freebsd-dev/sys/dev/netmap/netmap_offloadings.c
Luigi Rizzo 4bf50f18eb 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

402 lines
12 KiB
C

/*
* Copyright (C) 2014 Vincenzo Maffione. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
/* $FreeBSD$ */
#if defined(__FreeBSD__)
#include <sys/cdefs.h> /* prerequisite */
#include <sys/types.h>
#include <sys/errno.h>
#include <sys/param.h> /* defines used in kernel.h */
#include <sys/kernel.h> /* types used in module initialization */
#include <sys/sockio.h>
#include <sys/socketvar.h> /* struct socket */
#include <sys/socket.h> /* sockaddrs */
#include <net/if.h>
#include <net/if_var.h>
#include <machine/bus.h> /* bus_dmamap_* */
#include <sys/endian.h>
#elif defined(linux)
#include "bsd_glue.h"
#elif defined(__APPLE__)
#warning OSX support is only partial
#include "osx_glue.h"
#else
#error Unsupported platform
#endif /* unsupported */
#include <net/netmap.h>
#include <dev/netmap/netmap_kern.h>
/* This routine is called by bdg_mismatch_datapath() when it finishes
* accumulating bytes for a segment, in order to fix some fields in the
* segment headers (which still contain the same content as the header
* of the original GSO packet). 'buf' points to the beginning (e.g.
* the ethernet header) of the segment, and 'len' is its length.
*/
static void gso_fix_segment(uint8_t *buf, size_t len, u_int idx,
u_int segmented_bytes, u_int last_segment,
u_int tcp, u_int iphlen)
{
struct nm_iphdr *iph = (struct nm_iphdr *)(buf + 14);
struct nm_ipv6hdr *ip6h = (struct nm_ipv6hdr *)(buf + 14);
uint16_t *check = NULL;
uint8_t *check_data = NULL;
if (iphlen == 20) {
/* Set the IPv4 "Total Length" field. */
iph->tot_len = htobe16(len-14);
ND("ip total length %u", be16toh(ip->tot_len));
/* Set the IPv4 "Identification" field. */
iph->id = htobe16(be16toh(iph->id) + idx);
ND("ip identification %u", be16toh(iph->id));
/* Compute and insert the IPv4 header checksum. */
iph->check = 0;
iph->check = nm_csum_ipv4(iph);
ND("IP csum %x", be16toh(iph->check));
} else {/* if (iphlen == 40) */
/* Set the IPv6 "Payload Len" field. */
ip6h->payload_len = htobe16(len-14-iphlen);
}
if (tcp) {
struct nm_tcphdr *tcph = (struct nm_tcphdr *)(buf + 14 + iphlen);
/* Set the TCP sequence number. */
tcph->seq = htobe32(be32toh(tcph->seq) + segmented_bytes);
ND("tcp seq %u", be32toh(tcph->seq));
/* Zero the PSH and FIN TCP flags if this is not the last
segment. */
if (!last_segment)
tcph->flags &= ~(0x8 | 0x1);
ND("last_segment %u", last_segment);
check = &tcph->check;
check_data = (uint8_t *)tcph;
} else { /* UDP */
struct nm_udphdr *udph = (struct nm_udphdr *)(buf + 14 + iphlen);
/* Set the UDP 'Length' field. */
udph->len = htobe16(len-14-iphlen);
check = &udph->check;
check_data = (uint8_t *)udph;
}
/* Compute and insert TCP/UDP checksum. */
*check = 0;
if (iphlen == 20)
nm_csum_tcpudp_ipv4(iph, check_data, len-14-iphlen, check);
else
nm_csum_tcpudp_ipv6(ip6h, check_data, len-14-iphlen, check);
ND("TCP/UDP csum %x", be16toh(*check));
}
/* The VALE mismatch datapath implementation. */
void bdg_mismatch_datapath(struct netmap_vp_adapter *na,
struct netmap_vp_adapter *dst_na,
struct nm_bdg_fwd *ft_p, struct netmap_ring *ring,
u_int *j, u_int lim, u_int *howmany)
{
struct netmap_slot *slot = NULL;
struct nm_vnet_hdr *vh = NULL;
/* Number of source slots to process. */
u_int frags = ft_p->ft_frags;
struct nm_bdg_fwd *ft_end = ft_p + frags;
/* Source and destination pointers. */
uint8_t *dst, *src;
size_t src_len, dst_len;
u_int j_start = *j;
u_int dst_slots = 0;
/* If the source port uses the offloadings, while destination doesn't,
* we grab the source virtio-net header and do the offloadings here.
*/
if (na->virt_hdr_len && !dst_na->virt_hdr_len) {
vh = (struct nm_vnet_hdr *)ft_p->ft_buf;
}
/* Init source and dest pointers. */
src = ft_p->ft_buf;
src_len = ft_p->ft_len;
slot = &ring->slot[*j];
dst = NMB(&dst_na->up, slot);
dst_len = src_len;
/* We are processing the first input slot and there is a mismatch
* between source and destination virt_hdr_len (SHL and DHL).
* When the a client is using virtio-net headers, the header length
* can be:
* - 10: the header corresponds to the struct nm_vnet_hdr
* - 12: the first 10 bytes correspond to the struct
* virtio_net_hdr, and the last 2 bytes store the
* "mergeable buffers" info, which is an optional
* hint that can be zeroed for compability
*
* The destination header is therefore built according to the
* following table:
*
* SHL | DHL | destination header
* -----------------------------
* 0 | 10 | zero
* 0 | 12 | zero
* 10 | 0 | doesn't exist
* 10 | 12 | first 10 bytes are copied from source header, last 2 are zero
* 12 | 0 | doesn't exist
* 12 | 10 | copied from the first 10 bytes of source header
*/
bzero(dst, dst_na->virt_hdr_len);
if (na->virt_hdr_len && dst_na->virt_hdr_len)
memcpy(dst, src, sizeof(struct nm_vnet_hdr));
/* Skip the virtio-net headers. */
src += na->virt_hdr_len;
src_len -= na->virt_hdr_len;
dst += dst_na->virt_hdr_len;
dst_len = dst_na->virt_hdr_len + src_len;
/* Here it could be dst_len == 0 (which implies src_len == 0),
* so we avoid passing a zero length fragment.
*/
if (dst_len == 0) {
ft_p++;
src = ft_p->ft_buf;
src_len = ft_p->ft_len;
dst_len = src_len;
}
if (vh && vh->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
u_int gso_bytes = 0;
/* Length of the GSO packet header. */
u_int gso_hdr_len = 0;
/* Pointer to the GSO packet header. Assume it is in a single fragment. */
uint8_t *gso_hdr = NULL;
/* Index of the current segment. */
u_int gso_idx = 0;
/* Payload data bytes segmented so far (e.g. TCP data bytes). */
u_int segmented_bytes = 0;
/* Length of the IP header (20 if IPv4, 40 if IPv6). */
u_int iphlen = 0;
/* Is this a TCP or an UDP GSO packet? */
u_int tcp = ((vh->gso_type & ~VIRTIO_NET_HDR_GSO_ECN)
== VIRTIO_NET_HDR_GSO_UDP) ? 0 : 1;
/* Segment the GSO packet contained into the input slots (frags). */
while (ft_p != ft_end) {
size_t copy;
/* Grab the GSO header if we don't have it. */
if (!gso_hdr) {
uint16_t ethertype;
gso_hdr = src;
/* Look at the 'Ethertype' field to see if this packet
* is IPv4 or IPv6.
*/
ethertype = be16toh(*((uint16_t *)(gso_hdr + 12)));
if (ethertype == 0x0800)
iphlen = 20;
else /* if (ethertype == 0x86DD) */
iphlen = 40;
ND(3, "type=%04x", ethertype);
/* Compute gso_hdr_len. For TCP we need to read the
* content of the 'Data Offset' field.
*/
if (tcp) {
struct nm_tcphdr *tcph =
(struct nm_tcphdr *)&gso_hdr[14+iphlen];
gso_hdr_len = 14 + iphlen + 4*(tcph->doff >> 4);
} else
gso_hdr_len = 14 + iphlen + 8; /* UDP */
ND(3, "gso_hdr_len %u gso_mtu %d", gso_hdr_len,
dst_na->mfs);
/* Advance source pointers. */
src += gso_hdr_len;
src_len -= gso_hdr_len;
if (src_len == 0) {
ft_p++;
if (ft_p == ft_end)
break;
src = ft_p->ft_buf;
src_len = ft_p->ft_len;
continue;
}
}
/* Fill in the header of the current segment. */
if (gso_bytes == 0) {
memcpy(dst, gso_hdr, gso_hdr_len);
gso_bytes = gso_hdr_len;
}
/* Fill in data and update source and dest pointers. */
copy = src_len;
if (gso_bytes + copy > dst_na->mfs)
copy = dst_na->mfs - gso_bytes;
memcpy(dst + gso_bytes, src, copy);
gso_bytes += copy;
src += copy;
src_len -= copy;
/* A segment is complete or we have processed all the
the GSO payload bytes. */
if (gso_bytes >= dst_na->mfs ||
(src_len == 0 && ft_p + 1 == ft_end)) {
/* After raw segmentation, we must fix some header
* fields and compute checksums, in a protocol dependent
* way. */
gso_fix_segment(dst, gso_bytes, gso_idx,
segmented_bytes,
src_len == 0 && ft_p + 1 == ft_end,
tcp, iphlen);
ND("frame %u completed with %d bytes", gso_idx, (int)gso_bytes);
slot->len = gso_bytes;
slot->flags = 0;
segmented_bytes += gso_bytes - gso_hdr_len;
dst_slots++;
/* Next destination slot. */
*j = nm_next(*j, lim);
slot = &ring->slot[*j];
dst = NMB(&dst_na->up, slot);
gso_bytes = 0;
gso_idx++;
}
/* Next input slot. */
if (src_len == 0) {
ft_p++;
if (ft_p == ft_end)
break;
src = ft_p->ft_buf;
src_len = ft_p->ft_len;
}
}
ND(3, "%d bytes segmented", segmented_bytes);
} else {
/* Address of a checksum field into a destination slot. */
uint16_t *check = NULL;
/* Accumulator for an unfolded checksum. */
rawsum_t csum = 0;
/* Process a non-GSO packet. */
/* Init 'check' if necessary. */
if (vh && (vh->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)) {
if (unlikely(vh->csum_offset + vh->csum_start > src_len))
D("invalid checksum request");
else
check = (uint16_t *)(dst + vh->csum_start +
vh->csum_offset);
}
while (ft_p != ft_end) {
/* Init/update the packet checksum if needed. */
if (vh && (vh->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)) {
if (!dst_slots)
csum = nm_csum_raw(src + vh->csum_start,
src_len - vh->csum_start, 0);
else
csum = nm_csum_raw(src, src_len, csum);
}
/* Round to a multiple of 64 */
src_len = (src_len + 63) & ~63;
if (ft_p->ft_flags & NS_INDIRECT) {
if (copyin(src, dst, src_len)) {
/* Invalid user pointer, pretend len is 0. */
dst_len = 0;
}
} else {
memcpy(dst, src, (int)src_len);
}
slot->len = dst_len;
dst_slots++;
/* Next destination slot. */
*j = nm_next(*j, lim);
slot = &ring->slot[*j];
dst = NMB(&dst_na->up, slot);
/* Next source slot. */
ft_p++;
src = ft_p->ft_buf;
dst_len = src_len = ft_p->ft_len;
}
/* Finalize (fold) the checksum if needed. */
if (check && vh && (vh->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)) {
*check = nm_csum_fold(csum);
}
ND(3, "using %u dst_slots", dst_slots);
/* A second pass on the desitations slots to set the slot flags,
* using the right number of destination slots.
*/
while (j_start != *j) {
slot = &ring->slot[j_start];
slot->flags = (dst_slots << 8)| NS_MOREFRAG;
j_start = nm_next(j_start, lim);
}
/* Clear NS_MOREFRAG flag on last entry. */
slot->flags = (dst_slots << 8);
}
/* Update howmany. */
if (unlikely(dst_slots > *howmany)) {
dst_slots = *howmany;
D("Slot allocation error: Should never happen");
}
*howmany -= dst_slots;
}