2 * Copyright (C) 2013-2014 Universita` di Pisa. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/types.h>
29 #include <sys/module.h>
30 #include <sys/errno.h>
31 #include <sys/param.h> /* defines used in kernel.h */
32 #include <sys/poll.h> /* POLLIN, POLLOUT */
33 #include <sys/kernel.h> /* types used in module initialization */
34 #include <sys/conf.h> /* DEV_MODULE */
35 #include <sys/endian.h>
37 #include <sys/rwlock.h>
39 #include <vm/vm.h> /* vtophys */
40 #include <vm/pmap.h> /* vtophys */
41 #include <vm/vm_param.h>
42 #include <vm/vm_object.h>
43 #include <vm/vm_page.h>
44 #include <vm/vm_pager.h>
48 #include <sys/malloc.h>
49 #include <sys/socket.h> /* sockaddrs */
50 #include <sys/selinfo.h>
52 #include <net/if_var.h>
53 #include <machine/bus.h> /* bus_dmamap_* */
54 #include <netinet/in.h> /* in6_cksum_pseudo() */
55 #include <machine/in_cksum.h> /* in_pseudo(), in_cksum_hdr() */
57 #include <net/netmap.h>
58 #include <dev/netmap/netmap_kern.h>
59 #include <dev/netmap/netmap_mem2.h>
62 /* ======================== FREEBSD-SPECIFIC ROUTINES ================== */
65 nm_csum_raw(uint8_t *data, size_t len, rawsum_t cur_sum)
67 /* TODO XXX please use the FreeBSD implementation for this. */
68 uint16_t *words = (uint16_t *)data;
72 for (i = 0; i < nw; i++)
73 cur_sum += be16toh(words[i]);
76 cur_sum += (data[len-1] << 8);
81 /* Fold a raw checksum: 'cur_sum' is in host byte order, while the
82 * return value is in network byte order.
85 nm_csum_fold(rawsum_t cur_sum)
87 /* TODO XXX please use the FreeBSD implementation for this. */
89 cur_sum = (cur_sum & 0xFFFF) + (cur_sum >> 16);
91 return htobe16((~cur_sum) & 0xFFFF);
95 nm_csum_ipv4(struct nm_iphdr *iph)
98 return in_cksum_hdr((void *)iph);
100 return nm_csum_fold(nm_csum_raw((uint8_t*)iph, sizeof(struct nm_iphdr), 0));
105 nm_csum_tcpudp_ipv4(struct nm_iphdr *iph, void *data,
106 size_t datalen, uint16_t *check)
109 uint16_t pseudolen = datalen + iph->protocol;
111 /* Compute and insert the pseudo-header cheksum. */
112 *check = in_pseudo(iph->saddr, iph->daddr,
114 /* Compute the checksum on TCP/UDP header + payload
115 * (includes the pseudo-header).
117 *check = nm_csum_fold(nm_csum_raw(data, datalen, 0));
119 static int notsupported = 0;
122 D("inet4 segmentation not supported");
128 nm_csum_tcpudp_ipv6(struct nm_ipv6hdr *ip6h, void *data,
129 size_t datalen, uint16_t *check)
132 *check = in6_cksum_pseudo((void*)ip6h, datalen, ip6h->nexthdr, 0);
133 *check = nm_csum_fold(nm_csum_raw(data, datalen, 0));
135 static int notsupported = 0;
138 D("inet6 segmentation not supported");
145 * Intercept the rx routine in the standard device driver.
146 * Second argument is non-zero to intercept, 0 to restore
149 netmap_catch_rx(struct netmap_adapter *na, int intercept)
151 struct netmap_generic_adapter *gna =
152 (struct netmap_generic_adapter *)na;
153 struct ifnet *ifp = na->ifp;
156 if (gna->save_if_input) {
157 D("cannot intercept again");
158 return EINVAL; /* already set */
160 gna->save_if_input = ifp->if_input;
161 ifp->if_input = generic_rx_handler;
163 if (!gna->save_if_input){
165 return EINVAL; /* not saved */
167 ifp->if_input = gna->save_if_input;
168 gna->save_if_input = NULL;
176 * Intercept the packet steering routine in the tx path,
177 * so that we can decide which queue is used for an mbuf.
178 * Second argument is non-zero to intercept, 0 to restore.
179 * On freebsd we just intercept if_transmit.
182 netmap_catch_tx(struct netmap_generic_adapter *gna, int enable)
184 struct netmap_adapter *na = &gna->up.up;
185 struct ifnet *ifp = na->ifp;
188 na->if_transmit = ifp->if_transmit;
189 ifp->if_transmit = netmap_transmit;
191 ifp->if_transmit = na->if_transmit;
197 * Transmit routine used by generic_netmap_txsync(). Returns 0 on success
198 * and non-zero on error (which may be packet drops or other errors).
199 * addr and len identify the netmap buffer, m is the (preallocated)
200 * mbuf to use for transmissions.
202 * We should add a reference to the mbuf so the m_freem() at the end
203 * of the transmission does not consume resources.
205 * On FreeBSD, and on multiqueue cards, we can force the queue using
206 * if ((m->m_flags & M_FLOWID) != 0)
207 * i = m->m_pkthdr.flowid % adapter->num_queues;
209 * i = curcpu % adapter->num_queues;
213 generic_xmit_frame(struct ifnet *ifp, struct mbuf *m,
214 void *addr, u_int len, u_int ring_nr)
219 * The mbuf should be a cluster from our special pool,
220 * so we do not need to do an m_copyback but just copy
221 * (and eventually, just reference the netmap buffer)
224 if (*m->m_ext.ref_cnt != 1) {
225 D("invalid refcnt %d for %p",
226 *m->m_ext.ref_cnt, m);
227 panic("in generic_xmit_frame");
229 // XXX the ext_size check is unnecessary if we link the netmap buf
230 if (m->m_ext.ext_size < len) {
231 RD(5, "size %d < len %d", m->m_ext.ext_size, len);
232 len = m->m_ext.ext_size;
234 if (0) { /* XXX seems to have negligible benefits */
235 m->m_ext.ext_buf = m->m_data = addr;
237 bcopy(addr, m->m_data, len);
239 m->m_len = m->m_pkthdr.len = len;
240 // inc refcount. All ours, we could skip the atomic
241 atomic_fetchadd_int(m->m_ext.ref_cnt, 1);
242 m->m_flags |= M_FLOWID;
243 m->m_pkthdr.flowid = ring_nr;
244 m->m_pkthdr.rcvif = ifp; /* used for tx notification */
245 ret = NA(ifp)->if_transmit(ifp, m);
250 #if __FreeBSD_version >= 1100005
251 struct netmap_adapter *
252 netmap_getna(if_t ifp)
254 return (NA((struct ifnet *)ifp));
256 #endif /* __FreeBSD_version >= 1100005 */
259 * The following two functions are empty until we have a generic
260 * way to extract the info from the ifp
263 generic_find_num_desc(struct ifnet *ifp, unsigned int *tx, unsigned int *rx)
265 D("called, in tx %d rx %d", *tx, *rx);
271 generic_find_num_queues(struct ifnet *ifp, u_int *txq, u_int *rxq)
273 D("called, in txq %d rxq %d", *txq, *rxq);
274 *txq = netmap_generic_rings;
275 *rxq = netmap_generic_rings;
280 netmap_mitigation_init(struct nm_generic_mit *mit, struct netmap_adapter *na)
283 mit->mit_pending = 0;
289 netmap_mitigation_start(struct nm_generic_mit *mit)
296 netmap_mitigation_restart(struct nm_generic_mit *mit)
303 netmap_mitigation_active(struct nm_generic_mit *mit)
311 netmap_mitigation_cleanup(struct nm_generic_mit *mit)
318 * In order to track whether pages are still mapped, we hook into
319 * the standard cdev_pager and intercept the constructor and
323 struct netmap_vm_handle_t {
325 struct netmap_priv_d *priv;
330 netmap_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
331 vm_ooffset_t foff, struct ucred *cred, u_short *color)
333 struct netmap_vm_handle_t *vmh = handle;
336 D("handle %p size %jd prot %d foff %jd",
337 handle, (intmax_t)size, prot, (intmax_t)foff);
344 netmap_dev_pager_dtor(void *handle)
346 struct netmap_vm_handle_t *vmh = handle;
347 struct cdev *dev = vmh->dev;
348 struct netmap_priv_d *priv = vmh->priv;
351 D("handle %p", handle);
359 netmap_dev_pager_fault(vm_object_t object, vm_ooffset_t offset,
360 int prot, vm_page_t *mres)
362 struct netmap_vm_handle_t *vmh = object->handle;
363 struct netmap_priv_d *priv = vmh->priv;
366 vm_memattr_t memattr;
369 ND("object %p offset %jd prot %d mres %p",
370 object, (intmax_t)offset, prot, mres);
371 memattr = object->memattr;
372 pidx = OFF_TO_IDX(offset);
373 paddr = netmap_mem_ofstophys(priv->np_mref, offset);
375 return VM_PAGER_FAIL;
377 if (((*mres)->flags & PG_FICTITIOUS) != 0) {
379 * If the passed in result page is a fake page, update it with
380 * the new physical address.
383 vm_page_updatefake(page, paddr, memattr);
386 * Replace the passed in reqpage page with our own fake page and
387 * free up the all of the original pages.
389 #ifndef VM_OBJECT_WUNLOCK /* FreeBSD < 10.x */
390 #define VM_OBJECT_WUNLOCK VM_OBJECT_UNLOCK
391 #define VM_OBJECT_WLOCK VM_OBJECT_LOCK
392 #endif /* VM_OBJECT_WUNLOCK */
394 VM_OBJECT_WUNLOCK(object);
395 page = vm_page_getfake(paddr, memattr);
396 VM_OBJECT_WLOCK(object);
399 vm_page_unlock(*mres);
401 vm_page_insert(page, object, pidx);
403 page->valid = VM_PAGE_BITS_ALL;
404 return (VM_PAGER_OK);
408 static struct cdev_pager_ops netmap_cdev_pager_ops = {
409 .cdev_pg_ctor = netmap_dev_pager_ctor,
410 .cdev_pg_dtor = netmap_dev_pager_dtor,
411 .cdev_pg_fault = netmap_dev_pager_fault,
416 netmap_mmap_single(struct cdev *cdev, vm_ooffset_t *foff,
417 vm_size_t objsize, vm_object_t *objp, int prot)
420 struct netmap_vm_handle_t *vmh;
421 struct netmap_priv_d *priv;
425 D("cdev %p foff %jd size %jd objp %p prot %d", cdev,
426 (intmax_t )*foff, (intmax_t )objsize, objp, prot);
428 vmh = malloc(sizeof(struct netmap_vm_handle_t), M_DEVBUF,
435 error = devfs_get_cdevpriv((void**)&priv);
442 error = netmap_get_memory(priv);
446 obj = cdev_pager_allocate(vmh, OBJT_DEVICE,
447 &netmap_cdev_pager_ops, objsize, prot,
450 D("cdev_pager_allocate failed");
469 // XXX can we remove this ?
471 netmap_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
474 D("dev %p fflag 0x%x devtype %d td %p",
475 dev, fflag, devtype, td);
481 netmap_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
483 struct netmap_priv_d *priv;
491 // XXX wait or nowait ?
492 priv = malloc(sizeof(struct netmap_priv_d), M_DEVBUF,
497 error = devfs_set_cdevpriv(priv, netmap_dtor);
501 priv->np_refcount = 1;
506 /******************** kqueue support ****************/
509 * The OS_selwakeup also needs to issue a KNOTE_UNLOCKED.
510 * We use a non-zero argument to distinguish the call from the one
511 * in kevent_scan() which instead also needs to run netmap_poll().
512 * The knote uses a global mutex for the time being. We might
513 * try to reuse the one in the si, but it is not allocated
514 * permanently so it might be a bit tricky.
516 * The *kqfilter function registers one or another f_event
517 * depending on read or write mode.
518 * In the call to f_event() td_fpop is NULL so any child function
519 * calling devfs_get_cdevpriv() would fail - and we need it in
520 * netmap_poll(). As a workaround we store priv into kn->kn_hook
521 * and pass it as first argument to netmap_poll(), which then
522 * uses the failure to tell that we are called from f_event()
523 * and do not need the selrecord().
526 void freebsd_selwakeup(struct selinfo *si, int pri);
529 freebsd_selwakeup(struct selinfo *si, int pri)
532 D("on knote %p", &si->si_note);
533 selwakeuppri(si, pri);
534 /* use a non-zero hint to tell the notification from the
535 * call done in kqueue_scan() which uses 0
537 KNOTE_UNLOCKED(&si->si_note, 0x100 /* notification */);
541 netmap_knrdetach(struct knote *kn)
543 struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook;
544 struct selinfo *si = priv->np_rxsi;
546 D("remove selinfo %p", si);
547 knlist_remove(&si->si_note, kn, 0);
551 netmap_knwdetach(struct knote *kn)
553 struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook;
554 struct selinfo *si = priv->np_txsi;
556 D("remove selinfo %p", si);
557 knlist_remove(&si->si_note, kn, 0);
561 * callback from notifies (generated externally) and our
562 * calls to kevent(). The former we just return 1 (ready)
563 * since we do not know better.
564 * In the latter we call netmap_poll and return 0/1 accordingly.
567 netmap_knrw(struct knote *kn, long hint, int events)
569 struct netmap_priv_d *priv;
573 ND(5, "call from notify");
574 return 1; /* assume we are ready */
577 /* the notification may come from an external thread,
578 * in which case we do not want to run the netmap_poll
579 * This should be filtered above, but check just in case.
581 if (curthread != priv->np_td) { /* should not happen */
582 RD(5, "curthread changed %p %p", curthread, priv->np_td);
585 revents = netmap_poll((void *)priv, events, curthread);
586 return (events & revents) ? 1 : 0;
591 netmap_knread(struct knote *kn, long hint)
593 return netmap_knrw(kn, hint, POLLIN);
597 netmap_knwrite(struct knote *kn, long hint)
599 return netmap_knrw(kn, hint, POLLOUT);
602 static struct filterops netmap_rfiltops = {
604 .f_detach = netmap_knrdetach,
605 .f_event = netmap_knread,
608 static struct filterops netmap_wfiltops = {
610 .f_detach = netmap_knwdetach,
611 .f_event = netmap_knwrite,
616 * This is called when a thread invokes kevent() to record
617 * a change in the configuration of the kqueue().
618 * The 'priv' should be the same as in the netmap device.
621 netmap_kqfilter(struct cdev *dev, struct knote *kn)
623 struct netmap_priv_d *priv;
625 struct netmap_adapter *na;
627 int ev = kn->kn_filter;
629 if (ev != EVFILT_READ && ev != EVFILT_WRITE) {
630 D("bad filter request %d", ev);
633 error = devfs_get_cdevpriv((void**)&priv);
635 D("device not yet setup");
640 D("no netmap adapter for this file descriptor");
643 /* the si is indicated in the priv */
644 si = (ev == EVFILT_WRITE) ? priv->np_txsi : priv->np_rxsi;
646 kn->kn_fop = (ev == EVFILT_WRITE) ?
647 &netmap_wfiltops : &netmap_rfiltops;
649 knlist_add(&si->si_note, kn, 1);
651 ND("register %p %s td %p priv %p kn %p np_nifp %p kn_fp/fpop %s",
652 na, na->ifp->if_xname, curthread, priv, kn,
654 kn->kn_fp == curthread->td_fpop ? "match" : "MISMATCH");
658 struct cdevsw netmap_cdevsw = {
659 .d_version = D_VERSION,
661 .d_open = netmap_open,
662 .d_mmap_single = netmap_mmap_single,
663 .d_ioctl = netmap_ioctl,
664 .d_poll = netmap_poll,
665 .d_kqfilter = netmap_kqfilter,
666 .d_close = netmap_close,
668 /*--- end of kqueue support ----*/
671 * Kernel entry point.
673 * Initialize/finalize the module and return.
675 * Return 0 on success, errno on failure.
678 netmap_loader(__unused struct module *module, int event, __unused void *arg)
684 error = netmap_init();
700 DEV_MODULE(netmap, netmap_loader, NULL);