(conf, td); if (error != 0) break; if (conf->kda_encryption != KERNELDUMP_ENC_NONE) { if (conf->kda_encryptedkeysize <= 0 || conf->kda_encryptedkeysize > KERNELDUMP_ENCKEY_MAX_SIZE) { error = EINVAL; break; } encryptedkey = malloc(conf->kda_encryptedkeysize, M_TEMP, M_WAITOK); error = copyin(conf->kda_encryptedkey, encryptedkey, conf->kda_encryptedkeysize); if (error != 0) { free(encryptedkey, M_TEMP); break; } conf->kda_encryptedkey = encryptedkey; } memset(&dumper, 0, sizeof(dumper)); dumper.dumper_start = netdump_start; dumper.dumper_hdr = netdump_write_headers; dumper.dumper = netdump_dumper; dumper.priv = NULL; dumper.blocksize = NETDUMP_DATASIZE; dumper.maxiosize = MAXDUMPPGS * PAGE_SIZE; dumper.mediaoffset = 0; dumper.mediasize = 0; error = dumper_insert(&dumper, conf->kda_iface, conf); if (encryptedkey != NULL) { explicit_bzero(encryptedkey, conf->kda_encryptedkeysize); free(encryptedkey, M_TEMP); } if (error != 0) netdump_unconfigure(); break; default: error = ENOTTY; break; } explicit_bzero(&kda_copy, sizeof(kda_copy)); if (conf != NULL) explicit_bzero(conf, sizeof(*conf)); NETDUMP_WUNLOCK(); return (error); } /* * Called upon system init or kld load. Initializes the netdump parameters to * sane defaults (locates the first available NIC and uses the first IPv4 IP on * that card as the client IP). Leaves the server IP unconfigured. * * Parameters: * mod, Unused. * what, The module event type. * priv, Unused. * * Returns: * int, An errno value if an error occured, 0 otherwise. */ static int netdump_modevent(module_t mod __unused, int what, void *priv __unused) { struct diocskerneldump_arg conf; char *arg; int error; error = 0; switch (what) { case MOD_LOAD: error = make_dev_p(MAKEDEV_WAITOK, &netdump_cdev, &netdump_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "netdump"); if (error != 0) return (error); nd_detach_cookie = EVENTHANDLER_REGISTER(ifnet_departure_event, netdump_ifdetach, NULL, EVENTHANDLER_PRI_ANY); if ((arg = kern_getenv("net.dump.iface")) != NULL) { strlcpy(conf.kda_iface, arg, sizeof(conf.kda_iface)); freeenv(arg); if ((arg = kern_getenv("net.dump.server")) != NULL) { inet_aton(arg, &conf.kda_server.in4); freeenv(arg); } if ((arg = kern_getenv("net.dump.client")) != NULL) { inet_aton(arg, &conf.kda_client.in4); freeenv(arg); } if ((arg = kern_getenv("net.dump.gateway")) != NULL) { inet_aton(arg, &conf.kda_gateway.in4); freeenv(arg); } conf.kda_af = AF_INET; /* Ignore errors; we print a message to the console. */ NETDUMP_WLOCK(); (void)netdump_configure(&conf, curthread); NETDUMP_WUNLOCK(); } break; case MOD_UNLOAD: NETDUMP_WLOCK(); if (netdump_enabled()) { printf("netdump: disabling dump device for unload\n"); netdump_unconfigure(); } NETDUMP_WUNLOCK(); destroy_dev(netdump_cdev); EVENTHANDLER_DEREGISTER(ifnet_departure_event, nd_detach_cookie); break; default: error = EOPNOTSUPP; break; } return (error); } static moduledata_t netdump_mod = { "netdump", netdump_modevent, NULL, }; MODULE_VERSION(netdump, 1); DECLARE_MODULE(netdump, netdump_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);