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
|
|
|
/*
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
* Copyright (C) 2014-2016 Giuseppe Lettieri
|
|
|
|
* All rights reserved.
|
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
|
|
|
*
|
|
|
|
* 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
|
|
|
|
*
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
* netmap monitors can be used to do monitoring of network traffic
|
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
|
|
|
* 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.
|
|
|
|
*
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
* If the monitor is not able to cope with the stream of frames, excess traffic
|
|
|
|
* will be dropped.
|
|
|
|
*
|
|
|
|
* If the monitored adapter leaves netmap mode, the monitor has to be restarted.
|
|
|
|
*
|
|
|
|
* Monitors can be either zero-copy or copy-based.
|
|
|
|
*
|
|
|
|
* Copy monitors see the frames before they are consumed:
|
|
|
|
*
|
|
|
|
* - For tx traffic, this is when the application sends them, before they are
|
|
|
|
* passed down to the adapter.
|
|
|
|
*
|
|
|
|
* - For rx traffic, this is when they are received by the adapter, before
|
|
|
|
* they are sent up to the application, if any (note that, if no
|
|
|
|
* application is reading from a monitored ring, the ring will eventually
|
|
|
|
* fill up and traffic will stop).
|
|
|
|
*
|
|
|
|
* Zero-copy monitors only see the frames after they have been consumed:
|
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
|
|
|
*
|
|
|
|
* - 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.
|
|
|
|
*
|
2018-04-09 09:24:26 +00:00
|
|
|
* Several copy or zero-copy monitors may be active on any ring.
|
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
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
|
|
#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"
|
|
|
|
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
#elif defined(_WIN32)
|
|
|
|
#include "win_glue.h"
|
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
|
|
|
#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
|
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
/*
|
|
|
|
********************************************************************
|
|
|
|
* functions common to both kind of monitors
|
|
|
|
********************************************************************
|
|
|
|
*/
|
|
|
|
|
2017-06-12 22:53:18 +00:00
|
|
|
static int netmap_zmon_reg(struct netmap_adapter *, int);
|
|
|
|
static int
|
|
|
|
nm_is_zmon(struct netmap_adapter *na)
|
|
|
|
{
|
|
|
|
return na->nm_register == netmap_zmon_reg;
|
|
|
|
}
|
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
/* 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)
|
|
|
|
{
|
2018-05-18 03:38:17 +00:00
|
|
|
RD(1, "%s %x", kring->name, flags);
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
return EIO;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* nm_sync callback for the monitor's own rx rings.
|
|
|
|
* Note that the lock in netmap_zmon_parent_sync only protects
|
|
|
|
* writers among themselves. Synchronization between writers
|
|
|
|
* (i.e., netmap_zmon_parent_txsync and netmap_zmon_parent_rxsync)
|
|
|
|
* and readers (i.e., netmap_zmon_rxsync) relies on memory barriers.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
netmap_monitor_rxsync(struct netmap_kring *kring, int flags)
|
|
|
|
{
|
2018-05-18 03:38:17 +00:00
|
|
|
ND("%s %x", kring->name, flags);
|
2017-06-12 22:53:18 +00:00
|
|
|
kring->nr_hwcur = kring->rhead;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
mb();
|
2018-05-18 03:38:17 +00:00
|
|
|
return 0;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* nm_krings_create callbacks for monitors.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
netmap_monitor_krings_create(struct netmap_adapter *na)
|
|
|
|
{
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
int error = netmap_krings_create(na, 0);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
/* override the host rings callbacks */
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
na->tx_rings[na->num_tx_rings]->nm_sync = netmap_monitor_txsync;
|
|
|
|
na->rx_rings[na->num_rx_rings]->nm_sync = netmap_monitor_rxsync;
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
return 0;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* nm_krings_delete callback for monitors */
|
|
|
|
static void
|
|
|
|
netmap_monitor_krings_delete(struct netmap_adapter *na)
|
|
|
|
{
|
|
|
|
netmap_krings_delete(na);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static u_int
|
|
|
|
nm_txrx2flag(enum txrx t)
|
|
|
|
{
|
|
|
|
return (t == NR_RX ? NR_MONITOR_RX : NR_MONITOR_TX);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* allocate the monitors array in the monitored kring */
|
|
|
|
static int
|
|
|
|
nm_monitor_alloc(struct netmap_kring *kring, u_int n)
|
|
|
|
{
|
2017-06-12 22:53:18 +00:00
|
|
|
size_t old_len, len;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
struct netmap_kring **nm;
|
|
|
|
|
|
|
|
if (n <= kring->max_monitors)
|
|
|
|
/* we already have more entries that requested */
|
|
|
|
return 0;
|
2016-10-18 15:41:57 +00:00
|
|
|
|
2017-06-12 22:53:18 +00:00
|
|
|
old_len = sizeof(struct netmap_kring *)*kring->max_monitors;
|
2018-05-18 03:38:17 +00:00
|
|
|
len = sizeof(struct netmap_kring *) * n;
|
2017-06-12 22:53:18 +00:00
|
|
|
nm = nm_os_realloc(kring->monitors, len, old_len);
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
if (nm == NULL)
|
|
|
|
return ENOMEM;
|
|
|
|
|
|
|
|
kring->monitors = nm;
|
|
|
|
kring->max_monitors = n;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* deallocate the parent array in the parent adapter */
|
|
|
|
static void
|
|
|
|
nm_monitor_dealloc(struct netmap_kring *kring)
|
|
|
|
{
|
|
|
|
if (kring->monitors) {
|
|
|
|
if (kring->n_monitors > 0) {
|
|
|
|
D("freeing not empty monitor array for %s (%d dangling monitors)!", kring->name,
|
|
|
|
kring->n_monitors);
|
|
|
|
}
|
2017-06-12 22:53:18 +00:00
|
|
|
nm_os_free(kring->monitors);
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
kring->monitors = NULL;
|
|
|
|
kring->max_monitors = 0;
|
|
|
|
kring->n_monitors = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-06-12 22:53:18 +00:00
|
|
|
/* returns 1 iff kring has no monitors */
|
|
|
|
static inline int
|
|
|
|
nm_monitor_none(struct netmap_kring *kring)
|
|
|
|
{
|
|
|
|
return kring->n_monitors == 0 &&
|
|
|
|
kring->zmon_list[NR_TX].next == NULL &&
|
|
|
|
kring->zmon_list[NR_RX].next == NULL;
|
|
|
|
}
|
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
/*
|
|
|
|
* monitors work by replacing the nm_sync() and possibly the
|
|
|
|
* nm_notify() callbacks in the monitored rings.
|
|
|
|
*/
|
|
|
|
static int netmap_zmon_parent_txsync(struct netmap_kring *, int);
|
|
|
|
static int netmap_zmon_parent_rxsync(struct netmap_kring *, int);
|
|
|
|
static int netmap_monitor_parent_txsync(struct netmap_kring *, int);
|
|
|
|
static int netmap_monitor_parent_rxsync(struct netmap_kring *, int);
|
|
|
|
static int netmap_monitor_parent_notify(struct netmap_kring *, int);
|
|
|
|
|
|
|
|
/* add the monitor mkring to the list of monitors of kring.
|
|
|
|
* If this is the first monitor, intercept the callbacks
|
|
|
|
*/
|
|
|
|
static int
|
2017-06-12 22:53:18 +00:00
|
|
|
netmap_monitor_add(struct netmap_kring *mkring, struct netmap_kring *kring, int zmon)
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
{
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
int error = NM_IRQ_COMPLETED;
|
2017-06-12 22:53:18 +00:00
|
|
|
enum txrx t = kring->tx;
|
|
|
|
struct netmap_zmon_list *z = &kring->zmon_list[t];
|
|
|
|
struct netmap_zmon_list *mz = &mkring->zmon_list[t];
|
|
|
|
|
|
|
|
/* a zero-copy monitor which is not the first in the list
|
|
|
|
* must monitor the previous monitor
|
|
|
|
*/
|
|
|
|
if (zmon && z->prev != NULL)
|
|
|
|
kring = z->prev;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
|
2018-04-09 09:24:26 +00:00
|
|
|
/* synchronize with concurrently running nm_sync()s */
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
nm_kr_stop(kring, NM_KR_LOCKED);
|
2017-06-12 22:53:18 +00:00
|
|
|
|
|
|
|
if (nm_monitor_none(kring)) {
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
/* this is the first monitor, intercept callbacks */
|
2017-06-12 22:53:18 +00:00
|
|
|
ND("intercept callbacks on %s", kring->name);
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
kring->mon_sync = kring->nm_sync;
|
|
|
|
kring->mon_notify = kring->nm_notify;
|
|
|
|
if (kring->tx == NR_TX) {
|
2017-06-12 22:53:18 +00:00
|
|
|
kring->nm_sync = netmap_monitor_parent_txsync;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
} else {
|
2017-06-12 22:53:18 +00:00
|
|
|
kring->nm_sync = netmap_monitor_parent_rxsync;
|
|
|
|
kring->nm_notify = netmap_monitor_parent_notify;
|
|
|
|
kring->mon_tail = kring->nr_hwtail;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-06-12 22:53:18 +00:00
|
|
|
if (zmon) {
|
|
|
|
/* append the zmon to the list */
|
|
|
|
struct netmap_monitor_adapter *mna =
|
|
|
|
(struct netmap_monitor_adapter *)mkring->na;
|
|
|
|
struct netmap_adapter *pna;
|
|
|
|
|
|
|
|
if (z->prev != NULL)
|
|
|
|
z->prev->zmon_list[t].next = mkring;
|
|
|
|
mz->prev = z->prev;
|
|
|
|
z->prev = mkring;
|
|
|
|
if (z->next == NULL)
|
|
|
|
z->next = mkring;
|
|
|
|
|
|
|
|
/* grap a reference to the previous netmap adapter
|
|
|
|
* in the chain (this may be the monitored port
|
|
|
|
* or another zero-copy monitor)
|
|
|
|
*/
|
|
|
|
pna = kring->na;
|
|
|
|
netmap_adapter_get(pna);
|
|
|
|
netmap_adapter_put(mna->priv.np_na);
|
|
|
|
mna->priv.np_na = pna;
|
|
|
|
} else {
|
|
|
|
/* make sure the monitor array exists and is big enough */
|
|
|
|
error = nm_monitor_alloc(kring, kring->n_monitors + 1);
|
|
|
|
if (error)
|
|
|
|
goto out;
|
|
|
|
kring->monitors[kring->n_monitors] = mkring;
|
|
|
|
mkring->mon_pos[kring->tx] = kring->n_monitors;
|
|
|
|
kring->n_monitors++;
|
|
|
|
}
|
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
out:
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
nm_kr_start(kring);
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* remove the monitor mkring from the list of monitors of kring.
|
|
|
|
* If this is the last monitor, restore the original callbacks
|
|
|
|
*/
|
|
|
|
static void
|
|
|
|
netmap_monitor_del(struct netmap_kring *mkring, struct netmap_kring *kring)
|
|
|
|
{
|
2017-06-12 22:53:18 +00:00
|
|
|
struct netmap_zmon_list *mz = &mkring->zmon_list[kring->tx];
|
|
|
|
int zmon = nm_is_zmon(mkring->na);
|
|
|
|
|
|
|
|
|
|
|
|
if (zmon && mz->prev != NULL)
|
|
|
|
kring = mz->prev;
|
|
|
|
|
2018-04-09 09:24:26 +00:00
|
|
|
/* synchronize with concurrently running nm_sync()s */
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
nm_kr_stop(kring, NM_KR_LOCKED);
|
2017-06-12 22:53:18 +00:00
|
|
|
|
|
|
|
if (zmon) {
|
|
|
|
/* remove the monitor from the list */
|
|
|
|
if (mz->prev != NULL)
|
|
|
|
mz->prev->zmon_list[kring->tx].next = mz->next;
|
|
|
|
else
|
|
|
|
kring->zmon_list[kring->tx].next = mz->next;
|
|
|
|
if (mz->next != NULL) {
|
|
|
|
mz->next->zmon_list[kring->tx].prev = mz->prev;
|
|
|
|
} else {
|
|
|
|
kring->zmon_list[kring->tx].prev = mz->prev;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
/* this is a copy monitor */
|
|
|
|
uint32_t mon_pos = mkring->mon_pos[kring->tx];
|
|
|
|
kring->n_monitors--;
|
|
|
|
if (mon_pos != kring->n_monitors) {
|
|
|
|
kring->monitors[mon_pos] =
|
|
|
|
kring->monitors[kring->n_monitors];
|
|
|
|
kring->monitors[mon_pos]->mon_pos[kring->tx] = mon_pos;
|
|
|
|
}
|
|
|
|
kring->monitors[kring->n_monitors] = NULL;
|
|
|
|
if (kring->n_monitors == 0) {
|
|
|
|
nm_monitor_dealloc(kring);
|
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
}
|
2017-06-12 22:53:18 +00:00
|
|
|
|
|
|
|
if (nm_monitor_none(kring)) {
|
|
|
|
/* this was the last monitor, restore the callbacks */
|
|
|
|
ND("%s: restoring sync on %s: %p", mkring->name, kring->name,
|
|
|
|
kring->mon_sync);
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
kring->nm_sync = kring->mon_sync;
|
|
|
|
kring->mon_sync = NULL;
|
|
|
|
if (kring->tx == NR_RX) {
|
2016-10-18 15:41:57 +00:00
|
|
|
ND("%s: restoring notify on %s: %p",
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
mkring->name, kring->name, kring->mon_notify);
|
|
|
|
kring->nm_notify = kring->mon_notify;
|
|
|
|
kring->mon_notify = NULL;
|
|
|
|
}
|
|
|
|
}
|
2017-06-12 22:53:18 +00:00
|
|
|
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
nm_kr_start(kring);
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* This is called when the monitored adapter leaves netmap mode
|
|
|
|
* (see netmap_do_unregif).
|
|
|
|
* We need to notify the monitors that the monitored rings are gone.
|
|
|
|
* We do this by setting their mna->priv.np_na to NULL.
|
|
|
|
* Note that the rings are already stopped when this happens, so
|
|
|
|
* no monitor ring callback can be active.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
netmap_monitor_stop(struct netmap_adapter *na)
|
|
|
|
{
|
|
|
|
enum txrx t;
|
|
|
|
|
|
|
|
for_rx_tx(t) {
|
|
|
|
u_int i;
|
|
|
|
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
for (i = 0; i < nma_get_nrings(na, t) + 1; i++) {
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
struct netmap_kring *kring = NMR(na, t)[i];
|
2017-06-12 22:53:18 +00:00
|
|
|
struct netmap_kring *zkring;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
u_int j;
|
|
|
|
|
|
|
|
for (j = 0; j < kring->n_monitors; j++) {
|
|
|
|
struct netmap_kring *mkring =
|
|
|
|
kring->monitors[j];
|
|
|
|
struct netmap_monitor_adapter *mna =
|
|
|
|
(struct netmap_monitor_adapter *)mkring->na;
|
|
|
|
/* forget about this adapter */
|
2017-06-12 22:53:18 +00:00
|
|
|
if (mna->priv.np_na != NULL) {
|
|
|
|
netmap_adapter_put(mna->priv.np_na);
|
|
|
|
mna->priv.np_na = NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
zkring = kring->zmon_list[kring->tx].next;
|
|
|
|
if (zkring != NULL) {
|
|
|
|
struct netmap_monitor_adapter *next =
|
|
|
|
(struct netmap_monitor_adapter *)zkring->na;
|
|
|
|
struct netmap_monitor_adapter *this =
|
|
|
|
(struct netmap_monitor_adapter *)na;
|
|
|
|
struct netmap_adapter *pna = this->priv.np_na;
|
|
|
|
/* let the next monitor forget about us */
|
|
|
|
if (next->priv.np_na != NULL) {
|
|
|
|
netmap_adapter_put(next->priv.np_na);
|
|
|
|
}
|
|
|
|
if (pna != NULL && nm_is_zmon(na)) {
|
|
|
|
/* we are a monitor ourselves and we may
|
|
|
|
* need to pass down the reference to
|
|
|
|
* the previous adapter in the chain
|
|
|
|
*/
|
|
|
|
netmap_adapter_get(pna);
|
|
|
|
next->priv.np_na = pna;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
next->priv.np_na = NULL;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* common functions for the nm_register() callbacks of both kind of
|
|
|
|
* monitors.
|
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
|
|
|
*/
|
|
|
|
static int
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
netmap_monitor_reg_common(struct netmap_adapter *na, int onoff, int zmon)
|
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
|
|
|
{
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
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, *mkring;
|
|
|
|
int i;
|
2017-06-12 22:53:18 +00:00
|
|
|
enum txrx t, s;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
|
|
|
|
ND("%p: onoff %d", na, onoff);
|
|
|
|
if (onoff) {
|
|
|
|
if (pna == NULL) {
|
|
|
|
/* parent left netmap mode, fatal */
|
|
|
|
D("%s: internal error", na->name);
|
|
|
|
return ENXIO;
|
|
|
|
}
|
|
|
|
for_rx_tx(t) {
|
2017-06-12 22:53:18 +00:00
|
|
|
for (i = 0; i < nma_get_nrings(na, t) + 1; i++) {
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
mkring = NMR(na, t)[i];
|
2017-06-12 22:53:18 +00:00
|
|
|
if (!nm_kring_pending_on(mkring))
|
|
|
|
continue;
|
|
|
|
mkring->nr_mode = NKR_NETMAP_ON;
|
|
|
|
if (t == NR_TX)
|
|
|
|
continue;
|
|
|
|
for_rx_tx(s) {
|
|
|
|
if (i > nma_get_nrings(pna, s))
|
|
|
|
continue;
|
|
|
|
if (mna->flags & nm_txrx2flag(s)) {
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
kring = NMR(pna, s)[i];
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
netmap_monitor_add(mkring, kring, zmon);
|
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
na->na_flags |= NAF_NETMAP_ON;
|
|
|
|
} else {
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
if (na->active_fds == 0)
|
|
|
|
na->na_flags &= ~NAF_NETMAP_ON;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
for_rx_tx(t) {
|
2017-06-12 22:53:18 +00:00
|
|
|
for (i = 0; i < nma_get_nrings(na, t) + 1; i++) {
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
mkring = NMR(na, t)[i];
|
2017-06-12 22:53:18 +00:00
|
|
|
if (!nm_kring_pending_off(mkring))
|
|
|
|
continue;
|
|
|
|
mkring->nr_mode = NKR_NETMAP_OFF;
|
|
|
|
if (t == NR_TX)
|
|
|
|
continue;
|
|
|
|
/* we cannot access the parent krings if the parent
|
|
|
|
* has left netmap mode. This is signaled by a NULL
|
|
|
|
* pna pointer
|
|
|
|
*/
|
|
|
|
if (pna == NULL)
|
|
|
|
continue;
|
|
|
|
for_rx_tx(s) {
|
|
|
|
if (i > nma_get_nrings(pna, s))
|
|
|
|
continue;
|
|
|
|
if (mna->flags & nm_txrx2flag(s)) {
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
kring = NMR(pna, s)[i];
|
2017-06-12 22:53:18 +00:00
|
|
|
netmap_monitor_del(mkring, kring);
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
****************************************************************
|
2016-10-18 15:41:57 +00:00
|
|
|
* functions specific for zero-copy monitors
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
****************************************************************
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Common function for both zero-copy tx and rx nm_sync()
|
|
|
|
* callbacks
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
netmap_zmon_parent_sync(struct netmap_kring *kring, int flags, enum txrx tx)
|
|
|
|
{
|
2017-06-12 22:53:18 +00:00
|
|
|
struct netmap_kring *mkring = kring->zmon_list[tx].next;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
struct netmap_ring *ring = kring->ring, *mring;
|
|
|
|
int error = 0;
|
|
|
|
int rel_slots, free_slots, busy, sent = 0;
|
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
|
|
|
u_int beg, end, i;
|
|
|
|
u_int lim = kring->nkr_num_slots - 1,
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
mlim; // = mkring->nkr_num_slots - 1;
|
|
|
|
|
|
|
|
if (mkring == NULL) {
|
|
|
|
RD(5, "NULL monitor on %s", kring->name);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
mring = mkring->ring;
|
|
|
|
mlim = mkring->nkr_num_slots - 1;
|
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
|
|
|
|
|
|
|
/* get the relased slots (rel_slots) */
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
if (tx == NR_TX) {
|
2017-06-12 22:53:18 +00:00
|
|
|
beg = kring->nr_hwtail + 1;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
error = kring->mon_sync(kring, flags);
|
|
|
|
if (error)
|
|
|
|
return error;
|
2017-06-12 22:53:18 +00:00
|
|
|
end = kring->nr_hwtail + 1;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
} else { /* NR_RX */
|
|
|
|
beg = kring->nr_hwcur;
|
|
|
|
end = kring->rhead;
|
|
|
|
}
|
|
|
|
|
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
|
|
|
rel_slots = end - beg;
|
|
|
|
if (rel_slots < 0)
|
|
|
|
rel_slots += kring->nkr_num_slots;
|
|
|
|
|
|
|
|
if (!rel_slots) {
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
/* no released slots, but we still need
|
|
|
|
* to call rxsync if this is a rx ring
|
|
|
|
*/
|
|
|
|
goto out_rxsync;
|
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
|
|
|
}
|
|
|
|
|
|
|
|
/* 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;
|
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
if (!free_slots)
|
|
|
|
goto out;
|
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
|
|
|
|
|
|
|
/* swap min(free_slots, rel_slots) slots */
|
|
|
|
if (free_slots < rel_slots) {
|
|
|
|
beg += (rel_slots - free_slots);
|
|
|
|
rel_slots = free_slots;
|
|
|
|
}
|
2017-06-12 22:53:18 +00:00
|
|
|
if (unlikely(beg >= kring->nkr_num_slots))
|
|
|
|
beg -= kring->nkr_num_slots;
|
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
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
sent = rel_slots;
|
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
|
|
|
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;
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
ND(5, "beg %d buf_idx %d", beg, tmp);
|
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
|
|
|
|
|
|
|
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;
|
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
out:
|
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
|
|
|
mtx_unlock(&mkring->q_lock);
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
|
|
|
|
if (sent) {
|
|
|
|
/* notify the new frames to the monitor */
|
|
|
|
mkring->nm_notify(mkring, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
out_rxsync:
|
|
|
|
if (tx == NR_RX)
|
|
|
|
error = kring->mon_sync(kring, flags);
|
|
|
|
|
|
|
|
return error;
|
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
|
|
|
}
|
|
|
|
|
|
|
|
/* callback used to replace the nm_sync callback in the monitored tx rings */
|
|
|
|
static int
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
netmap_zmon_parent_txsync(struct netmap_kring *kring, int flags)
|
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
|
|
|
{
|
2018-05-18 03:38:17 +00:00
|
|
|
return netmap_zmon_parent_sync(kring, flags, NR_TX);
|
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
|
|
|
}
|
|
|
|
|
|
|
|
/* callback used to replace the nm_sync callback in the monitored rx rings */
|
|
|
|
static int
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
netmap_zmon_parent_rxsync(struct netmap_kring *kring, int flags)
|
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
|
|
|
{
|
2018-05-18 03:38:17 +00:00
|
|
|
return netmap_zmon_parent_sync(kring, flags, NR_RX);
|
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
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
netmap_zmon_reg(struct netmap_adapter *na, int onoff)
|
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
|
|
|
{
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
return netmap_monitor_reg_common(na, onoff, 1 /* zcopy */);
|
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
|
|
|
}
|
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
/* nm_dtor callback for monitors */
|
|
|
|
static void
|
|
|
|
netmap_zmon_dtor(struct netmap_adapter *na)
|
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
|
|
|
{
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
struct netmap_monitor_adapter *mna =
|
|
|
|
(struct netmap_monitor_adapter *)na;
|
|
|
|
struct netmap_priv_d *priv = &mna->priv;
|
|
|
|
struct netmap_adapter *pna = priv->np_na;
|
|
|
|
|
|
|
|
netmap_adapter_put(pna);
|
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
|
|
|
}
|
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
/*
|
|
|
|
****************************************************************
|
2016-10-18 15:41:57 +00:00
|
|
|
* functions specific for copy monitors
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
****************************************************************
|
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
|
|
|
*/
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
|
|
|
|
static void
|
|
|
|
netmap_monitor_parent_sync(struct netmap_kring *kring, u_int first_new, int new_slots)
|
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
|
|
|
{
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
u_int j;
|
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
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
for (j = 0; j < kring->n_monitors; j++) {
|
|
|
|
struct netmap_kring *mkring = kring->monitors[j];
|
|
|
|
u_int i, mlim, beg;
|
|
|
|
int free_slots, busy, sent = 0, m;
|
|
|
|
u_int lim = kring->nkr_num_slots - 1;
|
|
|
|
struct netmap_ring *ring = kring->ring, *mring = mkring->ring;
|
|
|
|
u_int max_len = NETMAP_BUF_SIZE(mkring->na);
|
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
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
mlim = mkring->nkr_num_slots - 1;
|
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
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
/* 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)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
/* copy min(free_slots, new_slots) slots */
|
|
|
|
m = new_slots;
|
|
|
|
beg = first_new;
|
|
|
|
if (free_slots < m) {
|
|
|
|
beg += (m - free_slots);
|
|
|
|
if (beg >= kring->nkr_num_slots)
|
|
|
|
beg -= kring->nkr_num_slots;
|
|
|
|
m = free_slots;
|
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
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
|
|
|
|
for ( ; m; m--) {
|
|
|
|
struct netmap_slot *s = &ring->slot[beg];
|
|
|
|
struct netmap_slot *ms = &mring->slot[i];
|
|
|
|
u_int copy_len = s->len;
|
|
|
|
char *src = NMB(kring->na, s),
|
|
|
|
*dst = NMB(mkring->na, ms);
|
|
|
|
|
|
|
|
if (unlikely(copy_len > max_len)) {
|
|
|
|
RD(5, "%s->%s: truncating %d to %d", kring->name,
|
|
|
|
mkring->name, copy_len, max_len);
|
|
|
|
copy_len = max_len;
|
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
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
|
|
|
|
memcpy(dst, src, copy_len);
|
|
|
|
ms->len = copy_len;
|
|
|
|
sent++;
|
|
|
|
|
|
|
|
beg = nm_next(beg, lim);
|
|
|
|
i = nm_next(i, mlim);
|
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
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
mb();
|
|
|
|
mkring->nr_hwtail = i;
|
|
|
|
out:
|
|
|
|
mtx_unlock(&mkring->q_lock);
|
|
|
|
|
|
|
|
if (sent) {
|
|
|
|
/* notify the new frames to the monitor */
|
|
|
|
mkring->nm_notify(mkring, 0);
|
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
|
|
|
}
|
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* 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)
|
|
|
|
{
|
|
|
|
u_int first_new;
|
|
|
|
int new_slots;
|
|
|
|
|
|
|
|
/* get the new slots */
|
2017-06-12 22:53:18 +00:00
|
|
|
if (kring->n_monitors > 0) {
|
|
|
|
first_new = kring->nr_hwcur;
|
|
|
|
new_slots = kring->rhead - first_new;
|
|
|
|
if (new_slots < 0)
|
|
|
|
new_slots += kring->nkr_num_slots;
|
|
|
|
if (new_slots)
|
|
|
|
netmap_monitor_parent_sync(kring, first_new, new_slots);
|
|
|
|
}
|
|
|
|
if (kring->zmon_list[NR_TX].next != NULL) {
|
|
|
|
return netmap_zmon_parent_txsync(kring, flags);
|
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
return kring->mon_sync(kring, flags);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* 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)
|
|
|
|
{
|
|
|
|
u_int first_new;
|
|
|
|
int new_slots, error;
|
|
|
|
|
|
|
|
/* get the new slots */
|
2017-06-12 22:53:18 +00:00
|
|
|
if (kring->zmon_list[NR_RX].next != NULL) {
|
|
|
|
error = netmap_zmon_parent_rxsync(kring, flags);
|
|
|
|
} else {
|
|
|
|
error = kring->mon_sync(kring, flags);
|
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
if (error)
|
|
|
|
return error;
|
2017-06-12 22:53:18 +00:00
|
|
|
if (kring->n_monitors > 0) {
|
|
|
|
first_new = kring->mon_tail;
|
|
|
|
new_slots = kring->nr_hwtail - first_new;
|
|
|
|
if (new_slots < 0)
|
|
|
|
new_slots += kring->nkr_num_slots;
|
|
|
|
if (new_slots)
|
|
|
|
netmap_monitor_parent_sync(kring, first_new, new_slots);
|
|
|
|
kring->mon_tail = kring->nr_hwtail;
|
|
|
|
}
|
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
|
|
|
return 0;
|
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
|
|
|
|
/* callback used to replace the nm_notify() callback in the monitored rx rings */
|
|
|
|
static int
|
|
|
|
netmap_monitor_parent_notify(struct netmap_kring *kring, int flags)
|
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
|
|
|
{
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
int (*notify)(struct netmap_kring*, int);
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
ND(5, "%s %x", kring->name, flags);
|
|
|
|
/* ?xsync callbacks have tryget called by their callers
|
|
|
|
* (NIOCREGIF and poll()), but here we have to call it
|
|
|
|
* by ourself
|
|
|
|
*/
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
if (nm_kr_tryget(kring, 0, NULL)) {
|
|
|
|
/* in all cases, just skip the sync */
|
|
|
|
return NM_IRQ_COMPLETED;
|
|
|
|
}
|
|
|
|
if (kring->n_monitors > 0) {
|
|
|
|
netmap_monitor_parent_rxsync(kring, NAF_FORCE_READ);
|
2017-06-12 22:53:18 +00:00
|
|
|
}
|
|
|
|
if (nm_monitor_none(kring)) {
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
/* we are no longer monitoring this ring, so both
|
|
|
|
* mon_sync and mon_notify are NULL
|
|
|
|
*/
|
|
|
|
notify = kring->nm_notify;
|
2017-06-12 22:53:18 +00:00
|
|
|
} else {
|
|
|
|
notify = kring->mon_notify;
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
nm_kr_put(kring);
|
2018-05-18 03:38:17 +00:00
|
|
|
return notify(kring, flags);
|
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
|
|
|
}
|
|
|
|
|
|
|
|
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
static int
|
|
|
|
netmap_monitor_reg(struct netmap_adapter *na, int onoff)
|
|
|
|
{
|
|
|
|
return netmap_monitor_reg_common(na, onoff, 0 /* no zcopy */);
|
|
|
|
}
|
|
|
|
|
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
|
|
|
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;
|
|
|
|
|
|
|
|
netmap_adapter_put(pna);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
/* check if req is a request for a monitor adapter that we can satisfy */
|
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
|
|
|
int
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
netmap_get_monitor_na(struct nmreq_header *hdr, struct netmap_adapter **na,
|
|
|
|
struct netmap_mem_d *nmd, int create)
|
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
|
|
|
{
|
2018-05-18 03:38:17 +00:00
|
|
|
struct nmreq_register *req = (struct nmreq_register *)(uintptr_t)hdr->nr_body;
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
struct nmreq_register preq;
|
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
|
|
|
struct netmap_adapter *pna; /* parent adapter */
|
|
|
|
struct netmap_monitor_adapter *mna;
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
struct ifnet *ifp = NULL;
|
2017-06-12 22:53:18 +00:00
|
|
|
int error;
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
int zcopy = (req->nr_flags & NR_ZCOPY_MON);
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
char monsuff[10] = "";
|
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
|
|
|
|
2017-06-12 22:53:18 +00:00
|
|
|
if (zcopy) {
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
req->nr_flags |= (NR_MONITOR_TX | NR_MONITOR_RX);
|
2017-06-12 22:53:18 +00:00
|
|
|
}
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
if ((req->nr_flags & (NR_MONITOR_TX | NR_MONITOR_RX)) == 0) {
|
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
|
|
|
ND("not a monitor");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
/* this is a request for a monitor adapter */
|
|
|
|
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
ND("flags %lx", req->nr_flags);
|
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
|
|
|
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
/* First, try to find the adapter that we want to monitor.
|
|
|
|
* We use the same req, after we have turned off the monitor flags.
|
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
|
|
|
* In this way we can potentially monitor everything netmap understands,
|
|
|
|
* except other monitors.
|
|
|
|
*/
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
memcpy(&preq, req, sizeof(preq));
|
|
|
|
preq.nr_flags &= ~(NR_MONITOR_TX | NR_MONITOR_RX | NR_ZCOPY_MON);
|
2018-05-18 03:38:17 +00:00
|
|
|
hdr->nr_body = (uintptr_t)&preq;
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
error = netmap_get_na(hdr, &pna, &ifp, nmd, create);
|
2018-05-18 03:38:17 +00:00
|
|
|
hdr->nr_body = (uintptr_t)req;
|
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
|
|
|
if (error) {
|
|
|
|
D("parent lookup failed: %d", error);
|
|
|
|
return error;
|
|
|
|
}
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
ND("found parent: %s", pna->name);
|
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
|
|
|
|
|
|
|
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;
|
|
|
|
}
|
|
|
|
|
2017-06-12 22:53:18 +00:00
|
|
|
mna = nm_os_malloc(sizeof(*mna));
|
|
|
|
if (mna == NULL) {
|
|
|
|
D("memory error");
|
|
|
|
error = ENOMEM;
|
|
|
|
goto put_out;
|
|
|
|
}
|
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
|
|
|
mna->priv.np_na = pna;
|
2017-06-12 22:53:18 +00:00
|
|
|
|
|
|
|
/* grab all the rings we need in the parent */
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
error = netmap_interp_ringid(&mna->priv, req->nr_mode, req->nr_ringid,
|
|
|
|
req->nr_flags);
|
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
|
|
|
if (error) {
|
|
|
|
D("ringid error");
|
2017-06-12 22:53:18 +00:00
|
|
|
goto free_out;
|
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
|
|
|
}
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
if (mna->priv.np_qlast[NR_TX] - mna->priv.np_qfirst[NR_TX] == 1) {
|
|
|
|
snprintf(monsuff, 10, "-%d", mna->priv.np_qfirst[NR_TX]);
|
|
|
|
}
|
|
|
|
snprintf(mna->up.name, sizeof(mna->up.name), "%s%s/%s%s%s", pna->name,
|
|
|
|
monsuff,
|
|
|
|
zcopy ? "z" : "",
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
(req->nr_flags & NR_MONITOR_RX) ? "r" : "",
|
|
|
|
(req->nr_flags & NR_MONITOR_TX) ? "t" : "");
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
|
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
|
|
|
/* the monitor supports the host rings iff the parent does */
|
2017-06-12 22:53:18 +00:00
|
|
|
mna->up.na_flags |= (pna->na_flags & NAF_HOST_RINGS);
|
Sync netmap sources with the version in our private tree.
This commit contains large contributions from Giuseppe Lettieri and
Stefano Garzarella, is partly supported by grants from Verisign and Cisco,
and brings in the following:
- fix zerocopy monitor ports and introduce copying monitor ports
(the latter are lower performance but give access to all traffic
in parallel with the application)
- exclusive open mode, useful to implement solutions that recover
from crashes of the main netmap client (suggested by Patrick Kelsey)
- revised memory allocator in preparation for the 'passthrough mode'
(ptnetmap) recently presented at bsdcan. ptnetmap is described in
S. Garzarella, G. Lettieri, L. Rizzo;
Virtual device passthrough for high speed VM networking,
ACM/IEEE ANCS 2015, Oakland (CA) May 2015
http://info.iet.unipi.it/~luigi/research.html
- fix rx CRC handing on ixl
- add module dependencies for netmap when building drivers as modules
- minor simplifications to device-specific routines (*txsync, *rxsync)
- general code cleanup (remove unused variables, introduce macros
to access rings and remove duplicate code,
Applications do not need to be recompiled, unless of course
they want to use the new features (monitors and exclusive open).
Those willing to try this code on stable/10 can just update the
sys/dev/netmap/*, sys/net/netmap* with the version in HEAD
and apply the small patches to individual device drivers.
MFC after: 1 month
Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
|
|
|
/* a do-nothing txsync: monitors cannot be used to inject packets */
|
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
|
|
|
mna->up.nm_txsync = netmap_monitor_txsync;
|
|
|
|
mna->up.nm_rxsync = netmap_monitor_rxsync;
|
|
|
|
mna->up.nm_krings_create = netmap_monitor_krings_create;
|
|
|
|
mna->up.nm_krings_delete = netmap_monitor_krings_delete;
|
2017-06-12 22:53:18 +00:00
|
|
|
mna->up.num_tx_rings = 1; // XXX what should we do here with chained zmons?
|
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
|
|
|
/* 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
|
|
|
|
*/
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
mna->up.num_tx_desc = req->nr_tx_slots;
|
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_bound_var(&mna->up.num_tx_desc, pna->num_tx_desc,
|
|
|
|
1, NM_MONITOR_MAXSLOTS, NULL);
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
mna->up.num_rx_desc = req->nr_rx_slots;
|
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_bound_var(&mna->up.num_rx_desc, pna->num_rx_desc,
|
|
|
|
1, NM_MONITOR_MAXSLOTS, NULL);
|
2017-06-12 22:53:18 +00:00
|
|
|
if (zcopy) {
|
|
|
|
mna->up.nm_register = netmap_zmon_reg;
|
|
|
|
mna->up.nm_dtor = netmap_zmon_dtor;
|
|
|
|
/* to have zero copy, we need to use the same memory allocator
|
|
|
|
* as the monitored port
|
|
|
|
*/
|
|
|
|
mna->up.nm_mem = netmap_mem_get(pna->nm_mem);
|
|
|
|
/* and the allocator cannot be changed */
|
|
|
|
mna->up.na_flags |= NAF_MEM_OWNER;
|
|
|
|
} else {
|
|
|
|
mna->up.nm_register = netmap_monitor_reg;
|
|
|
|
mna->up.nm_dtor = netmap_monitor_dtor;
|
|
|
|
mna->up.nm_mem = netmap_mem_private_new(
|
|
|
|
mna->up.num_tx_rings,
|
|
|
|
mna->up.num_tx_desc,
|
|
|
|
mna->up.num_rx_rings,
|
|
|
|
mna->up.num_rx_desc,
|
|
|
|
0, /* extra bufs */
|
|
|
|
0, /* pipes */
|
|
|
|
&error);
|
|
|
|
if (mna->up.nm_mem == NULL)
|
|
|
|
goto put_out;
|
|
|
|
}
|
|
|
|
|
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
|
|
|
error = netmap_attach_common(&mna->up);
|
|
|
|
if (error) {
|
|
|
|
D("attach_common error");
|
2017-06-12 22:53:18 +00:00
|
|
|
goto mem_put_out;
|
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
|
|
|
}
|
|
|
|
|
|
|
|
/* remember the traffic directions we have to monitor */
|
netmap: align codebase to the current upstream (commit id 3fb001303718146)
Changelist:
- Turn tx_rings and rx_rings arrays into arrays of pointers to kring
structs. This patch includes fixes for ixv, ixl, ix, re, cxgbe, iflib,
vtnet and ptnet drivers to cope with the change.
- Generalize the nm_config() callback to accept a struct containing many
parameters.
- Introduce NKR_FAKERING to support buffers sharing (used for netmap
pipes)
- Improved API for external VALE modules.
- Various bug fixes and improvements to the netmap memory allocator,
including support for externally (userspace) allocated memory.
- Refactoring of netmap pipes: now linked rings share the same netmap
buffers, with a separate set of kring pointers (rhead, rcur, rtail).
Buffer swapping does not need to happen anymore.
- Large refactoring of the control API towards an extensible solution;
the goal is to allow the addition of more commands and extension of
existing ones (with new options) without the need of hacks or the
risk of running out of configuration space.
A new NIOCCTRL ioctl has been added to handle all the requests of the
new control API, which cover all the functionalities so far supported.
The netmap API bumps from 11 to 12 with this patch. Full backward
compatibility is provided for the old control command (NIOCREGIF), by
means of a new netmap_legacy module. Many parts of the old netmap.h
header has now been moved to netmap_legacy.h (included by netmap.h).
Approved by: hrs (mentor)
2018-04-12 07:20:50 +00:00
|
|
|
mna->flags = (req->nr_flags & (NR_MONITOR_TX | NR_MONITOR_RX | NR_ZCOPY_MON));
|
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
|
|
|
|
|
|
|
*na = &mna->up;
|
|
|
|
netmap_adapter_get(*na);
|
|
|
|
|
|
|
|
/* keep the reference to the parent */
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
ND("monitor ok");
|
|
|
|
|
|
|
|
/* drop the reference to the ifp, if any */
|
|
|
|
if (ifp)
|
|
|
|
if_rele(ifp);
|
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
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
2017-06-12 22:53:18 +00:00
|
|
|
mem_put_out:
|
|
|
|
netmap_mem_put(mna->up.nm_mem);
|
|
|
|
free_out:
|
|
|
|
nm_os_free(mna);
|
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
|
|
|
put_out:
|
Import the current version of netmap, aligned with the one on github.
This commit, long overdue, contains contributions in the last 2 years
from Stefano Garzarella, Giuseppe Lettieri, Vincenzo Maffione, including:
+ fixes on monitor ports
+ the 'ptnet' virtual device driver, and ptnetmap backend, for
high speed virtual passthrough on VMs (bhyve fixes in an upcoming commit)
+ improved emulated netmap mode
+ more robust error handling
+ removal of stale code
+ various fixes to code and documentation (some mixup between RX and TX
parameters, and private and public variables)
We also include an additional tool, nmreplay, which is functionally
equivalent to tcpreplay but operating on netmap ports.
2016-10-16 14:13:32 +00:00
|
|
|
netmap_unget_na(pna, ifp);
|
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
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
#endif /* WITH_MONITOR */
|