d/numam-dpdk
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

kni: add vhost backend

Browse Source 
Attach to vhost-net as raw socket backend.

Signed-off-by: Intel
This commit is contained in:
Intel 2013-09-18 12:00:00 +02:00 committed by Thomas Monjalon
parent 904d29a135
commit b23ffbaa82
10 changed files with 880 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
config/defconfig_i686-default-linuxapp-gcc
View File

@ -277,6 +277,11 @@ CONFIG_RTE_SCHED_PORT_N_GRINDERS=8
CONFIG_RTE_LIBRTE_KNI=y
CONFIG_RTE_LIBRTE_KNI_DEBUG=n
CONFIG_RTE_KNI_KO_DEBUG=n
CONFIG_RTE_KNI_VHOST=n
CONFIG_RTE_KNI_VHOST_MAX_CACHE_SIZE=1024
CONFIG_RTE_KNI_VHOST_VNET_HDR_EN=n
CONFIG_RTE_KNI_VHOST_DEBUG_RX=n
CONFIG_RTE_KNI_VHOST_DEBUG_TX=n
#
# Enable warning directives

5
config/defconfig_i686-default-linuxapp-icc
View File

@ -277,6 +277,11 @@ CONFIG_RTE_SCHED_PORT_N_GRINDERS=8
CONFIG_RTE_LIBRTE_KNI=y
CONFIG_RTE_LIBRTE_KNI_DEBUG=n
CONFIG_RTE_KNI_KO_DEBUG=n
CONFIG_RTE_KNI_VHOST=n
CONFIG_RTE_KNI_VHOST_MAX_CACHE_SIZE=1024
CONFIG_RTE_KNI_VHOST_VNET_HDR_EN=n
CONFIG_RTE_KNI_VHOST_DEBUG_RX=n
CONFIG_RTE_KNI_VHOST_DEBUG_TX=n
#
# Enable warning directives

5
config/defconfig_x86_64-default-linuxapp-gcc
View File

@ -289,6 +289,11 @@ CONFIG_RTE_SCHED_PORT_N_GRINDERS=8
CONFIG_RTE_LIBRTE_KNI=y
CONFIG_RTE_LIBRTE_KNI_DEBUG=n
CONFIG_RTE_KNI_KO_DEBUG=n
CONFIG_RTE_KNI_VHOST=n
CONFIG_RTE_KNI_VHOST_MAX_CACHE_SIZE=1024
CONFIG_RTE_KNI_VHOST_VNET_HDR_EN=n
CONFIG_RTE_KNI_VHOST_DEBUG_RX=n
CONFIG_RTE_KNI_VHOST_DEBUG_TX=n
#
# Enable warning directives

5
config/defconfig_x86_64-default-linuxapp-icc
View File

@ -277,6 +277,11 @@ CONFIG_RTE_SCHED_PORT_N_GRINDERS=8
CONFIG_RTE_LIBRTE_KNI=y
CONFIG_RTE_LIBRTE_KNI_DEBUG=n
CONFIG_RTE_KNI_KO_DEBUG=n
CONFIG_RTE_KNI_VHOST=n
CONFIG_RTE_KNI_VHOST_MAX_CACHE_SIZE=1024
CONFIG_RTE_KNI_VHOST_VNET_HDR_EN=n
CONFIG_RTE_KNI_VHOST_DEBUG_RX=n
CONFIG_RTE_KNI_VHOST_DEBUG_TX=n
#
# Enable warning directives

1
lib/librte_eal/linuxapp/kni/Makefile
View File

@ -79,5 +79,6 @@ SRCS-y += ethtool/igb/igb_vmdq.c
SRCS-y += kni_misc.c
SRCS-y += kni_net.c
SRCS-y += kni_ethtool.c
SRCS-$(CONFIG_RTE_KNI_VHOST) += kni_vhost.c
include $(RTE_SDK)/mk/rte.module.mk

47
lib/librte_eal/linuxapp/kni/kni_dev.h
View File

@ -32,6 +32,10 @@
#include <linux/spinlock.h>
#include <linux/list.h>
#ifdef RTE_KNI_VHOST
#include <net/sock.h>
#endif
#include <exec-env/rte_kni_common.h>
#define KNI_KTHREAD_RESCHEDULE_INTERVAL 5 /* us */
@ -91,8 +95,16 @@ struct kni_dev {
/* synchro for request processing */
unsigned long synchro;
};
#ifdef RTE_KNI_VHOST
struct kni_vhost_queue* vhost_queue;
volatile enum {
BE_STOP = 0x1,
BE_START = 0x2,
BE_FINISH = 0x4,
}vq_status;
#endif
};
#define KNI_ERR(args...) printk(KERN_DEBUG "KNI: Error: " args)
#define KNI_PRINT(args...) printk(KERN_DEBUG "KNI: " args)
@ -102,4 +114,37 @@ struct kni_dev {
#define KNI_DBG(args...)
#endif
#ifdef RTE_KNI_VHOST
unsigned int
kni_poll(struct file *file, struct socket *sock, poll_table * wait);
int kni_chk_vhost_rx(struct kni_dev *kni);
int kni_vhost_init(struct kni_dev *kni);
int kni_vhost_backend_release(struct kni_dev *kni);
struct kni_vhost_queue {
struct sock sk;
struct socket *sock;
int vnet_hdr_sz;
struct kni_dev *kni;
int sockfd;
unsigned int flags;
struct sk_buff* cache;
struct rte_kni_fifo* fifo;
};
#endif
#ifdef RTE_KNI_VHOST_DEBUG_RX
#define KNI_DBG_RX(args...) printk(KERN_DEBUG "KNI RX: " args)
#else
#define KNI_DBG_RX(args...)
#endif
#ifdef RTE_KNI_VHOST_DEBUG_TX
#define KNI_DBG_TX(args...) printk(KERN_DEBUG "KNI TX: " args)
#else
#define KNI_DBG_TX(args...)
#endif
#endif

14
lib/librte_eal/linuxapp/kni/kni_fifo.h
View File

@ -91,4 +91,18 @@ kni_fifo_free_count(struct rte_kni_fifo *fifo)
return (fifo->read - fifo->write - 1) & (fifo->len - 1);
}
#ifdef RTE_KNI_VHOST
/**
* Initializes the kni fifo structure
*/
static inline void
kni_fifo_init(struct rte_kni_fifo *fifo, unsigned size)
{
fifo->write = 0;
fifo->read = 0;
fifo->len = size;
fifo->elem_size = sizeof(void *);
}
#endif
#endif /* _KNI_FIFO_H_ */

22
lib/librte_eal/linuxapp/kni/kni_misc.c
View File

@ -193,6 +193,9 @@ kni_release(struct inode *inode, struct file *file)
dev->pthread = NULL;
}
#ifdef RTE_KNI_VHOST
kni_vhost_backend_release(dev);
#endif
kni_dev_remove(dev);
list_del(&dev->list);
}
@ -217,7 +220,11 @@ kni_thread_single(void *unused)
for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
list_for_each_entry_safe(dev, n,
&kni_list_head, list) {
#ifdef RTE_KNI_VHOST
kni_chk_vhost_rx(dev);
#else
kni_net_rx(dev);
#endif
kni_net_poll_resp(dev);
}
}
@ -238,7 +245,11 @@ kni_thread_multiple(void *param)
while (!kthread_should_stop()) {
for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
#ifdef RTE_KNI_VHOST
kni_chk_vhost_rx(dev);
#else
kni_net_rx(dev);
#endif
kni_net_poll_resp(dev);
}
schedule_timeout_interruptible(usecs_to_jiffies( \
@ -361,6 +372,10 @@ kni_ioctl_create(unsigned int ioctl_num, unsigned long ioctl_param)
kni->mbuf_kva = phys_to_virt(dev_info.mbuf_phys);
kni->mbuf_va = dev_info.mbuf_va;
#ifdef RTE_KNI_VHOST
kni->vhost_queue = NULL;
kni->vq_status = BE_STOP;
#endif
kni->mbuf_size = dev_info.mbuf_size;
KNI_PRINT("tx_phys: 0x%016llx, tx_q addr: 0x%p\n",
@ -443,6 +458,10 @@ kni_ioctl_create(unsigned int ioctl_num, unsigned long ioctl_param)
return -ENODEV;
}
#ifdef RTE_KNI_VHOST
kni_vhost_init(kni);
#endif
/**
* Create a new kernel thread for multiple mode, set its core affinity,
* and finally wake it up.
@ -497,6 +516,9 @@ kni_ioctl_release(unsigned int ioctl_num, unsigned long ioctl_param)
dev->pthread = NULL;
}
#ifdef RTE_KNI_VHOST
kni_vhost_backend_release(dev);
#endif
kni_dev_remove(dev);
list_del(&dev->list);
ret = 0;

13
lib/librte_eal/linuxapp/kni/kni_net.c
View File

@ -379,6 +379,18 @@ kni_net_rx(struct kni_dev *kni)
/*
* Transmit a packet (called by the kernel)
*/
#ifdef RTE_KNI_VHOST
static int
kni_net_tx(struct sk_buff *skb, struct net_device *dev)
{
struct kni_dev *kni = netdev_priv(dev);
dev_kfree_skb(skb);
kni->stats.tx_dropped++;
return NETDEV_TX_OK;
}
#else
static int
kni_net_tx(struct sk_buff *skb, struct net_device *dev)
{
@ -451,6 +463,7 @@ kni_net_tx(struct sk_buff *skb, struct net_device *dev)
return NETDEV_TX_OK;
}
#endif
/*
* Deal with a transmit timeout.

764
lib/librte_eal/linuxapp/kni/kni_vhost.c Normal file
View File

@ -0,0 +1,764 @@
/*-
* GPL LICENSE SUMMARY
*
* Copyright(c) 2010-2013 Intel Corporation. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of version 2 of the GNU General Public License as
* published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
* The full GNU General Public License is included in this distribution
* in the file called LICENSE.GPL.
*
* Contact Information:
* Intel Corporation
*/
#include <linux/module.h>
#include <linux/net.h>
#include <net/sock.h>
#include <linux/virtio_net.h>
#include <linux/wait.h>
#include <linux/mm.h>
#include <linux/nsproxy.h>
#include <linux/sched.h>
#include <linux/if_tun.h>
#include "kni_dev.h"
#include "kni_fifo.h"
#define RX_BURST_SZ 4
extern void put_unused_fd(unsigned int fd);
static struct proto kni_raw_proto = {
.name = "kni_vhost",
.owner = THIS_MODULE,
.obj_size = sizeof(struct kni_vhost_queue),
};
static inline int
kni_vhost_net_tx(struct kni_dev *kni, struct iovec *iov,
unsigned offset, unsigned len)
{
struct rte_kni_mbuf *pkt_kva = NULL;
struct rte_kni_mbuf *pkt_va = NULL;
int ret;
KNI_DBG_TX("tx offset=%d, len=%d, iovlen=%d\n",
offset, len, (int)iov->iov_len);
/**
* Check if it has at least one free entry in tx_q and
* one entry in alloc_q.
*/
if (kni_fifo_free_count(kni->tx_q) == 0 ||
kni_fifo_count(kni->alloc_q) == 0) {
/**
* If no free entry in tx_q or no entry in alloc_q,
* drops skb and goes out.
*/
goto drop;
}
/* dequeue a mbuf from alloc_q */
ret = kni_fifo_get(kni->alloc_q, (void **)&pkt_va, 1);
if (likely(ret == 1)) {
void *data_kva;
pkt_kva = (void *)pkt_va - kni->mbuf_va + kni->mbuf_kva;
data_kva = pkt_kva->data - kni->mbuf_va + kni->mbuf_kva;
memcpy_fromiovecend(data_kva, iov, offset, len);
if (unlikely(len < ETH_ZLEN)) {
memset(data_kva + len, 0, ETH_ZLEN - len);
len = ETH_ZLEN;
}
pkt_kva->pkt_len = len;
pkt_kva->data_len = len;
/* enqueue mbuf into tx_q */
ret = kni_fifo_put(kni->tx_q, (void **)&pkt_va, 1);
if (unlikely(ret != 1)) {
/* Failing should not happen */
KNI_ERR("Fail to enqueue mbuf into tx_q\n");
goto drop;
}
} else {
/* Failing should not happen */
KNI_ERR("Fail to dequeue mbuf from alloc_q\n");
goto drop;
}
/* update statistics */
kni->stats.tx_bytes += len;
kni->stats.tx_packets++;
return 0;
drop:
/* update statistics */
kni->stats.tx_dropped++;
return 0;
}
static inline int
kni_vhost_net_rx(struct kni_dev *kni, struct iovec *iov,
unsigned offset, unsigned len)
{
uint32_t pkt_len;
struct rte_kni_mbuf *kva;
struct rte_kni_mbuf *va;
void * data_kva;
struct sk_buff *skb;
struct kni_vhost_queue *q = kni->vhost_queue;
if (unlikely(q == NULL))
return 0;
/* ensure at least one entry in free_q */
if (unlikely(kni_fifo_free_count(kni->free_q) == 0))
return 0;
skb = skb_dequeue(&q->sk.sk_receive_queue);
if (unlikely(skb == NULL))
return 0;
kva = (struct rte_kni_mbuf*)skb->data;
/* free skb to cache */
skb->data = NULL;
if (unlikely(1 != kni_fifo_put(q->fifo, (void **)&skb, 1)))
/* Failing should not happen */
KNI_ERR("Fail to enqueue entries into rx cache fifo\n");
pkt_len = kva->data_len;
if (unlikely(pkt_len > len))
goto drop;
KNI_DBG_RX("rx offset=%d, len=%d, pkt_len=%d, iovlen=%d\n",
offset, len, pkt_len, (int)iov->iov_len);
data_kva = kva->data - kni->mbuf_va + kni->mbuf_kva;
if (unlikely(memcpy_toiovecend(iov, data_kva, offset, pkt_len)))
goto drop;
/* Update statistics */
kni->stats.rx_bytes += pkt_len;
kni->stats.rx_packets++;
/* enqueue mbufs into free_q */
va = (void*)kva - kni->mbuf_kva + kni->mbuf_va;
if (unlikely(1 != kni_fifo_put(kni->free_q, (void **)&va, 1)))
/* Failing should not happen */
KNI_ERR("Fail to enqueue entries into free_q\n");
KNI_DBG_RX("receive done %d\n", pkt_len);
return pkt_len;
drop:
/* Update drop statistics */
kni->stats.rx_dropped++;
return 0;
}
static unsigned int
kni_sock_poll(struct file *file, struct socket *sock, poll_table * wait)
{
struct kni_vhost_queue *q =
container_of(sock->sk, struct kni_vhost_queue, sk);
struct kni_dev *kni;
unsigned int mask = 0;
if (unlikely(q == NULL || q->kni == NULL))
return POLLERR;
kni = q->kni;
KNI_DBG("start kni_poll on group %d, wq 0x%16llx\n",
kni->group_id, (uint64_t)sock->wq);
poll_wait(file, &sock->wq->wait, wait);
if (kni_fifo_count(kni->rx_q) > 0)
mask |= POLLIN | POLLRDNORM;
if (sock_writeable(&q->sk) ||
(!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock->flags) &&
sock_writeable(&q->sk)))
mask |= POLLOUT | POLLWRNORM;
return mask;
}
static inline void
kni_vhost_enqueue(struct kni_dev *kni, struct kni_vhost_queue *q,
struct sk_buff *skb, struct rte_kni_mbuf *va)
{
struct rte_kni_mbuf *kva;
kva = (void *)(va) - kni->mbuf_va + kni->mbuf_kva;
(skb)->data = (unsigned char*)kva;
(skb)->len = kva->data_len;
skb_queue_tail(&q->sk.sk_receive_queue, skb);
}
static inline void
kni_vhost_enqueue_burst(struct kni_dev *kni, struct kni_vhost_queue *q,
struct sk_buff **skb, struct rte_kni_mbuf **va)
{
int i;
for (i = 0; i < RX_BURST_SZ; skb++, va++, i++)
kni_vhost_enqueue(kni, q, *skb, *va);
}
int
kni_chk_vhost_rx(struct kni_dev *kni)
{
struct kni_vhost_queue *q = kni->vhost_queue;
unsigned nb_in, nb_mbuf, nb_skb;
const unsigned BURST_MASK = RX_BURST_SZ - 1;
unsigned nb_burst, nb_backlog, i;
struct sk_buff *skb[RX_BURST_SZ];
struct rte_kni_mbuf *va[RX_BURST_SZ];
if (unlikely(BE_STOP & kni->vq_status)) {
kni->vq_status |= BE_FINISH;
return 0;
}
if (unlikely(q == NULL))
return 0;
nb_skb = kni_fifo_count(q->fifo);
nb_mbuf = kni_fifo_count(kni->rx_q);
nb_in = min(nb_mbuf, nb_skb);
nb_in = min(nb_in, (unsigned)RX_BURST_SZ);
nb_burst = (nb_in & ~BURST_MASK);
nb_backlog = (nb_in & BURST_MASK);
/* enqueue skb_queue per BURST_SIZE bulk */
if (0 != nb_burst) {
if (unlikely(RX_BURST_SZ != kni_fifo_get(
kni->rx_q, (void **)&va,
RX_BURST_SZ)))
goto except;
if (unlikely(RX_BURST_SZ != kni_fifo_get(
q->fifo, (void **)&skb,
RX_BURST_SZ)))
goto except;
kni_vhost_enqueue_burst(kni, q, skb, va);
}
/* all leftover, do one by one */
for (i = 0; i < nb_backlog; ++i) {
if (unlikely(1 != kni_fifo_get(
kni->rx_q,(void **)&va, 1)))
goto except;
if (unlikely(1 != kni_fifo_get(
q->fifo, (void **)&skb, 1)))
goto except;
kni_vhost_enqueue(kni, q, *skb, *va);
}
/* Ondemand wake up */
if ((nb_in == RX_BURST_SZ) || (nb_skb == 0) ||
((nb_mbuf < RX_BURST_SZ) && (nb_mbuf != 0))) {
wake_up_interruptible_poll(sk_sleep(&q->sk),
POLLIN | POLLRDNORM | POLLRDBAND);
KNI_DBG_RX("RX CHK KICK nb_mbuf %d, nb_skb %d, nb_in %d\n",
nb_mbuf, nb_skb, nb_in);
}
return 0;
except:
/* Failing should not happen */
KNI_ERR("Fail to enqueue fifo, it shouldn't happen \n");
BUG_ON(1);
return 0;
}
static int
kni_sock_sndmsg(struct kiocb *iocb, struct socket *sock,
struct msghdr *m, size_t total_len)
{
struct kni_vhost_queue *q =
container_of(sock->sk, struct kni_vhost_queue, sk);
int vnet_hdr_len = 0;
unsigned long len = total_len;
if (unlikely(q == NULL || q->kni == NULL))
return 0;
KNI_DBG_TX("kni_sndmsg len %ld, flags 0x%08x, nb_iov %d\n",
len, q->flags, (int)m->msg_iovlen);
#ifdef RTE_KNI_VHOST_VNET_HDR_EN
if (likely(q->flags & IFF_VNET_HDR)) {
vnet_hdr_len = q->vnet_hdr_sz;
if (unlikely(len < vnet_hdr_len))
return -EINVAL;
len -= vnet_hdr_len;
}
#endif
if (unlikely(len < ETH_HLEN + q->vnet_hdr_sz))
return -EINVAL;
return kni_vhost_net_tx(q->kni, m->msg_iov, vnet_hdr_len, len);
}
static int
kni_sock_rcvmsg(struct kiocb *iocb, struct socket *sock,
struct msghdr *m, size_t len, int flags)
{
int vnet_hdr_len = 0;
int pkt_len = 0;
struct kni_vhost_queue *q =
container_of(sock->sk, struct kni_vhost_queue, sk);
static struct virtio_net_hdr
__attribute__ ((unused)) vnet_hdr = {
.flags = 0,
.gso_type = VIRTIO_NET_HDR_GSO_NONE
};
if (unlikely(q == NULL || q->kni == NULL))
return 0;
#ifdef RTE_KNI_VHOST_VNET_HDR_EN
if (likely(q->flags & IFF_VNET_HDR)) {
vnet_hdr_len = q->vnet_hdr_sz;
if ((len -= vnet_hdr_len) < 0)
return -EINVAL;
}
#endif
if (unlikely(0 == (pkt_len = kni_vhost_net_rx(q->kni,
m->msg_iov, vnet_hdr_len, len))))
return 0;
#ifdef RTE_KNI_VHOST_VNET_HDR_EN
/* no need to copy hdr when no pkt received */
if (unlikely(memcpy_toiovecend(m->msg_iov,
(void *)&vnet_hdr, 0, vnet_hdr_len)))
return -EFAULT;
#endif
KNI_DBG_RX("kni_rcvmsg expect_len %ld, flags 0x%08x, pkt_len %d\n",
(unsigned long)len, q->flags, pkt_len);
return (pkt_len + vnet_hdr_len);
}
/* dummy tap like ioctl */
static int
kni_sock_ioctl(struct socket *sock, unsigned int cmd,
unsigned long arg)
{
void __user *argp = (void __user *)arg;
struct ifreq __user *ifr = argp;
unsigned int __user *up = argp;
struct kni_vhost_queue *q =
container_of(sock->sk, struct kni_vhost_queue, sk);
struct kni_dev *kni;
unsigned int u;
int __user *sp = argp;
int s;
int ret;
KNI_DBG("tap ioctl cmd 0x%08x\n", cmd);
switch (cmd) {
case TUNSETIFF:
KNI_DBG("TUNSETIFF\n");
/* ignore the name, just look at flags */
if (get_user(u, &ifr->ifr_flags))
return -EFAULT;
ret = 0;
if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
ret = -EINVAL;
else
q->flags = u;
return ret;
case TUNGETIFF:
KNI_DBG("TUNGETIFF\n");
rcu_read_lock_bh();
kni = rcu_dereference_bh(q->kni);
if (kni)
dev_hold(kni->net_dev);
rcu_read_unlock_bh();
if (!kni)
return -ENOLINK;
ret = 0;
if (copy_to_user(&ifr->ifr_name, kni->net_dev->name, IFNAMSIZ) ||
put_user(q->flags, &ifr->ifr_flags))
ret = -EFAULT;
dev_put(kni->net_dev);
return ret;
case TUNGETFEATURES:
KNI_DBG("TUNGETFEATURES\n");
u = IFF_TAP | IFF_NO_PI;
#ifdef RTE_KNI_VHOST_VNET_HDR_EN
u |= IFF_VNET_HDR;
#endif
if (put_user(u, up))
return -EFAULT;
return 0;
case TUNSETSNDBUF:
KNI_DBG("TUNSETSNDBUF\n");
if (get_user(u, up))
return -EFAULT;
q->sk.sk_sndbuf = u;
return 0;
case TUNGETVNETHDRSZ:
s = q->vnet_hdr_sz;
if (put_user(s, sp))
return -EFAULT;
KNI_DBG("TUNGETVNETHDRSZ %d\n", s);
return 0;
case TUNSETVNETHDRSZ:
if (get_user(s, sp))
return -EFAULT;
if (s < (int)sizeof(struct virtio_net_hdr))
return -EINVAL;
KNI_DBG("TUNSETVNETHDRSZ %d\n", s);
q->vnet_hdr_sz = s;
return 0;
case TUNSETOFFLOAD:
KNI_DBG("TUNSETOFFLOAD %lx\n", arg);
#ifdef RTE_KNI_VHOST_VNET_HDR_EN
/* not support any offload yet */
if (!(q->flags & IFF_VNET_HDR))
return -EINVAL;
return 0;
#else
return -EINVAL;
#endif
default:
KNI_DBG("NOT SUPPORT\n");
return -EINVAL;
}
}
static int
kni_sock_compat_ioctl(struct socket *sock, unsigned int cmd,
unsigned long arg)
{
/* 32 bits app on 64 bits OS to be supported later */
KNI_PRINT("Not implemented.\n");
return -EINVAL;
}
#define KNI_VHOST_WAIT_WQ_SAFE() \
do { \
while ((BE_FINISH | BE_STOP) == kni->vq_status) \
msleep(1); \
}while(0) \
static int
kni_sock_release(struct socket *sock)
{
struct kni_vhost_queue *q =
container_of(sock->sk, struct kni_vhost_queue, sk);
struct kni_dev *kni;
if (q == NULL)
return 0;
if (NULL != (kni = q->kni)) {
kni->vq_status = BE_STOP;
KNI_VHOST_WAIT_WQ_SAFE();
kni->vhost_queue = NULL;
q->kni = NULL;
}
if (q->sockfd != -1)
q->sockfd = -1;
sk_set_socket(&q->sk, NULL);
sock->sk = NULL;
sock_put(&q->sk);
KNI_DBG("dummy sock release done\n");
return 0;
}
int
kni_sock_getname (struct socket *sock,
struct sockaddr *addr,
int *sockaddr_len, int peer)
{
KNI_DBG("dummy sock getname\n");
((struct sockaddr_ll*)addr)->sll_family = AF_PACKET;
return 0;
}
static const struct proto_ops kni_socket_ops = {
.getname = kni_sock_getname,
.sendmsg = kni_sock_sndmsg,
.recvmsg = kni_sock_rcvmsg,
.release = kni_sock_release,
.poll = kni_sock_poll,
.ioctl = kni_sock_ioctl,
.compat_ioctl = kni_sock_compat_ioctl,
};
static void
kni_sk_write_space(struct sock *sk)
{
wait_queue_head_t *wqueue;
if (!sock_writeable(sk) ||
!test_and_clear_bit(SOCK_ASYNC_NOSPACE,
&sk->sk_socket->flags))
return;
wqueue = sk_sleep(sk);
if (wqueue && waitqueue_active(wqueue))
wake_up_interruptible_poll(
wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
}
static void
kni_sk_destruct(struct sock *sk)
{
struct kni_vhost_queue *q =
container_of(sk, struct kni_vhost_queue, sk);
if (!q)
return;
/* make sure there's no packet in buffer */
while (skb_dequeue(&sk->sk_receive_queue) != NULL)
;
mb();
if (q->fifo != NULL) {
kfree(q->fifo);
q->fifo = NULL;
}
if (q->cache != NULL) {
kfree(q->cache);
q->cache = NULL;
}
}
static int
kni_vhost_backend_init(struct kni_dev *kni)
{
struct kni_vhost_queue *q;
struct net *net = current->nsproxy->net_ns;
int err, i, sockfd;
struct rte_kni_fifo *fifo;
struct sk_buff *elem;
if (kni->vhost_queue != NULL)
return -1;
if (!(q = (struct kni_vhost_queue *)sk_alloc(
net, AF_UNSPEC, GFP_KERNEL, &kni_raw_proto)))
return -ENOMEM;
err = sock_create_lite(AF_UNSPEC, SOCK_RAW, IPPROTO_RAW, &q->sock);
if (err)
goto free_sk;
sockfd = sock_map_fd(q->sock, 0);
if (sockfd < 0) {
err = sockfd;
goto free_sock;
}
/* cache init */
q->cache = (struct sk_buff*)
kzalloc(RTE_KNI_VHOST_MAX_CACHE_SIZE * sizeof(struct sk_buff),
GFP_KERNEL);
if (!q->cache)
goto free_fd;
fifo = (struct rte_kni_fifo*)
kzalloc(RTE_KNI_VHOST_MAX_CACHE_SIZE * sizeof(void *)
+ sizeof(struct rte_kni_fifo), GFP_KERNEL);
if (!fifo)
goto free_cache;
kni_fifo_init(fifo, RTE_KNI_VHOST_MAX_CACHE_SIZE);
for (i = 0; i < RTE_KNI_VHOST_MAX_CACHE_SIZE; i++) {
elem = &q->cache[i];
kni_fifo_put(fifo, (void**)&elem, 1);
}
q->fifo = fifo;
/* store sockfd in vhost_queue */
q->sockfd = sockfd;
/* init socket */
q->sock->type = SOCK_RAW;
q->sock->state = SS_CONNECTED;
q->sock->ops = &kni_socket_ops;
sock_init_data(q->sock, &q->sk);
/* init sock data */
q->sk.sk_write_space = kni_sk_write_space;
q->sk.sk_destruct = kni_sk_destruct;
q->flags = IFF_NO_PI | IFF_TAP;
q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
#ifdef RTE_KNI_VHOST_VNET_HDR_EN
q->flags |= IFF_VNET_HDR;
#endif
/* bind kni_dev with vhost_queue */
q->kni = kni;
kni->vhost_queue = q;
wmb();
kni->vq_status = BE_START;
KNI_DBG("backend init sockfd=%d, sock->wq=0x%16llx,"
"sk->sk_wq=0x%16llx",
q->sockfd, (uint64_t)q->sock->wq,
(uint64_t)q->sk.sk_wq);
return 0;
free_cache:
kfree(q->cache);
q->cache = NULL;
free_fd:
put_unused_fd(sockfd);
free_sock:
q->kni = NULL;
kni->vhost_queue = NULL;
kni->vq_status |= BE_FINISH;
sock_release(q->sock);
q->sock->ops = NULL;
q->sock = NULL;
free_sk:
sk_free((struct sock*)q);
return err;
}
/* kni vhost sock sysfs */
static ssize_t
show_sock_fd(struct device *dev, struct device_attribute *attr,
char *buf)
{
struct net_device *net_dev = container_of(dev, struct net_device, dev);
struct kni_dev *kni = netdev_priv(net_dev);
int sockfd = -1;
if (kni->vhost_queue != NULL)
sockfd = kni->vhost_queue->sockfd;
return snprintf(buf, 10, "%d\n", sockfd);
}
static ssize_t
show_sock_en(struct device *dev, struct device_attribute *attr,
char *buf)
{
struct net_device *net_dev = container_of(dev, struct net_device, dev);
struct kni_dev *kni = netdev_priv(net_dev);
return snprintf(buf, 10, "%u\n", (kni->vhost_queue == NULL ? 0 : 1));
}
static ssize_t
set_sock_en(struct device *dev, struct device_attribute *attr,
const char *buf, size_t count)
{
struct net_device *net_dev = container_of(dev, struct net_device, dev);
struct kni_dev *kni = netdev_priv(net_dev);
unsigned long en;
int err = 0;
if (0 != strict_strtoul(buf, 0, &en))
return -EINVAL;
if (en)
err = kni_vhost_backend_init(kni);
return err ? err : count;
}
static DEVICE_ATTR(sock_fd, S_IRUGO | S_IRUSR, show_sock_fd, NULL);
static DEVICE_ATTR(sock_en, S_IRUGO | S_IWUSR, show_sock_en, set_sock_en);
static struct attribute *dev_attrs[] = {
&dev_attr_sock_fd.attr,
&dev_attr_sock_en.attr,
NULL,
};
static const struct attribute_group dev_attr_grp = {
.attrs = dev_attrs,
};
int
kni_vhost_backend_release(struct kni_dev *kni)
{
struct kni_vhost_queue *q = kni->vhost_queue;
if (q == NULL)
return 0;
/* dettach from kni */
q->kni = NULL;
KNI_DBG("release backend done\n");
return 0;
}
int
kni_vhost_init(struct kni_dev *kni)
{
struct net_device *dev = kni->net_dev;
if (sysfs_create_group(&dev->dev.kobj, &dev_attr_grp))
sysfs_remove_group(&dev->dev.kobj, &dev_attr_grp);
kni->vq_status = BE_STOP;
KNI_DBG("kni_vhost_init done\n");
return 0;
}