4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of version 2 of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 * The full GNU General Public License is included in this distribution
19 * in the file called LICENSE.GPL.
21 * Contact Information:
25 #include <linux/module.h>
26 #include <linux/net.h>
28 #include <linux/virtio_net.h>
29 #include <linux/wait.h>
31 #include <linux/nsproxy.h>
32 #include <linux/sched.h>
33 #include <linux/if_tun.h>
34 #include <linux/version.h>
42 extern void put_unused_fd(unsigned int fd);
44 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,7,0)
46 sock_alloc_file(struct socket *sock,
47 int flags, const char *dname);
49 extern int get_unused_fd_flags(unsigned flags);
51 extern void fd_install(unsigned int fd, struct file *file);
53 static int kni_sock_map_fd(struct socket *sock)
56 int fd = get_unused_fd_flags(0);
60 file = sock_alloc_file(sock, 0, NULL);
69 #define kni_sock_map_fd(s) sock_map_fd(s, 0)
72 static struct proto kni_raw_proto = {
75 .obj_size = sizeof(struct kni_vhost_queue),
79 kni_vhost_net_tx(struct kni_dev *kni, struct msghdr *m,
80 unsigned offset, unsigned len)
82 struct rte_kni_mbuf *pkt_kva = NULL;
83 struct rte_kni_mbuf *pkt_va = NULL;
86 KNI_DBG_TX("tx offset=%d, len=%d, iovlen=%d\n",
87 #ifdef HAVE_IOV_ITER_MSGHDR
88 offset, len, (int)m->msg_iter.iov->iov_len);
90 offset, len, (int)m->msg_iov->iov_len);
94 * Check if it has at least one free entry in tx_q and
95 * one entry in alloc_q.
97 if (kni_fifo_free_count(kni->tx_q) == 0 ||
98 kni_fifo_count(kni->alloc_q) == 0) {
100 * If no free entry in tx_q or no entry in alloc_q,
101 * drops skb and goes out.
106 /* dequeue a mbuf from alloc_q */
107 ret = kni_fifo_get(kni->alloc_q, (void **)&pkt_va, 1);
108 if (likely(ret == 1)) {
111 pkt_kva = (void *)pkt_va - kni->mbuf_va + kni->mbuf_kva;
112 data_kva = pkt_kva->buf_addr + pkt_kva->data_off
113 - kni->mbuf_va + kni->mbuf_kva;
115 #ifdef HAVE_IOV_ITER_MSGHDR
116 copy_from_iter(data_kva, len, &m->msg_iter);
118 memcpy_fromiovecend(data_kva, m->msg_iov, offset, len);
121 if (unlikely(len < ETH_ZLEN)) {
122 memset(data_kva + len, 0, ETH_ZLEN - len);
125 pkt_kva->pkt_len = len;
126 pkt_kva->data_len = len;
128 /* enqueue mbuf into tx_q */
129 ret = kni_fifo_put(kni->tx_q, (void **)&pkt_va, 1);
130 if (unlikely(ret != 1)) {
131 /* Failing should not happen */
132 KNI_ERR("Fail to enqueue mbuf into tx_q\n");
136 /* Failing should not happen */
137 KNI_ERR("Fail to dequeue mbuf from alloc_q\n");
141 /* update statistics */
142 kni->stats.tx_bytes += len;
143 kni->stats.tx_packets++;
148 /* update statistics */
149 kni->stats.tx_dropped++;
155 kni_vhost_net_rx(struct kni_dev *kni, struct msghdr *m,
156 unsigned offset, unsigned len)
159 struct rte_kni_mbuf *kva;
160 struct rte_kni_mbuf *va;
163 struct kni_vhost_queue *q = kni->vhost_queue;
165 if (unlikely(q == NULL))
168 /* ensure at least one entry in free_q */
169 if (unlikely(kni_fifo_free_count(kni->free_q) == 0))
172 skb = skb_dequeue(&q->sk.sk_receive_queue);
173 if (unlikely(skb == NULL))
176 kva = (struct rte_kni_mbuf*)skb->data;
178 /* free skb to cache */
180 if (unlikely(1 != kni_fifo_put(q->fifo, (void **)&skb, 1)))
181 /* Failing should not happen */
182 KNI_ERR("Fail to enqueue entries into rx cache fifo\n");
184 pkt_len = kva->data_len;
185 if (unlikely(pkt_len > len))
188 KNI_DBG_RX("rx offset=%d, len=%d, pkt_len=%d, iovlen=%d\n",
189 #ifdef HAVE_IOV_ITER_MSGHDR
190 offset, len, pkt_len, (int)m->msg_iter.iov->iov_len);
192 offset, len, pkt_len, (int)m->msg_iov->iov_len);
195 data_kva = kva->buf_addr + kva->data_off - kni->mbuf_va + kni->mbuf_kva;
196 #ifdef HAVE_IOV_ITER_MSGHDR
197 if (unlikely(copy_to_iter(data_kva, pkt_len, &m->msg_iter)))
199 if (unlikely(memcpy_toiovecend(m->msg_iov, data_kva, offset, pkt_len)))
203 /* Update statistics */
204 kni->stats.rx_bytes += pkt_len;
205 kni->stats.rx_packets++;
207 /* enqueue mbufs into free_q */
208 va = (void*)kva - kni->mbuf_kva + kni->mbuf_va;
209 if (unlikely(1 != kni_fifo_put(kni->free_q, (void **)&va, 1)))
210 /* Failing should not happen */
211 KNI_ERR("Fail to enqueue entries into free_q\n");
213 KNI_DBG_RX("receive done %d\n", pkt_len);
218 /* Update drop statistics */
219 kni->stats.rx_dropped++;
225 kni_sock_poll(struct file *file, struct socket *sock, poll_table * wait)
227 struct kni_vhost_queue *q =
228 container_of(sock->sk, struct kni_vhost_queue, sk);
230 unsigned int mask = 0;
232 if (unlikely(q == NULL || q->kni == NULL))
236 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,35)
237 KNI_DBG("start kni_poll on group %d, wq 0x%16llx\n",
238 kni->group_id, (uint64_t)sock->wq);
240 KNI_DBG("start kni_poll on group %d, wait at 0x%16llx\n",
241 kni->group_id, (uint64_t)&sock->wait);
244 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,35)
245 poll_wait(file, &sock->wq->wait, wait);
247 poll_wait(file, &sock->wait, wait);
250 if (kni_fifo_count(kni->rx_q) > 0)
251 mask |= POLLIN | POLLRDNORM;
253 if (sock_writeable(&q->sk) ||
254 #ifdef SOCKWQ_ASYNC_NOSPACE
255 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock->flags) &&
257 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock->flags) &&
259 sock_writeable(&q->sk)))
260 mask |= POLLOUT | POLLWRNORM;
266 kni_vhost_enqueue(struct kni_dev *kni, struct kni_vhost_queue *q,
267 struct sk_buff *skb, struct rte_kni_mbuf *va)
269 struct rte_kni_mbuf *kva;
271 kva = (void *)(va) - kni->mbuf_va + kni->mbuf_kva;
272 (skb)->data = (unsigned char*)kva;
273 (skb)->len = kva->data_len;
274 skb_queue_tail(&q->sk.sk_receive_queue, skb);
278 kni_vhost_enqueue_burst(struct kni_dev *kni, struct kni_vhost_queue *q,
279 struct sk_buff **skb, struct rte_kni_mbuf **va)
282 for (i = 0; i < RX_BURST_SZ; skb++, va++, i++)
283 kni_vhost_enqueue(kni, q, *skb, *va);
287 kni_chk_vhost_rx(struct kni_dev *kni)
289 struct kni_vhost_queue *q = kni->vhost_queue;
290 unsigned nb_in, nb_mbuf, nb_skb;
291 const unsigned BURST_MASK = RX_BURST_SZ - 1;
292 unsigned nb_burst, nb_backlog, i;
293 struct sk_buff *skb[RX_BURST_SZ];
294 struct rte_kni_mbuf *va[RX_BURST_SZ];
296 if (unlikely(BE_STOP & kni->vq_status)) {
297 kni->vq_status |= BE_FINISH;
301 if (unlikely(q == NULL))
304 nb_skb = kni_fifo_count(q->fifo);
305 nb_mbuf = kni_fifo_count(kni->rx_q);
307 nb_in = min(nb_mbuf, nb_skb);
308 nb_in = min(nb_in, (unsigned)RX_BURST_SZ);
309 nb_burst = (nb_in & ~BURST_MASK);
310 nb_backlog = (nb_in & BURST_MASK);
312 /* enqueue skb_queue per BURST_SIZE bulk */
314 if (unlikely(RX_BURST_SZ != kni_fifo_get(
315 kni->rx_q, (void **)&va,
319 if (unlikely(RX_BURST_SZ != kni_fifo_get(
320 q->fifo, (void **)&skb,
324 kni_vhost_enqueue_burst(kni, q, skb, va);
327 /* all leftover, do one by one */
328 for (i = 0; i < nb_backlog; ++i) {
329 if (unlikely(1 != kni_fifo_get(
330 kni->rx_q,(void **)&va, 1)))
333 if (unlikely(1 != kni_fifo_get(
334 q->fifo, (void **)&skb, 1)))
337 kni_vhost_enqueue(kni, q, *skb, *va);
340 /* Ondemand wake up */
341 if ((nb_in == RX_BURST_SZ) || (nb_skb == 0) ||
342 ((nb_mbuf < RX_BURST_SZ) && (nb_mbuf != 0))) {
343 wake_up_interruptible_poll(sk_sleep(&q->sk),
344 POLLIN | POLLRDNORM | POLLRDBAND);
345 KNI_DBG_RX("RX CHK KICK nb_mbuf %d, nb_skb %d, nb_in %d\n",
346 nb_mbuf, nb_skb, nb_in);
352 /* Failing should not happen */
353 KNI_ERR("Fail to enqueue fifo, it shouldn't happen \n");
360 #ifdef HAVE_KIOCB_MSG_PARAM
361 kni_sock_sndmsg(struct kiocb *iocb, struct socket *sock,
362 struct msghdr *m, size_t total_len)
364 kni_sock_sndmsg(struct socket *sock,
365 struct msghdr *m, size_t total_len)
366 #endif /* HAVE_KIOCB_MSG_PARAM */
368 struct kni_vhost_queue *q =
369 container_of(sock->sk, struct kni_vhost_queue, sk);
370 int vnet_hdr_len = 0;
371 unsigned long len = total_len;
373 if (unlikely(q == NULL || q->kni == NULL))
376 KNI_DBG_TX("kni_sndmsg len %ld, flags 0x%08x, nb_iov %d\n",
377 #ifdef HAVE_IOV_ITER_MSGHDR
378 len, q->flags, (int)m->msg_iter.iov->iov_len);
380 len, q->flags, (int)m->msg_iovlen);
383 #ifdef RTE_KNI_VHOST_VNET_HDR_EN
384 if (likely(q->flags & IFF_VNET_HDR)) {
385 vnet_hdr_len = q->vnet_hdr_sz;
386 if (unlikely(len < vnet_hdr_len))
392 if (unlikely(len < ETH_HLEN + q->vnet_hdr_sz))
395 return kni_vhost_net_tx(q->kni, m, vnet_hdr_len, len);
399 #ifdef HAVE_KIOCB_MSG_PARAM
400 kni_sock_rcvmsg(struct kiocb *iocb, struct socket *sock,
401 struct msghdr *m, size_t len, int flags)
403 kni_sock_rcvmsg(struct socket *sock,
404 struct msghdr *m, size_t len, int flags)
405 #endif /* HAVE_KIOCB_MSG_PARAM */
407 int vnet_hdr_len = 0;
409 struct kni_vhost_queue *q =
410 container_of(sock->sk, struct kni_vhost_queue, sk);
411 static struct virtio_net_hdr
412 __attribute__ ((unused)) vnet_hdr = {
414 .gso_type = VIRTIO_NET_HDR_GSO_NONE
417 if (unlikely(q == NULL || q->kni == NULL))
420 #ifdef RTE_KNI_VHOST_VNET_HDR_EN
421 if (likely(q->flags & IFF_VNET_HDR)) {
422 vnet_hdr_len = q->vnet_hdr_sz;
423 if ((len -= vnet_hdr_len) < 0)
428 if (unlikely(0 == (pkt_len = kni_vhost_net_rx(q->kni,
429 m, vnet_hdr_len, len))))
432 #ifdef RTE_KNI_VHOST_VNET_HDR_EN
433 /* no need to copy hdr when no pkt received */
434 #ifdef HAVE_IOV_ITER_MSGHDR
435 if (unlikely(copy_to_iter((void *)&vnet_hdr, vnet_hdr_len,
438 if (unlikely(memcpy_toiovecend(m->msg_iov,
439 (void *)&vnet_hdr, 0, vnet_hdr_len)))
440 #endif /* HAVE_IOV_ITER_MSGHDR */
442 #endif /* RTE_KNI_VHOST_VNET_HDR_EN */
443 KNI_DBG_RX("kni_rcvmsg expect_len %ld, flags 0x%08x, pkt_len %d\n",
444 (unsigned long)len, q->flags, pkt_len);
446 return pkt_len + vnet_hdr_len;
449 /* dummy tap like ioctl */
451 kni_sock_ioctl(struct socket *sock, unsigned int cmd,
454 void __user *argp = (void __user *)arg;
455 struct ifreq __user *ifr = argp;
456 unsigned int __user *up = argp;
457 struct kni_vhost_queue *q =
458 container_of(sock->sk, struct kni_vhost_queue, sk);
461 int __user *sp = argp;
465 KNI_DBG("tap ioctl cmd 0x%08x\n", cmd);
469 KNI_DBG("TUNSETIFF\n");
470 /* ignore the name, just look at flags */
471 if (get_user(u, &ifr->ifr_flags))
475 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
483 KNI_DBG("TUNGETIFF\n");
485 kni = rcu_dereference_bh(q->kni);
487 dev_hold(kni->net_dev);
488 rcu_read_unlock_bh();
494 if (copy_to_user(&ifr->ifr_name, kni->net_dev->name, IFNAMSIZ) ||
495 put_user(q->flags, &ifr->ifr_flags))
497 dev_put(kni->net_dev);
501 KNI_DBG("TUNGETFEATURES\n");
502 u = IFF_TAP | IFF_NO_PI;
503 #ifdef RTE_KNI_VHOST_VNET_HDR_EN
511 KNI_DBG("TUNSETSNDBUF\n");
518 case TUNGETVNETHDRSZ:
522 KNI_DBG("TUNGETVNETHDRSZ %d\n", s);
525 case TUNSETVNETHDRSZ:
528 if (s < (int)sizeof(struct virtio_net_hdr))
531 KNI_DBG("TUNSETVNETHDRSZ %d\n", s);
536 KNI_DBG("TUNSETOFFLOAD %lx\n", arg);
537 #ifdef RTE_KNI_VHOST_VNET_HDR_EN
538 /* not support any offload yet */
539 if (!(q->flags & IFF_VNET_HDR))
548 KNI_DBG("NOT SUPPORT\n");
554 kni_sock_compat_ioctl(struct socket *sock, unsigned int cmd,
557 /* 32 bits app on 64 bits OS to be supported later */
558 KNI_PRINT("Not implemented.\n");
563 #define KNI_VHOST_WAIT_WQ_SAFE() \
565 while ((BE_FINISH | BE_STOP) == kni->vq_status) \
571 kni_sock_release(struct socket *sock)
573 struct kni_vhost_queue *q =
574 container_of(sock->sk, struct kni_vhost_queue, sk);
580 if (NULL != (kni = q->kni)) {
581 kni->vq_status = BE_STOP;
582 KNI_VHOST_WAIT_WQ_SAFE();
583 kni->vhost_queue = NULL;
590 sk_set_socket(&q->sk, NULL);
595 KNI_DBG("dummy sock release done\n");
601 kni_sock_getname (struct socket *sock,
602 struct sockaddr *addr,
603 int *sockaddr_len, int peer)
605 KNI_DBG("dummy sock getname\n");
606 ((struct sockaddr_ll*)addr)->sll_family = AF_PACKET;
610 static const struct proto_ops kni_socket_ops = {
611 .getname = kni_sock_getname,
612 .sendmsg = kni_sock_sndmsg,
613 .recvmsg = kni_sock_rcvmsg,
614 .release = kni_sock_release,
615 .poll = kni_sock_poll,
616 .ioctl = kni_sock_ioctl,
617 .compat_ioctl = kni_sock_compat_ioctl,
621 kni_sk_write_space(struct sock *sk)
623 wait_queue_head_t *wqueue;
625 if (!sock_writeable(sk) ||
626 #ifdef SOCKWQ_ASYNC_NOSPACE
627 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
629 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
632 wqueue = sk_sleep(sk);
633 if (wqueue && waitqueue_active(wqueue))
634 wake_up_interruptible_poll(
635 wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
639 kni_sk_destruct(struct sock *sk)
641 struct kni_vhost_queue *q =
642 container_of(sk, struct kni_vhost_queue, sk);
647 /* make sure there's no packet in buffer */
648 while (skb_dequeue(&sk->sk_receive_queue) != NULL)
653 if (q->fifo != NULL) {
658 if (q->cache != NULL) {
665 kni_vhost_backend_init(struct kni_dev *kni)
667 struct kni_vhost_queue *q;
668 struct net *net = current->nsproxy->net_ns;
670 struct rte_kni_fifo *fifo;
671 struct sk_buff *elem;
673 if (kni->vhost_queue != NULL)
676 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 2, 0)
677 q = (struct kni_vhost_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
680 q = (struct kni_vhost_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
686 err = sock_create_lite(AF_UNSPEC, SOCK_RAW, IPPROTO_RAW, &q->sock);
690 sockfd = kni_sock_map_fd(q->sock);
697 q->cache = kzalloc(RTE_KNI_VHOST_MAX_CACHE_SIZE * sizeof(struct sk_buff),
702 fifo = kzalloc(RTE_KNI_VHOST_MAX_CACHE_SIZE * sizeof(void *)
703 + sizeof(struct rte_kni_fifo), GFP_KERNEL);
707 kni_fifo_init(fifo, RTE_KNI_VHOST_MAX_CACHE_SIZE);
709 for (i = 0; i < RTE_KNI_VHOST_MAX_CACHE_SIZE; i++) {
711 kni_fifo_put(fifo, (void**)&elem, 1);
715 /* store sockfd in vhost_queue */
719 q->sock->type = SOCK_RAW;
720 q->sock->state = SS_CONNECTED;
721 q->sock->ops = &kni_socket_ops;
722 sock_init_data(q->sock, &q->sk);
725 q->sk.sk_write_space = kni_sk_write_space;
726 q->sk.sk_destruct = kni_sk_destruct;
727 q->flags = IFF_NO_PI | IFF_TAP;
728 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
729 #ifdef RTE_KNI_VHOST_VNET_HDR_EN
730 q->flags |= IFF_VNET_HDR;
733 /* bind kni_dev with vhost_queue */
735 kni->vhost_queue = q;
739 kni->vq_status = BE_START;
741 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,35)
742 KNI_DBG("backend init sockfd=%d, sock->wq=0x%16llx,"
743 "sk->sk_wq=0x%16llx",
744 q->sockfd, (uint64_t)q->sock->wq,
745 (uint64_t)q->sk.sk_wq);
747 KNI_DBG("backend init sockfd=%d, sock->wait at 0x%16llx,"
748 "sk->sk_sleep=0x%16llx",
749 q->sockfd, (uint64_t)&q->sock->wait,
750 (uint64_t)q->sk.sk_sleep);
760 put_unused_fd(sockfd);
764 kni->vhost_queue = NULL;
765 kni->vq_status |= BE_FINISH;
766 sock_release(q->sock);
771 sk_free((struct sock*)q);
776 /* kni vhost sock sysfs */
778 show_sock_fd(struct device *dev, struct device_attribute *attr,
781 struct net_device *net_dev = container_of(dev, struct net_device, dev);
782 struct kni_dev *kni = netdev_priv(net_dev);
784 if (kni->vhost_queue != NULL)
785 sockfd = kni->vhost_queue->sockfd;
786 return snprintf(buf, 10, "%d\n", sockfd);
790 show_sock_en(struct device *dev, struct device_attribute *attr,
793 struct net_device *net_dev = container_of(dev, struct net_device, dev);
794 struct kni_dev *kni = netdev_priv(net_dev);
795 return snprintf(buf, 10, "%u\n", (kni->vhost_queue == NULL ? 0 : 1));
799 set_sock_en(struct device *dev, struct device_attribute *attr,
800 const char *buf, size_t count)
802 struct net_device *net_dev = container_of(dev, struct net_device, dev);
803 struct kni_dev *kni = netdev_priv(net_dev);
807 if (0 != kstrtoul(buf, 0, &en))
811 err = kni_vhost_backend_init(kni);
813 return err ? err : count;
816 static DEVICE_ATTR(sock_fd, S_IRUGO | S_IRUSR, show_sock_fd, NULL);
817 static DEVICE_ATTR(sock_en, S_IRUGO | S_IWUSR, show_sock_en, set_sock_en);
818 static struct attribute *dev_attrs[] = {
819 &dev_attr_sock_fd.attr,
820 &dev_attr_sock_en.attr,
824 static const struct attribute_group dev_attr_grp = {
829 kni_vhost_backend_release(struct kni_dev *kni)
831 struct kni_vhost_queue *q = kni->vhost_queue;
836 /* dettach from kni */
839 KNI_DBG("release backend done\n");
845 kni_vhost_init(struct kni_dev *kni)
847 struct net_device *dev = kni->net_dev;
849 if (sysfs_create_group(&dev->dev.kobj, &dev_attr_grp))
850 sysfs_remove_group(&dev->dev.kobj, &dev_attr_grp);
852 kni->vq_status = BE_STOP;
854 KNI_DBG("kni_vhost_init done\n");