freebsd-skq/sys/net/netmap_user.h

682 lines
19 KiB
C
Raw Normal View History

Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
/*
* Copyright (C) 2011-2014 Universita` di Pisa. All rights reserved.
*
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
* 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
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
* 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.
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
*/
/*
* $FreeBSD$
*
* Functions and macros to manipulate netmap structures and packets
* in userspace. See netmap(4) for more information.
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
*
* The address of the struct netmap_if, say nifp, is computed from the
* value returned from ioctl(.., NIOCREG, ...) and the mmap region:
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
* ioctl(fd, NIOCREG, &req);
* mem = mmap(0, ... );
* nifp = NETMAP_IF(mem, req.nr_nifp);
* (so simple, we could just do it manually)
*
* From there:
* struct netmap_ring *NETMAP_TXRING(nifp, index)
* struct netmap_ring *NETMAP_RXRING(nifp, index)
* we can access ring->cur, ring->head, ring->tail, etc.
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
*
* ring->slot[i] gives us the i-th slot (we can access
* directly len, flags, buf_idx)
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
*
2013-05-30 13:41:19 +00:00
* char *buf = NETMAP_BUF(ring, x) returns a pointer to
* the buffer numbered x
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
*
* All ring indexes (head, cur, tail) should always move forward.
* To compute the next index in a circular ring you can use
* i = nm_ring_next(ring, i);
*
* To ease porting apps from pcap to netmap we supply a few fuctions
* that can be called to open, close, read and write on netmap in a way
* similar to libpcap. Note that the read/write function depend on
* an ioctl()/select()/poll() being issued to refill rings or push
* packets out.
*
* In order to use these, include #define NETMAP_WITH_LIBS
* in the source file that invokes these functions.
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
*/
#ifndef _NET_NETMAP_USER_H_
#define _NET_NETMAP_USER_H_
#include <stdint.h>
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
#include <sys/socket.h> /* apple needs sockaddr */
#include <net/if.h> /* IFNAMSIZ */
#ifndef likely
#define likely(x) __builtin_expect(!!(x), 1)
#define unlikely(x) __builtin_expect(!!(x), 0)
#endif /* likely and unlikely */
#include <net/netmap.h>
/* helper macro */
#define _NETMAP_OFFSET(type, ptr, offset) \
((type)(void *)((char *)(ptr) + (offset)))
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
#define NETMAP_IF(_base, _ofs) _NETMAP_OFFSET(struct netmap_if *, _base, _ofs)
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
#define NETMAP_TXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
nifp, (nifp)->ring_ofs[index] )
#define NETMAP_RXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
nifp, (nifp)->ring_ofs[index + (nifp)->ni_tx_rings + 1] )
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
#define NETMAP_BUF(ring, index) \
((char *)(ring) + (ring)->buf_ofs + ((index)*(ring)->nr_buf_size))
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
#define NETMAP_BUF_IDX(ring, buf) \
( ((char *)(buf) - ((char *)(ring) + (ring)->buf_ofs) ) / \
(ring)->nr_buf_size )
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
static inline uint32_t
nm_ring_next(struct netmap_ring *r, uint32_t i)
{
return ( unlikely(i + 1 == r->num_slots) ? 0 : i + 1);
}
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
/*
* Return 1 if we have pending transmissions in the tx ring.
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
* When everything is complete ring->head = ring->tail + 1 (modulo ring size)
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
*/
static inline int
nm_tx_pending(struct netmap_ring *r)
{
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
return nm_ring_next(r, r->tail) != r->head;
}
static inline uint32_t
nm_ring_space(struct netmap_ring *ring)
{
int ret = ring->tail - ring->cur;
if (ret < 0)
ret += ring->num_slots;
return ret;
}
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
#ifdef NETMAP_WITH_LIBS
/*
* Support for simple I/O libraries.
* Include other system headers required for compiling this.
*/
#ifndef HAVE_NETMAP_WITH_LIBS
#define HAVE_NETMAP_WITH_LIBS
#include <stdio.h>
#include <sys/time.h>
#include <sys/mman.h>
#include <string.h> /* memset */
#include <sys/ioctl.h>
#include <sys/errno.h> /* EINVAL */
#include <fcntl.h> /* O_RDWR */
#include <unistd.h> /* close() */
2014-01-17 04:38:58 +00:00
#include <signal.h>
#include <stdlib.h>
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
#ifndef ND /* debug macros */
/* debug support */
#define ND(_fmt, ...) do {} while(0)
#define D(_fmt, ...) \
do { \
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 timeval _t0; \
gettimeofday(&_t0, NULL); \
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
fprintf(stderr, "%03d.%06d %s [%d] " _fmt "\n", \
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)(_t0.tv_sec % 1000), (int)_t0.tv_usec, \
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
__FUNCTION__, __LINE__, ##__VA_ARGS__); \
} while (0)
/* Rate limited version of "D", lps indicates how many per second */
#define RD(lps, format, ...) \
do { \
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 __t0, __cnt; \
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
struct timeval __xxts; \
gettimeofday(&__xxts, NULL); \
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 (__t0 != __xxts.tv_sec) { \
__t0 = __xxts.tv_sec; \
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
__cnt = 0; \
} \
if (__cnt++ < lps) { \
D(format, ##__VA_ARGS__); \
} \
} while (0)
#endif
struct nm_pkthdr { /* same as pcap_pkthdr */
struct timeval ts;
uint32_t caplen;
uint32_t len;
};
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
struct nm_stat { /* same as pcap_stat */
2014-01-17 04:38:58 +00:00
u_int ps_recv;
u_int ps_drop;
u_int ps_ifdrop;
#ifdef WIN32
u_int bs_capt;
#endif /* WIN32 */
};
#define NM_ERRBUF_SIZE 512
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
struct nm_desc {
struct nm_desc *self; /* point to self if netmap. */
int fd;
void *mem;
uint32_t memsize;
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
int done_mmap; /* set if mem is the result of mmap */
struct netmap_if * const nifp;
2014-01-17 04:38:58 +00:00
uint16_t first_tx_ring, last_tx_ring, cur_tx_ring;
uint16_t first_rx_ring, last_rx_ring, cur_rx_ring;
struct nmreq req; /* also contains the nr_name = ifname */
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
struct nm_pkthdr hdr;
2014-01-17 04:38:58 +00:00
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
/*
* The memory contains netmap_if, rings and then buffers.
* Given a pointer (e.g. to nm_inject) we can compare with
* mem/buf_start/buf_end to tell if it is a buffer or
* some other descriptor in our region.
* We also store a pointer to some ring as it helps in the
* translation from buffer indexes to addresses.
*/
struct netmap_ring * const some_ring;
void * const buf_start;
void * const buf_end;
2014-01-17 04:38:58 +00:00
/* parameters from pcap_open_live */
int snaplen;
int promisc;
int to_ms;
char *errbuf;
/* save flags so we can restore them on close */
uint32_t if_flags;
uint32_t if_reqcap;
uint32_t if_curcap;
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
struct nm_stat st;
2014-01-17 04:38:58 +00:00
char msg[NM_ERRBUF_SIZE];
};
/*
* when the descriptor is open correctly, d->self == d
* Eventually we should also use some magic number.
*/
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
#define P2NMD(p) ((struct nm_desc *)(p))
#define IS_NETMAP_DESC(d) ((d) && P2NMD(d)->self == P2NMD(d))
#define NETMAP_FD(d) (P2NMD(d)->fd)
/*
* this is a slightly optimized copy routine which rounds
* to multiple of 64 bytes and is often faster than dealing
* with other odd sizes. We assume there is enough room
* in the source and destination buffers.
*
* XXX only for multiples of 64 bytes, non overlapped.
*/
static inline void
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
nm_pkt_copy(const void *_src, void *_dst, int l)
{
const uint64_t *src = (const uint64_t *)_src;
uint64_t *dst = (uint64_t *)_dst;
if (unlikely(l >= 1024)) {
memcpy(dst, src, l);
return;
}
for (; likely(l > 0); l-=64) {
*dst++ = *src++;
*dst++ = *src++;
*dst++ = *src++;
*dst++ = *src++;
*dst++ = *src++;
*dst++ = *src++;
*dst++ = *src++;
*dst++ = *src++;
}
}
/*
* The callback, invoked on each received packet. Same as libpcap
*/
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
typedef void (*nm_cb_t)(u_char *, const struct nm_pkthdr *, const u_char *d);
/*
*--- the pcap-like API ---
*
* nm_open() opens a file descriptor, binds to a port and maps memory.
*
* ifname (netmap:foo or vale:foo) is the port name
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
* a suffix can indicate the follwing:
* ^ bind the host (sw) ring pair
* * bind host and NIC ring pairs (transparent)
* -NN bind individual NIC ring pair
* {NN bind master side of pipe NN
* }NN bind slave side of pipe NN
*
* req provides the initial values of nmreq before parsing ifname.
* Remember that the ifname parsing will override the ring
* number in nm_ringid, and part of nm_flags;
* flags special functions, normally 0
* indicates which fields of *arg are significant
* arg special functions, normally NULL
* if passed a netmap_desc with mem != NULL,
* use that memory instead of mmap.
*/
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
static struct nm_desc *nm_open(const char *ifname, const struct nmreq *req,
uint64_t flags, const struct nm_desc *arg);
/*
* nm_open can import some fields from the parent descriptor.
* These flags control which ones.
* Also in flags you can specify NETMAP_NO_TX_POLL and NETMAP_DO_RX_POLL,
* which set the initial value for these flags.
* Note that the 16 low bits of the flags are reserved for data
* that may go into the nmreq.
*/
enum {
NM_OPEN_NO_MMAP = 0x040000, /* reuse mmap from parent */
NM_OPEN_IFNAME = 0x080000, /* nr_name, nr_ringid, nr_flags */
NM_OPEN_ARG1 = 0x100000,
NM_OPEN_ARG2 = 0x200000,
NM_OPEN_ARG3 = 0x400000,
NM_OPEN_RING_CFG = 0x800000, /* tx|rx rings|slots */
};
/*
* nm_close() closes and restores the port to its previous state
*/
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
static int nm_close(struct nm_desc *);
/*
* nm_inject() is the same as pcap_inject()
* nm_dispatch() is the same as pcap_dispatch()
* nm_nextpkt() is the same as pcap_next()
*/
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
static int nm_inject(struct nm_desc *, const void *, size_t);
static int nm_dispatch(struct nm_desc *, int, nm_cb_t, u_char *);
static u_char *nm_nextpkt(struct nm_desc *, struct nm_pkthdr *);
/*
* Try to open, return descriptor if successful, NULL otherwise.
* An invalid netmap name will return errno = 0;
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
* You can pass a pointer to a pre-filled nm_desc to add special
* parameters. Flags is used as follows
* NM_OPEN_NO_MMAP use the memory from arg, only
* if the nr_arg2 (memory block) matches.
* NM_OPEN_ARG1 use req.nr_arg1 from arg
* NM_OPEN_ARG2 use req.nr_arg2 from arg
* NM_OPEN_RING_CFG user ring config from arg
*/
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
static struct nm_desc *
nm_open(const char *ifname, const struct nmreq *req,
uint64_t new_flags, const struct nm_desc *arg)
{
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
struct nm_desc *d = NULL;
const struct nm_desc *parent = arg;
u_int namelen;
uint32_t nr_ringid = 0, nr_flags;
const char *port = NULL;
const char *errmsg = NULL;
if (strncmp(ifname, "netmap:", 7) && strncmp(ifname, "vale", 4)) {
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
errno = 0; /* name not recognised, not an error */
return NULL;
}
if (ifname[0] == 'n')
ifname += 7;
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
/* scan for a separator */
for (port = ifname; *port && !index("-*^{}", *port); port++)
;
namelen = port - ifname;
if (namelen >= sizeof(d->req.nr_name)) {
errmsg = "name too long";
goto fail;
2014-01-17 04:38:58 +00:00
}
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
switch (*port) {
default: /* '\0', no suffix */
nr_flags = NR_REG_ALL_NIC;
break;
case '-': /* one NIC */
nr_flags = NR_REG_ONE_NIC;
nr_ringid = atoi(port + 1);
break;
case '*': /* NIC and SW, ignore port */
nr_flags = NR_REG_NIC_SW;
if (port[1]) {
errmsg = "invalid port for nic+sw";
goto fail;
}
break;
case '^': /* only sw ring */
nr_flags = NR_REG_SW;
if (port[1]) {
errmsg = "invalid port for sw ring";
goto fail;
}
break;
case '{':
nr_flags = NR_REG_PIPE_MASTER;
nr_ringid = atoi(port + 1);
break;
case '}':
nr_flags = NR_REG_PIPE_SLAVE;
nr_ringid = atoi(port + 1);
break;
}
if (nr_ringid >= NETMAP_RING_MASK) {
errmsg = "invalid ringid";
goto fail;
}
2014-01-17 04:38:58 +00:00
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
d = (struct nm_desc *)calloc(1, sizeof(*d));
if (d == NULL) {
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
errmsg = "nm_desc alloc failure";
errno = ENOMEM;
return NULL;
}
d->self = d; /* set this early so nm_close() works */
d->fd = open("/dev/netmap", O_RDWR);
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
if (d->fd < 0) {
errmsg = "cannot open /dev/netmap";
goto fail;
}
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
if (req)
d->req = *req;
d->req.nr_version = NETMAP_API;
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
d->req.nr_ringid &= ~NETMAP_RING_MASK;
/* these fields are overridden by ifname and flags processing */
d->req.nr_ringid |= nr_ringid;
d->req.nr_flags = nr_flags;
2014-01-17 04:38:58 +00:00
memcpy(d->req.nr_name, ifname, namelen);
d->req.nr_name[namelen] = '\0';
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
/* optionally import info from parent */
if (IS_NETMAP_DESC(parent) && new_flags) {
if (new_flags & NM_OPEN_ARG1)
D("overriding ARG1 %d", parent->req.nr_arg1);
d->req.nr_arg1 = new_flags & NM_OPEN_ARG1 ?
parent->req.nr_arg1 : 4;
if (new_flags & NM_OPEN_ARG2)
D("overriding ARG2 %d", parent->req.nr_arg2);
d->req.nr_arg2 = new_flags & NM_OPEN_ARG2 ?
parent->req.nr_arg2 : 0;
if (new_flags & NM_OPEN_ARG3)
D("overriding ARG3 %d", parent->req.nr_arg3);
d->req.nr_arg3 = new_flags & NM_OPEN_ARG3 ?
parent->req.nr_arg3 : 0;
if (new_flags & NM_OPEN_RING_CFG) {
D("overriding RING_CFG");
d->req.nr_tx_slots = parent->req.nr_tx_slots;
d->req.nr_rx_slots = parent->req.nr_rx_slots;
d->req.nr_tx_rings = parent->req.nr_tx_rings;
d->req.nr_rx_rings = parent->req.nr_rx_rings;
}
if (new_flags & NM_OPEN_IFNAME) {
D("overriding ifname %s ringid 0x%x flags 0x%x",
parent->req.nr_name, parent->req.nr_ringid,
parent->req.nr_flags);
memcpy(d->req.nr_name, parent->req.nr_name,
sizeof(d->req.nr_name));
d->req.nr_ringid = parent->req.nr_ringid;
d->req.nr_flags = parent->req.nr_flags;
}
}
/* add the *XPOLL flags */
d->req.nr_ringid |= new_flags & (NETMAP_NO_TX_POLL | NETMAP_DO_RX_POLL);
2014-01-17 04:38:58 +00:00
if (ioctl(d->fd, NIOCREGIF, &d->req)) {
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
errmsg = "NIOCREGIF failed";
goto fail;
2014-01-17 04:38:58 +00:00
}
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
if (IS_NETMAP_DESC(parent) && parent->mem &&
parent->req.nr_arg2 == d->req.nr_arg2) {
/* do not mmap, inherit from parent */
d->memsize = parent->memsize;
d->mem = parent->mem;
} else {
/* XXX TODO: check if memsize is too large (or there is overflow) */
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
d->memsize = d->req.nr_memsize;
d->mem = mmap(0, d->memsize, PROT_WRITE | PROT_READ, MAP_SHARED,
d->fd, 0);
if (d->mem == MAP_FAILED) {
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
errmsg = "mmap failed";
goto fail;
}
d->done_mmap = 1;
}
{
struct netmap_if *nifp = NETMAP_IF(d->mem, d->req.nr_offset);
struct netmap_ring *r = NETMAP_RXRING(nifp, );
*(struct netmap_if **)(uintptr_t)&(d->nifp) = nifp;
*(struct netmap_ring **)(uintptr_t)&d->some_ring = r;
*(void **)(uintptr_t)&d->buf_start = NETMAP_BUF(r, 0);
*(void **)(uintptr_t)&d->buf_end =
(char *)d->mem + d->memsize;
}
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 (d->req.nr_flags == NR_REG_SW) { /* host stack */
2014-01-17 04:38:58 +00:00
d->first_tx_ring = d->last_tx_ring = d->req.nr_tx_rings;
d->first_rx_ring = d->last_rx_ring = d->req.nr_rx_rings;
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 if (d->req.nr_flags == NR_REG_ALL_NIC) { /* only nic */
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
d->first_tx_ring = 0;
d->first_rx_ring = 0;
2014-01-17 04:38:58 +00:00
d->last_tx_ring = d->req.nr_tx_rings - 1;
d->last_rx_ring = d->req.nr_rx_rings - 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
} else if (d->req.nr_flags == NR_REG_NIC_SW) {
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
d->first_tx_ring = 0;
d->first_rx_ring = 0;
d->last_tx_ring = d->req.nr_tx_rings;
d->last_rx_ring = d->req.nr_rx_rings;
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 if (d->req.nr_flags == NR_REG_ONE_NIC) {
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
/* XXX check validity */
d->first_tx_ring = d->last_tx_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
d->first_rx_ring = d->last_rx_ring = d->req.nr_ringid & NETMAP_RING_MASK;
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
} else { /* pipes */
d->first_tx_ring = d->last_tx_ring = 0;
d->first_rx_ring = d->last_rx_ring = 0;
}
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
#ifdef DEBUG_NETMAP_USER
{ /* debugging code */
int i;
D("%s tx %d .. %d %d rx %d .. %d %d", ifname,
d->first_tx_ring, d->last_tx_ring, d->req.nr_tx_rings,
d->first_rx_ring, d->last_rx_ring, d->req.nr_rx_rings);
for (i = 0; i <= d->req.nr_tx_rings; i++) {
struct netmap_ring *r = NETMAP_TXRING(d->nifp, i);
D("TX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
2014-01-17 04:38:58 +00:00
}
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
for (i = 0; i <= d->req.nr_rx_rings; i++) {
struct netmap_ring *r = NETMAP_RXRING(d->nifp, i);
D("RX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
}
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
}
#endif /* debugging */
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
d->cur_tx_ring = d->first_tx_ring;
d->cur_rx_ring = d->first_rx_ring;
return d;
fail:
nm_close(d);
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
if (errmsg)
D("%s %s", errmsg, ifname);
if (errno == 0)
errno = EINVAL;
return NULL;
}
static int
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
nm_close(struct nm_desc *d)
{
/*
* ugly trick to avoid unused warnings
*/
static void *__xxzt[] __attribute__ ((unused)) =
{ (void *)nm_open, (void *)nm_inject,
(void *)nm_dispatch, (void *)nm_nextpkt } ;
if (d == NULL || d->self != d)
return EINVAL;
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
if (d->done_mmap && d->mem)
munmap(d->mem, d->memsize);
if (d->fd != -1)
close(d->fd);
bzero(d, sizeof(*d));
free(d);
return 0;
}
/*
* Same prototype as pcap_inject(), only need to cast.
*/
static int
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
nm_inject(struct nm_desc *d, const void *buf, size_t size)
{
2014-01-17 04:38:58 +00:00
u_int c, n = d->last_tx_ring - d->first_tx_ring + 1;
for (c = 0; c < n ; c++) {
/* compute current ring to use */
struct netmap_ring *ring;
uint32_t i, idx;
2014-01-17 04:38:58 +00:00
uint32_t ri = d->cur_tx_ring + c;
2014-01-17 04:38:58 +00:00
if (ri > d->last_tx_ring)
ri = d->first_tx_ring;
ring = NETMAP_TXRING(d->nifp, ri);
if (nm_ring_empty(ring)) {
continue;
}
i = ring->cur;
idx = ring->slot[i].buf_idx;
ring->slot[i].len = size;
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
nm_pkt_copy(buf, NETMAP_BUF(ring, idx), size);
2014-01-17 04:38:58 +00:00
d->cur_tx_ring = ri;
ring->head = ring->cur = nm_ring_next(ring, i);
return size;
}
return 0; /* fail */
}
/*
* Same prototype as pcap_dispatch(), only need to cast.
*/
static int
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
nm_dispatch(struct nm_desc *d, int cnt, nm_cb_t cb, u_char *arg)
{
2014-01-17 04:38:58 +00:00
int n = d->last_rx_ring - d->first_rx_ring + 1;
int c, got = 0, ri = d->cur_rx_ring;
if (cnt == 0)
cnt = -1;
/* cnt == -1 means infinite, but rings have a finite amount
* of buffers and the int is large enough that we never wrap,
* so we can omit checking for -1
*/
for (c=0; c < n && cnt != got; c++) {
/* compute current ring to use */
struct netmap_ring *ring;
2014-01-17 04:38:58 +00:00
ri = d->cur_rx_ring + c;
if (ri > d->last_rx_ring)
ri = d->first_rx_ring;
ring = NETMAP_RXRING(d->nifp, ri);
for ( ; !nm_ring_empty(ring) && cnt != got; got++) {
u_int i = ring->cur;
u_int idx = ring->slot[i].buf_idx;
u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
2014-01-17 04:38:58 +00:00
// __builtin_prefetch(buf);
d->hdr.len = d->hdr.caplen = ring->slot[i].len;
d->hdr.ts = ring->ts;
cb(arg, &d->hdr, buf);
ring->head = ring->cur = nm_ring_next(ring, i);
}
}
2014-01-17 04:38:58 +00:00
d->cur_rx_ring = ri;
return got;
}
static u_char *
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
nm_nextpkt(struct nm_desc *d, struct nm_pkthdr *hdr)
{
2014-01-17 04:38:58 +00:00
int ri = d->cur_rx_ring;
do {
/* compute current ring to use */
struct netmap_ring *ring = NETMAP_RXRING(d->nifp, ri);
if (!nm_ring_empty(ring)) {
u_int i = ring->cur;
u_int idx = ring->slot[i].buf_idx;
u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
2014-01-17 04:38:58 +00:00
// __builtin_prefetch(buf);
hdr->ts = ring->ts;
hdr->len = hdr->caplen = ring->slot[i].len;
ring->cur = nm_ring_next(ring, i);
/* we could postpone advancing head if we want
* to hold the buffer. This can be supported in
* the future.
*/
ring->head = ring->cur;
2014-01-17 04:38:58 +00:00
d->cur_rx_ring = ri;
return buf;
}
ri++;
2014-01-17 04:38:58 +00:00
if (ri > d->last_rx_ring)
ri = d->first_rx_ring;
} while (ri != d->cur_rx_ring);
return NULL; /* nothing found */
}
#endif /* !HAVE_NETMAP_WITH_LIBS */
#endif /* NETMAP_WITH_LIBS */
Bring in support for netmap, a framework for very efficient packet I/O from userspace, capable of line rate at 10G, see http://info.iet.unipi.it/~luigi/netmap/ At this time I am bringing in only the generic code (sys/dev/netmap/ plus two headers under sys/net/), and some sample applications in tools/tools/netmap. There is also a manpage in share/man/man4 [1] In order to make use of the framework you need to build a kernel with "device netmap", and patch individual drivers with the code that you can find in sys/dev/netmap/head.diff The file will go away as the relevant pieces are committed to the various device drivers, which should happen in a few days after talking to the driver maintainers. Netmap support is available at the moment for Intel 10G and 1G cards (ixgbe, em/lem/igb), and for the Realtek 1G card ("re"). I have partial patches for "bge" and am starting to work on "cxgbe". Hopefully changes are trivial enough so interested third parties can submit their patches. Interested people can contact me for advice on how to add netmap support to specific devices. CREDITS: Netmap has been developed by Luigi Rizzo and other collaborators at the Universita` di Pisa, and supported by EU project CHANGE (http://www.change-project.eu/) The code is distributed under a BSD Copyright. [1] In my opinion is a bad idea to have all manpage in one directory. We should place kernel documentation in the same dir that contains the code, which would make it much simpler to keep doc and code in sync, reduce the clutter in share/man/ and incidentally is the policy used for all of userspace code. Makefiles and doc tools can be trivially adjusted to find the manpages in the relevant subdirs.
2011-11-17 12:17:39 +00:00
#endif /* _NET_NETMAP_USER_H_ */