freebsd-nq/sys/dev/netmap/netmap_monitor.c

499 lines
14 KiB
C
Raw Normal View History

Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
/*
* Copyright (C) 2014 Giuseppe Lettieri. 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.
*/
/*
* $FreeBSD$
*
* Monitors
*
* netmap monitors can be used to do zero-copy monitoring of network traffic
* on another adapter, when the latter adapter is working in netmap mode.
*
* Monitors offer to userspace the same interface as any other netmap port,
* with as many pairs of netmap rings as the monitored adapter.
* However, only the rx rings are actually used. Each monitor rx ring receives
* the traffic transiting on both the tx and rx corresponding rings in the
* monitored adapter. During registration, the user can choose if she wants
* to intercept tx only, rx only, or both tx and rx traffic.
*
* The monitor only sees the frames after they have been consumed in the
* monitored adapter:
*
* - For tx traffic, this is after the slots containing the frames have been
* marked as free. Note that this may happen at a considerably delay after
* frame transmission, since freeing of slots is often done lazily.
*
* - For rx traffic, this is after the consumer on the monitored adapter
* has released them. In most cases, the consumer is a userspace
* application which may have modified the frame contents.
*
* If the monitor is not able to cope with the stream of frames, excess traffic
* will be dropped.
*
* Each ring can be monitored by at most one monitor. This may change in the
* future, if we implement monitor chaining.
*
*/
#if defined(__FreeBSD__)
#include <sys/cdefs.h> /* prerequisite */
#include <sys/types.h>
#include <sys/errno.h>
#include <sys/param.h> /* defines used in kernel.h */
#include <sys/kernel.h> /* types used in module initialization */
#include <sys/malloc.h>
#include <sys/poll.h>
#include <sys/lock.h>
#include <sys/rwlock.h>
#include <sys/selinfo.h>
#include <sys/sysctl.h>
#include <sys/socket.h> /* sockaddrs */
#include <net/if.h>
#include <net/if_var.h>
#include <machine/bus.h> /* bus_dmamap_* */
#include <sys/refcount.h>
#elif defined(linux)
#include "bsd_glue.h"
#elif defined(__APPLE__)
#warning OSX support is only partial
#include "osx_glue.h"
#else
#error Unsupported platform
#endif /* unsupported */
/*
* common headers
*/
#include <net/netmap.h>
#include <dev/netmap/netmap_kern.h>
#include <dev/netmap/netmap_mem2.h>
#ifdef WITH_MONITOR
#define NM_MONITOR_MAXSLOTS 4096
/* monitor works by replacing the nm_sync callbacks in the monitored rings.
* The actions to be performed are the same on both tx and rx rings, so we
* have collected them here
*/
static int
netmap_monitor_parent_sync(struct netmap_kring *kring, int flags, u_int* ringptr)
{
struct netmap_monitor_adapter *mna = kring->monitor;
struct netmap_kring *mkring = &mna->up.rx_rings[kring->ring_id];
struct netmap_ring *ring = kring->ring, *mring = mkring->ring;
int error;
int rel_slots, free_slots, busy;
u_int beg, end, i;
u_int lim = kring->nkr_num_slots - 1,
mlim = mkring->nkr_num_slots - 1;
/* get the relased slots (rel_slots) */
beg = *ringptr;
error = kring->save_sync(kring, flags);
if (error)
return error;
end = *ringptr;
rel_slots = end - beg;
if (rel_slots < 0)
rel_slots += kring->nkr_num_slots;
if (!rel_slots) {
return 0;
}
/* we need to lock the monitor receive ring, since it
* is the target of bot tx and rx traffic from the monitored
* adapter
*/
mtx_lock(&mkring->q_lock);
/* get the free slots available on the monitor ring */
i = mkring->nr_hwtail;
busy = i - mkring->nr_hwcur;
if (busy < 0)
busy += mkring->nkr_num_slots;
free_slots = mlim - busy;
if (!free_slots) {
mtx_unlock(&mkring->q_lock);
return 0;
}
/* swap min(free_slots, rel_slots) slots */
if (free_slots < rel_slots) {
beg += (rel_slots - free_slots);
if (beg > lim)
beg = 0;
rel_slots = free_slots;
}
for ( ; rel_slots; rel_slots--) {
struct netmap_slot *s = &ring->slot[beg];
struct netmap_slot *ms = &mring->slot[i];
uint32_t tmp;
tmp = ms->buf_idx;
ms->buf_idx = s->buf_idx;
s->buf_idx = tmp;
tmp = ms->len;
ms->len = s->len;
s->len = tmp;
s->flags |= NS_BUF_CHANGED;
beg = nm_next(beg, lim);
i = nm_next(i, mlim);
}
2014-11-13 00:14:25 +00:00
mb();
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
mkring->nr_hwtail = i;
mtx_unlock(&mkring->q_lock);
/* notify the new frames to the monitor */
mna->up.nm_notify(&mna->up, mkring->ring_id, NR_RX, 0);
return 0;
}
/* callback used to replace the nm_sync callback in the monitored tx rings */
static int
netmap_monitor_parent_txsync(struct netmap_kring *kring, int flags)
{
ND("%s %x", kring->name, flags);
return netmap_monitor_parent_sync(kring, flags, &kring->nr_hwtail);
}
/* callback used to replace the nm_sync callback in the monitored rx rings */
static int
netmap_monitor_parent_rxsync(struct netmap_kring *kring, int flags)
{
ND("%s %x", kring->name, flags);
return netmap_monitor_parent_sync(kring, flags, &kring->rcur);
}
/* nm_sync callback for the monitor's own tx rings.
* This makes no sense and always returns error
*/
static int
netmap_monitor_txsync(struct netmap_kring *kring, int flags)
{
D("%s %x", kring->name, flags);
return EIO;
}
/* nm_sync callback for the monitor's own rx rings.
* Note that the lock in netmap_monitor_parent_sync only protects
* writers among themselves. Synchronization between writers
* (i.e., netmap_monitor_parent_txsync and netmap_monitor_parent_rxsync)
* and readers (i.e., netmap_monitor_rxsync) relies on memory barriers.
*/
static int
netmap_monitor_rxsync(struct netmap_kring *kring, int flags)
{
ND("%s %x", kring->name, flags);
kring->nr_hwcur = kring->rcur;
2014-11-13 00:14:25 +00:00
mb();
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
nm_rxsync_finalize(kring);
return 0;
}
/* nm_krings_create callbacks for monitors.
* We could use the default netmap_hw_krings_monitor, but
* we don't need the mbq.
*/
static int
netmap_monitor_krings_create(struct netmap_adapter *na)
{
return netmap_krings_create(na, 0);
}
/* nm_register callback for monitors.
*
* On registration, replace the nm_sync callbacks in the monitored
* rings with our own, saving the previous ones in the monitored
* rings themselves, where they are used by netmap_monitor_parent_sync.
*
* On de-registration, restore the original callbacks. We need to
* stop traffic while we are doing this, since the monitored adapter may
* have already started executing a netmap_monitor_parent_sync
* and may not like the kring->save_sync pointer to become NULL.
*/
static int
netmap_monitor_reg(struct netmap_adapter *na, int onoff)
{
struct netmap_monitor_adapter *mna =
(struct netmap_monitor_adapter *)na;
struct netmap_priv_d *priv = &mna->priv;
struct netmap_adapter *pna = priv->np_na;
struct netmap_kring *kring;
int i;
ND("%p: onoff %d", na, onoff);
if (onoff) {
if (!nm_netmap_on(pna)) {
/* parent left netmap mode, fatal */
return ENXIO;
}
if (mna->flags & NR_MONITOR_TX) {
for (i = priv->np_txqfirst; i < priv->np_txqlast; i++) {
kring = &pna->tx_rings[i];
kring->save_sync = kring->nm_sync;
kring->nm_sync = netmap_monitor_parent_txsync;
}
}
if (mna->flags & NR_MONITOR_RX) {
for (i = priv->np_rxqfirst; i < priv->np_rxqlast; i++) {
kring = &pna->rx_rings[i];
kring->save_sync = kring->nm_sync;
kring->nm_sync = netmap_monitor_parent_rxsync;
}
}
na->na_flags |= NAF_NETMAP_ON;
} else {
if (!nm_netmap_on(pna)) {
/* parent left netmap mode, nothing to restore */
return 0;
}
na->na_flags &= ~NAF_NETMAP_ON;
if (mna->flags & NR_MONITOR_TX) {
for (i = priv->np_txqfirst; i < priv->np_txqlast; i++) {
netmap_set_txring(pna, i, 1 /* stopped */);
kring = &pna->tx_rings[i];
kring->nm_sync = kring->save_sync;
kring->save_sync = NULL;
netmap_set_txring(pna, i, 0 /* enabled */);
}
}
if (mna->flags & NR_MONITOR_RX) {
for (i = priv->np_rxqfirst; i < priv->np_rxqlast; i++) {
netmap_set_rxring(pna, i, 1 /* stopped */);
kring = &pna->rx_rings[i];
kring->nm_sync = kring->save_sync;
kring->save_sync = NULL;
netmap_set_rxring(pna, i, 0 /* enabled */);
}
}
}
return 0;
}
/* nm_krings_delete callback for monitors */
static void
netmap_monitor_krings_delete(struct netmap_adapter *na)
{
netmap_krings_delete(na);
}
/* nm_dtor callback for monitors */
static void
netmap_monitor_dtor(struct netmap_adapter *na)
{
struct netmap_monitor_adapter *mna =
(struct netmap_monitor_adapter *)na;
struct netmap_priv_d *priv = &mna->priv;
struct netmap_adapter *pna = priv->np_na;
int i;
ND("%p", na);
if (nm_netmap_on(pna)) {
/* parent still in netmap mode, mark its krings as free */
if (mna->flags & NR_MONITOR_TX) {
for (i = priv->np_txqfirst; i < priv->np_txqlast; i++) {
pna->tx_rings[i].monitor = NULL;
}
}
if (mna->flags & NR_MONITOR_RX) {
for (i = priv->np_rxqfirst; i < priv->np_rxqlast; i++) {
pna->rx_rings[i].monitor = NULL;
}
}
}
netmap_adapter_put(pna);
}
/* check if nmr is a request for a monitor adapter that we can satisfy */
int
netmap_get_monitor_na(struct nmreq *nmr, struct netmap_adapter **na, int create)
{
struct nmreq pnmr;
struct netmap_adapter *pna; /* parent adapter */
struct netmap_monitor_adapter *mna;
int i, error;
if ((nmr->nr_flags & (NR_MONITOR_TX | NR_MONITOR_RX)) == 0) {
ND("not a monitor");
return 0;
}
/* this is a request for a monitor adapter */
D("flags %x", nmr->nr_flags);
mna = malloc(sizeof(*mna), M_DEVBUF, M_NOWAIT | M_ZERO);
if (mna == NULL) {
D("memory error");
return ENOMEM;
}
/* first, try to find the adapter that we want to monitor
* We use the same nmr, after we have turned off the monitor flags.
* In this way we can potentially monitor everything netmap understands,
* except other monitors.
*/
memcpy(&pnmr, nmr, sizeof(pnmr));
pnmr.nr_flags &= ~(NR_MONITOR_TX | NR_MONITOR_RX);
error = netmap_get_na(&pnmr, &pna, create);
if (error) {
D("parent lookup failed: %d", error);
return error;
}
D("found parent: %s", pna->name);
if (!nm_netmap_on(pna)) {
/* parent not in netmap mode */
/* XXX we can wait for the parent to enter netmap mode,
* by intercepting its nm_register callback (2014-03-16)
*/
D("%s not in netmap mode", pna->name);
error = EINVAL;
goto put_out;
}
/* grab all the rings we need in the parent */
mna->priv.np_na = pna;
error = netmap_interp_ringid(&mna->priv, nmr->nr_ringid, nmr->nr_flags);
if (error) {
D("ringid error");
goto put_out;
}
if (nmr->nr_flags & NR_MONITOR_TX) {
for (i = mna->priv.np_txqfirst; i < mna->priv.np_txqlast; i++) {
struct netmap_kring *kring = &pna->tx_rings[i];
if (kring->monitor) {
error = EBUSY;
D("ring busy");
goto release_out;
}
kring->monitor = mna;
}
}
if (nmr->nr_flags & NR_MONITOR_RX) {
for (i = mna->priv.np_rxqfirst; i < mna->priv.np_rxqlast; i++) {
struct netmap_kring *kring = &pna->rx_rings[i];
if (kring->monitor) {
error = EBUSY;
D("ring busy");
goto release_out;
}
kring->monitor = mna;
}
}
snprintf(mna->up.name, sizeof(mna->up.name), "mon:%s", pna->name);
/* the monitor supports the host rings iff the parent does */
mna->up.na_flags = (pna->na_flags & NAF_HOST_RINGS);
mna->up.nm_txsync = netmap_monitor_txsync;
mna->up.nm_rxsync = netmap_monitor_rxsync;
mna->up.nm_register = netmap_monitor_reg;
mna->up.nm_dtor = netmap_monitor_dtor;
mna->up.nm_krings_create = netmap_monitor_krings_create;
mna->up.nm_krings_delete = netmap_monitor_krings_delete;
mna->up.nm_mem = pna->nm_mem;
mna->up.na_lut = pna->na_lut;
mna->up.na_lut_objtotal = pna->na_lut_objtotal;
mna->up.na_lut_objsize = pna->na_lut_objsize;
mna->up.num_tx_rings = 1; // XXX we don't need it, but field can't be zero
/* we set the number of our rx_rings to be max(num_rx_rings, num_rx_rings)
* in the parent
*/
mna->up.num_rx_rings = pna->num_rx_rings;
if (pna->num_tx_rings > pna->num_rx_rings)
mna->up.num_rx_rings = pna->num_tx_rings;
/* by default, the number of slots is the same as in
* the parent rings, but the user may ask for a different
* number
*/
mna->up.num_tx_desc = nmr->nr_tx_slots;
nm_bound_var(&mna->up.num_tx_desc, pna->num_tx_desc,
1, NM_MONITOR_MAXSLOTS, NULL);
mna->up.num_rx_desc = nmr->nr_rx_slots;
nm_bound_var(&mna->up.num_rx_desc, pna->num_rx_desc,
1, NM_MONITOR_MAXSLOTS, NULL);
error = netmap_attach_common(&mna->up);
if (error) {
D("attach_common error");
goto release_out;
}
/* remember the traffic directions we have to monitor */
mna->flags = (nmr->nr_flags & (NR_MONITOR_TX | NR_MONITOR_RX));
*na = &mna->up;
netmap_adapter_get(*na);
/* write the configuration back */
nmr->nr_tx_rings = mna->up.num_tx_rings;
nmr->nr_rx_rings = mna->up.num_rx_rings;
nmr->nr_tx_slots = mna->up.num_tx_desc;
nmr->nr_rx_slots = mna->up.num_rx_desc;
/* keep the reference to the parent */
D("monitor ok");
return 0;
release_out:
D("monitor error");
for (i = mna->priv.np_txqfirst; i < mna->priv.np_txqlast; i++) {
if (pna->tx_rings[i].monitor == mna)
pna->tx_rings[i].monitor = NULL;
}
for (i = mna->priv.np_rxqfirst; i < mna->priv.np_rxqlast; i++) {
if (pna->rx_rings[i].monitor == mna)
pna->rx_rings[i].monitor = NULL;
}
put_out:
netmap_adapter_put(pna);
free(mna, M_DEVBUF);
return error;
}
#endif /* WITH_MONITOR */