freebsd-skq/sys/net/if_vlan.c

2171 lines
54 KiB
C
Raw Normal View History

/*-
* Copyright 1998 Massachusetts Institute of Technology
* Copyright 2012 ADARA Networks, Inc.
* Copyright 2017 Dell EMC Isilon
*
* Portions of this software were developed by Robert N. M. Watson under
* contract to ADARA Networks, Inc.
*
* Permission to use, copy, modify, and distribute this software and
* its documentation for any purpose and without fee is hereby
* granted, provided that both the above copyright notice and this
* permission notice appear in all copies, that both the above
* copyright notice and this permission notice appear in all
* supporting documentation, and that the name of M.I.T. not be used
* in advertising or publicity pertaining to distribution of the
* software without specific, written prior permission. M.I.T. makes
* no representations about the suitability of this software for any
* purpose. It is provided "as is" without express or implied
* warranty.
*
* THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS
* ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
* SHALL M.I.T. 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.
*/
/*
* if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
* This is sort of sneaky in the implementation, since
* we need to pretend to be enough of an Ethernet implementation
* to make arp work. The way we do this is by telling everyone
* that we are an Ethernet, and then catch the packets that
* ether_output() sends to us via if_transmit(), rewrite them for
* use by the real outgoing interface, and ask it to send them.
*/
2010-02-21 00:07:45 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_inet.h"
#include "opt_inet6.h"
Add kernel-side support for in-kernel TLS. KTLS adds support for in-kernel framing and encryption of Transport Layer Security (1.0-1.2) data on TCP sockets. KTLS only supports offload of TLS for transmitted data. Key negotation must still be performed in userland. Once completed, transmit session keys for a connection are provided to the kernel via a new TCP_TXTLS_ENABLE socket option. All subsequent data transmitted on the socket is placed into TLS frames and encrypted using the supplied keys. Any data written to a KTLS-enabled socket via write(2), aio_write(2), or sendfile(2) is assumed to be application data and is encoded in TLS frames with an application data type. Individual records can be sent with a custom type (e.g. handshake messages) via sendmsg(2) with a new control message (TLS_SET_RECORD_TYPE) specifying the record type. At present, rekeying is not supported though the in-kernel framework should support rekeying. KTLS makes use of the recently added unmapped mbufs to store TLS frames in the socket buffer. Each TLS frame is described by a single ext_pgs mbuf. The ext_pgs structure contains the header of the TLS record (and trailer for encrypted records) as well as references to the associated TLS session. KTLS supports two primary methods of encrypting TLS frames: software TLS and ifnet TLS. Software TLS marks mbufs holding socket data as not ready via M_NOTREADY similar to sendfile(2) when TLS framing information is added to an unmapped mbuf in ktls_frame(). ktls_enqueue() is then called to schedule TLS frames for encryption. In the case of sendfile_iodone() calls ktls_enqueue() instead of pru_ready() leaving the mbufs marked M_NOTREADY until encryption is completed. For other writes (vn_sendfile when pages are available, write(2), etc.), the PRUS_NOTREADY is set when invoking pru_send() along with invoking ktls_enqueue(). A pool of worker threads (the "KTLS" kernel process) encrypts TLS frames queued via ktls_enqueue(). Each TLS frame is temporarily mapped using the direct map and passed to a software encryption backend to perform the actual encryption. (Note: The use of PHYS_TO_DMAP could be replaced with sf_bufs if someone wished to make this work on architectures without a direct map.) KTLS supports pluggable software encryption backends. Internally, Netflix uses proprietary pure-software backends. This commit includes a simple backend in a new ktls_ocf.ko module that uses the kernel's OpenCrypto framework to provide AES-GCM encryption of TLS frames. As a result, software TLS is now a bit of a misnomer as it can make use of hardware crypto accelerators. Once software encryption has finished, the TLS frame mbufs are marked ready via pru_ready(). At this point, the encrypted data appears as regular payload to the TCP stack stored in unmapped mbufs. ifnet TLS permits a NIC to offload the TLS encryption and TCP segmentation. In this mode, a new send tag type (IF_SND_TAG_TYPE_TLS) is allocated on the interface a socket is routed over and associated with a TLS session. TLS records for a TLS session using ifnet TLS are not marked M_NOTREADY but are passed down the stack unencrypted. The ip_output_send() and ip6_output_send() helper functions that apply send tags to outbound IP packets verify that the send tag of the TLS record matches the outbound interface. If so, the packet is tagged with the TLS send tag and sent to the interface. The NIC device driver must recognize packets with the TLS send tag and schedule them for TLS encryption and TCP segmentation. If the the outbound interface does not match the interface in the TLS send tag, the packet is dropped. In addition, a task is scheduled to refresh the TLS send tag for the TLS session. If a new TLS send tag cannot be allocated, the connection is dropped. If a new TLS send tag is allocated, however, subsequent packets will be tagged with the correct TLS send tag. (This latter case has been tested by configuring both ports of a Chelsio T6 in a lagg and failing over from one port to another. As the connections migrated to the new port, new TLS send tags were allocated for the new port and connections resumed without being dropped.) ifnet TLS can be enabled and disabled on supported network interfaces via new '[-]txtls[46]' options to ifconfig(8). ifnet TLS is supported across both vlan devices and lagg interfaces using failover, lacp with flowid enabled, or lacp with flowid enabled. Applications may request the current KTLS mode of a connection via a new TCP_TXTLS_MODE socket option. They can also use this socket option to toggle between software and ifnet TLS modes. In addition, a testing tool is available in tools/tools/switch_tls. This is modeled on tcpdrop and uses similar syntax. However, instead of dropping connections, -s is used to force KTLS connections to switch to software TLS and -i is used to switch to ifnet TLS. Various sysctls and counters are available under the kern.ipc.tls sysctl node. The kern.ipc.tls.enable node must be set to true to enable KTLS (it is off by default). The use of unmapped mbufs must also be enabled via kern.ipc.mb_use_ext_pgs to enable KTLS. KTLS is enabled via the KERN_TLS kernel option. This patch is the culmination of years of work by several folks including Scott Long and Randall Stewart for the original design and implementation; Drew Gallatin for several optimizations including the use of ext_pgs mbufs, the M_NOTREADY mechanism for TLS records awaiting software encryption, and pluggable software crypto backends; and John Baldwin for modifications to support hardware TLS offload. Reviewed by: gallatin, hselasky, rrs Obtained from: Netflix Sponsored by: Netflix, Chelsio Communications Differential Revision: https://reviews.freebsd.org/D21277
2019-08-27 00:01:56 +00:00
#include "opt_kern_tls.h"
2006-01-30 13:45:15 +00:00
#include "opt_vlan.h"
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
#include "opt_ratelimit.h"
#include <sys/param.h>
#include <sys/eventhandler.h>
#include <sys/kernel.h>
2006-01-30 13:45:15 +00:00
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/module.h>
#include <sys/rmlock.h>
#include <sys/priv.h>
#include <sys/queue.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <sys/systm.h>
#include <sys/sx.h>
#include <sys/taskqueue.h>
#include <net/bpf.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <net/if_var.h>
#include <net/if_clone.h>
#include <net/if_dl.h>
#include <net/if_types.h>
#include <net/if_vlan_var.h>
#include <net/route.h>
#include <net/vnet.h>
#ifdef INET
#include <netinet/in.h>
#include <netinet/if_ether.h>
#endif
#ifdef INET6
/*
* XXX: declare here to avoid to include many inet6 related files..
* should be more generalized?
*/
extern void nd6_setmtu(struct ifnet *);
#endif
2006-01-30 13:45:15 +00:00
#define VLAN_DEF_HWIDTH 4
#define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST)
2006-01-30 13:45:15 +00:00
#define UP_AND_RUNNING(ifp) \
((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
CK_SLIST_HEAD(ifvlanhead, ifvlan);
2006-01-30 13:45:15 +00:00
struct ifvlantrunk {
struct ifnet *parent; /* parent interface of this trunk */
struct mtx lock;
2006-01-30 13:45:15 +00:00
#ifdef VLAN_ARRAY
#define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1)
struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */
2006-01-30 13:45:15 +00:00
#else
struct ifvlanhead *hash; /* dynamic hash-list table */
uint16_t hmask;
uint16_t hwidth;
#endif
int refcnt;
};
Add kernel-side support for in-kernel TLS. KTLS adds support for in-kernel framing and encryption of Transport Layer Security (1.0-1.2) data on TCP sockets. KTLS only supports offload of TLS for transmitted data. Key negotation must still be performed in userland. Once completed, transmit session keys for a connection are provided to the kernel via a new TCP_TXTLS_ENABLE socket option. All subsequent data transmitted on the socket is placed into TLS frames and encrypted using the supplied keys. Any data written to a KTLS-enabled socket via write(2), aio_write(2), or sendfile(2) is assumed to be application data and is encoded in TLS frames with an application data type. Individual records can be sent with a custom type (e.g. handshake messages) via sendmsg(2) with a new control message (TLS_SET_RECORD_TYPE) specifying the record type. At present, rekeying is not supported though the in-kernel framework should support rekeying. KTLS makes use of the recently added unmapped mbufs to store TLS frames in the socket buffer. Each TLS frame is described by a single ext_pgs mbuf. The ext_pgs structure contains the header of the TLS record (and trailer for encrypted records) as well as references to the associated TLS session. KTLS supports two primary methods of encrypting TLS frames: software TLS and ifnet TLS. Software TLS marks mbufs holding socket data as not ready via M_NOTREADY similar to sendfile(2) when TLS framing information is added to an unmapped mbuf in ktls_frame(). ktls_enqueue() is then called to schedule TLS frames for encryption. In the case of sendfile_iodone() calls ktls_enqueue() instead of pru_ready() leaving the mbufs marked M_NOTREADY until encryption is completed. For other writes (vn_sendfile when pages are available, write(2), etc.), the PRUS_NOTREADY is set when invoking pru_send() along with invoking ktls_enqueue(). A pool of worker threads (the "KTLS" kernel process) encrypts TLS frames queued via ktls_enqueue(). Each TLS frame is temporarily mapped using the direct map and passed to a software encryption backend to perform the actual encryption. (Note: The use of PHYS_TO_DMAP could be replaced with sf_bufs if someone wished to make this work on architectures without a direct map.) KTLS supports pluggable software encryption backends. Internally, Netflix uses proprietary pure-software backends. This commit includes a simple backend in a new ktls_ocf.ko module that uses the kernel's OpenCrypto framework to provide AES-GCM encryption of TLS frames. As a result, software TLS is now a bit of a misnomer as it can make use of hardware crypto accelerators. Once software encryption has finished, the TLS frame mbufs are marked ready via pru_ready(). At this point, the encrypted data appears as regular payload to the TCP stack stored in unmapped mbufs. ifnet TLS permits a NIC to offload the TLS encryption and TCP segmentation. In this mode, a new send tag type (IF_SND_TAG_TYPE_TLS) is allocated on the interface a socket is routed over and associated with a TLS session. TLS records for a TLS session using ifnet TLS are not marked M_NOTREADY but are passed down the stack unencrypted. The ip_output_send() and ip6_output_send() helper functions that apply send tags to outbound IP packets verify that the send tag of the TLS record matches the outbound interface. If so, the packet is tagged with the TLS send tag and sent to the interface. The NIC device driver must recognize packets with the TLS send tag and schedule them for TLS encryption and TCP segmentation. If the the outbound interface does not match the interface in the TLS send tag, the packet is dropped. In addition, a task is scheduled to refresh the TLS send tag for the TLS session. If a new TLS send tag cannot be allocated, the connection is dropped. If a new TLS send tag is allocated, however, subsequent packets will be tagged with the correct TLS send tag. (This latter case has been tested by configuring both ports of a Chelsio T6 in a lagg and failing over from one port to another. As the connections migrated to the new port, new TLS send tags were allocated for the new port and connections resumed without being dropped.) ifnet TLS can be enabled and disabled on supported network interfaces via new '[-]txtls[46]' options to ifconfig(8). ifnet TLS is supported across both vlan devices and lagg interfaces using failover, lacp with flowid enabled, or lacp with flowid enabled. Applications may request the current KTLS mode of a connection via a new TCP_TXTLS_MODE socket option. They can also use this socket option to toggle between software and ifnet TLS modes. In addition, a testing tool is available in tools/tools/switch_tls. This is modeled on tcpdrop and uses similar syntax. However, instead of dropping connections, -s is used to force KTLS connections to switch to software TLS and -i is used to switch to ifnet TLS. Various sysctls and counters are available under the kern.ipc.tls sysctl node. The kern.ipc.tls.enable node must be set to true to enable KTLS (it is off by default). The use of unmapped mbufs must also be enabled via kern.ipc.mb_use_ext_pgs to enable KTLS. KTLS is enabled via the KERN_TLS kernel option. This patch is the culmination of years of work by several folks including Scott Long and Randall Stewart for the original design and implementation; Drew Gallatin for several optimizations including the use of ext_pgs mbufs, the M_NOTREADY mechanism for TLS records awaiting software encryption, and pluggable software crypto backends; and John Baldwin for modifications to support hardware TLS offload. Reviewed by: gallatin, hselasky, rrs Obtained from: Netflix Sponsored by: Netflix, Chelsio Communications Differential Revision: https://reviews.freebsd.org/D21277
2019-08-27 00:01:56 +00:00
#if defined(KERN_TLS) || defined(RATELIMIT)
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
struct vlan_snd_tag {
struct m_snd_tag com;
struct m_snd_tag *tag;
};
static inline struct vlan_snd_tag *
mst_to_vst(struct m_snd_tag *mst)
{
return (__containerof(mst, struct vlan_snd_tag, com));
}
#endif
/*
* This macro provides a facility to iterate over every vlan on a trunk with
* the assumption that none will be added/removed during iteration.
*/
#ifdef VLAN_ARRAY
#define VLAN_FOREACH(_ifv, _trunk) \
size_t _i; \
for (_i = 0; _i < VLAN_ARRAY_SIZE; _i++) \
if (((_ifv) = (_trunk)->vlans[_i]) != NULL)
#else /* VLAN_ARRAY */
#define VLAN_FOREACH(_ifv, _trunk) \
struct ifvlan *_next; \
size_t _i; \
for (_i = 0; _i < (1 << (_trunk)->hwidth); _i++) \
CK_SLIST_FOREACH_SAFE((_ifv), &(_trunk)->hash[_i], ifv_list, _next)
#endif /* VLAN_ARRAY */
/*
* This macro provides a facility to iterate over every vlan on a trunk while
* also modifying the number of vlans on the trunk. The iteration continues
* until some condition is met or there are no more vlans on the trunk.
*/
#ifdef VLAN_ARRAY
/* The VLAN_ARRAY case is simple -- just a for loop using the condition. */
#define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
size_t _i; \
for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \
if (((_ifv) = (_trunk)->vlans[_i]))
#else /* VLAN_ARRAY */
/*
* The hash table case is more complicated. We allow for the hash table to be
* modified (i.e. vlans removed) while we are iterating over it. To allow for
* this we must restart the iteration every time we "touch" something during
* the iteration, since removal will resize the hash table and invalidate our
* current position. If acting on the touched element causes the trunk to be
* emptied, then iteration also stops.
*/
#define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
size_t _i; \
bool _touch = false; \
for (_i = 0; \
!(_cond) && _i < (1 << (_trunk)->hwidth); \
_i = (_touch && ((_trunk) != NULL) ? 0 : _i + 1), _touch = false) \
if (((_ifv) = CK_SLIST_FIRST(&(_trunk)->hash[_i])) != NULL && \
(_touch = true))
#endif /* VLAN_ARRAY */
struct vlan_mc_entry {
struct sockaddr_dl mc_addr;
CK_SLIST_ENTRY(vlan_mc_entry) mc_entries;
struct epoch_context mc_epoch_ctx;
};
struct ifvlan {
2006-01-30 13:45:15 +00:00
struct ifvlantrunk *ifv_trunk;
struct ifnet *ifv_ifp;
2006-01-30 13:45:15 +00:00
#define TRUNK(ifv) ((ifv)->ifv_trunk)
#define PARENT(ifv) (TRUNK(ifv)->parent)
void *ifv_cookie;
int ifv_pflags; /* special flags we have set on parent */
int ifv_capenable;
int ifv_encaplen; /* encapsulation length */
int ifv_mtufudge; /* MTU fudged by this much */
int ifv_mintu; /* min transmission unit */
struct ether_8021q_tag ifv_qtag;
#define ifv_proto ifv_qtag.proto
#define ifv_vid ifv_qtag.vid
#define ifv_pcp ifv_qtag.pcp
struct task lladdr_task;
CK_SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
#ifndef VLAN_ARRAY
CK_SLIST_ENTRY(ifvlan) ifv_list;
#endif
};
2006-01-30 13:45:15 +00:00
/* Special flags we should propagate to parent. */
static struct {
int flag;
int (*func)(struct ifnet *, int);
} vlan_pflags[] = {
{IFF_PROMISC, ifpromisc},
{IFF_ALLMULTI, if_allmulti},
{0, NULL}
};
extern int vlan_mtag_pcp;
static const char vlanname[] = "vlan";
static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
static eventhandler_tag ifdetach_tag;
static eventhandler_tag iflladdr_tag;
/*
* if_vlan uses two module-level synchronizations primitives to allow concurrent
* modification of vlan interfaces and (mostly) allow for vlans to be destroyed
* while they are being used for tx/rx. To accomplish this in a way that has
* acceptable performance and cooperation with other parts of the network stack
* there is a non-sleepable epoch(9) and an sx(9).
2006-01-30 13:45:15 +00:00
*
* The performance-sensitive paths that warrant using the epoch(9) are
* vlan_transmit and vlan_input. Both have to check for the vlan interface's
* existence using if_vlantrunk, and being in the network tx/rx paths the use
* of an epoch(9) gives a measureable improvement in performance.
*
* The reason for having an sx(9) is mostly because there are still areas that
* must be sleepable and also have safe concurrent access to a vlan interface.
* Since the sx(9) exists, it is used by default in most paths unless sleeping
* is not permitted, or if it is not clear whether sleeping is permitted.
2006-01-30 13:45:15 +00:00
*
*/
#define _VLAN_SX_ID ifv_sx
static struct sx _VLAN_SX_ID;
#define VLAN_LOCKING_INIT() \
sx_init_flags(&_VLAN_SX_ID, "vlan_sx", SX_RECURSE)
#define VLAN_LOCKING_DESTROY() \
sx_destroy(&_VLAN_SX_ID)
#define VLAN_SLOCK() sx_slock(&_VLAN_SX_ID)
#define VLAN_SUNLOCK() sx_sunlock(&_VLAN_SX_ID)
#define VLAN_XLOCK() sx_xlock(&_VLAN_SX_ID)
#define VLAN_XUNLOCK() sx_xunlock(&_VLAN_SX_ID)
#define VLAN_SLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_SLOCKED)
#define VLAN_XLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_XLOCKED)
#define VLAN_SXLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_LOCKED)
/*
* We also have a per-trunk mutex that should be acquired when changing
* its state.
*/
#define TRUNK_LOCK_INIT(trunk) mtx_init(&(trunk)->lock, vlanname, NULL, MTX_DEF)
#define TRUNK_LOCK_DESTROY(trunk) mtx_destroy(&(trunk)->lock)
#define TRUNK_WLOCK(trunk) mtx_lock(&(trunk)->lock)
#define TRUNK_WUNLOCK(trunk) mtx_unlock(&(trunk)->lock)
#define TRUNK_WLOCK_ASSERT(trunk) mtx_assert(&(trunk)->lock, MA_OWNED);
/*
2006-01-30 13:45:15 +00:00
* The VLAN_ARRAY substitutes the dynamic hash with a static array
2006-08-25 08:25:35 +00:00
* with 4096 entries. In theory this can give a boost in processing,
* however in practice it does not. Probably this is because the array
2006-01-30 13:45:15 +00:00
* is too big to fit into CPU cache.
*/
2006-01-30 13:45:15 +00:00
#ifndef VLAN_ARRAY
static void vlan_inithash(struct ifvlantrunk *trunk);
static void vlan_freehash(struct ifvlantrunk *trunk);
static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
uint16_t vid);
2006-01-30 13:45:15 +00:00
#endif
static void trunk_destroy(struct ifvlantrunk *trunk);
static void vlan_init(void *foo);
static void vlan_input(struct ifnet *ifp, struct mbuf *m);
static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
Add kernel-side support for in-kernel TLS. KTLS adds support for in-kernel framing and encryption of Transport Layer Security (1.0-1.2) data on TCP sockets. KTLS only supports offload of TLS for transmitted data. Key negotation must still be performed in userland. Once completed, transmit session keys for a connection are provided to the kernel via a new TCP_TXTLS_ENABLE socket option. All subsequent data transmitted on the socket is placed into TLS frames and encrypted using the supplied keys. Any data written to a KTLS-enabled socket via write(2), aio_write(2), or sendfile(2) is assumed to be application data and is encoded in TLS frames with an application data type. Individual records can be sent with a custom type (e.g. handshake messages) via sendmsg(2) with a new control message (TLS_SET_RECORD_TYPE) specifying the record type. At present, rekeying is not supported though the in-kernel framework should support rekeying. KTLS makes use of the recently added unmapped mbufs to store TLS frames in the socket buffer. Each TLS frame is described by a single ext_pgs mbuf. The ext_pgs structure contains the header of the TLS record (and trailer for encrypted records) as well as references to the associated TLS session. KTLS supports two primary methods of encrypting TLS frames: software TLS and ifnet TLS. Software TLS marks mbufs holding socket data as not ready via M_NOTREADY similar to sendfile(2) when TLS framing information is added to an unmapped mbuf in ktls_frame(). ktls_enqueue() is then called to schedule TLS frames for encryption. In the case of sendfile_iodone() calls ktls_enqueue() instead of pru_ready() leaving the mbufs marked M_NOTREADY until encryption is completed. For other writes (vn_sendfile when pages are available, write(2), etc.), the PRUS_NOTREADY is set when invoking pru_send() along with invoking ktls_enqueue(). A pool of worker threads (the "KTLS" kernel process) encrypts TLS frames queued via ktls_enqueue(). Each TLS frame is temporarily mapped using the direct map and passed to a software encryption backend to perform the actual encryption. (Note: The use of PHYS_TO_DMAP could be replaced with sf_bufs if someone wished to make this work on architectures without a direct map.) KTLS supports pluggable software encryption backends. Internally, Netflix uses proprietary pure-software backends. This commit includes a simple backend in a new ktls_ocf.ko module that uses the kernel's OpenCrypto framework to provide AES-GCM encryption of TLS frames. As a result, software TLS is now a bit of a misnomer as it can make use of hardware crypto accelerators. Once software encryption has finished, the TLS frame mbufs are marked ready via pru_ready(). At this point, the encrypted data appears as regular payload to the TCP stack stored in unmapped mbufs. ifnet TLS permits a NIC to offload the TLS encryption and TCP segmentation. In this mode, a new send tag type (IF_SND_TAG_TYPE_TLS) is allocated on the interface a socket is routed over and associated with a TLS session. TLS records for a TLS session using ifnet TLS are not marked M_NOTREADY but are passed down the stack unencrypted. The ip_output_send() and ip6_output_send() helper functions that apply send tags to outbound IP packets verify that the send tag of the TLS record matches the outbound interface. If so, the packet is tagged with the TLS send tag and sent to the interface. The NIC device driver must recognize packets with the TLS send tag and schedule them for TLS encryption and TCP segmentation. If the the outbound interface does not match the interface in the TLS send tag, the packet is dropped. In addition, a task is scheduled to refresh the TLS send tag for the TLS session. If a new TLS send tag cannot be allocated, the connection is dropped. If a new TLS send tag is allocated, however, subsequent packets will be tagged with the correct TLS send tag. (This latter case has been tested by configuring both ports of a Chelsio T6 in a lagg and failing over from one port to another. As the connections migrated to the new port, new TLS send tags were allocated for the new port and connections resumed without being dropped.) ifnet TLS can be enabled and disabled on supported network interfaces via new '[-]txtls[46]' options to ifconfig(8). ifnet TLS is supported across both vlan devices and lagg interfaces using failover, lacp with flowid enabled, or lacp with flowid enabled. Applications may request the current KTLS mode of a connection via a new TCP_TXTLS_MODE socket option. They can also use this socket option to toggle between software and ifnet TLS modes. In addition, a testing tool is available in tools/tools/switch_tls. This is modeled on tcpdrop and uses similar syntax. However, instead of dropping connections, -s is used to force KTLS connections to switch to software TLS and -i is used to switch to ifnet TLS. Various sysctls and counters are available under the kern.ipc.tls sysctl node. The kern.ipc.tls.enable node must be set to true to enable KTLS (it is off by default). The use of unmapped mbufs must also be enabled via kern.ipc.mb_use_ext_pgs to enable KTLS. KTLS is enabled via the KERN_TLS kernel option. This patch is the culmination of years of work by several folks including Scott Long and Randall Stewart for the original design and implementation; Drew Gallatin for several optimizations including the use of ext_pgs mbufs, the M_NOTREADY mechanism for TLS records awaiting software encryption, and pluggable software crypto backends; and John Baldwin for modifications to support hardware TLS offload. Reviewed by: gallatin, hselasky, rrs Obtained from: Netflix Sponsored by: Netflix, Chelsio Communications Differential Revision: https://reviews.freebsd.org/D21277
2019-08-27 00:01:56 +00:00
#if defined(KERN_TLS) || defined(RATELIMIT)
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
static int vlan_snd_tag_alloc(struct ifnet *,
union if_snd_tag_alloc_params *, struct m_snd_tag **);
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
static int vlan_snd_tag_modify(struct m_snd_tag *,
union if_snd_tag_modify_params *);
static int vlan_snd_tag_query(struct m_snd_tag *,
union if_snd_tag_query_params *);
static void vlan_snd_tag_free(struct m_snd_tag *);
static struct m_snd_tag *vlan_next_snd_tag(struct m_snd_tag *);
static void vlan_ratelimit_query(struct ifnet *,
struct if_ratelimit_query_results *);
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
#endif
static void vlan_qflush(struct ifnet *ifp);
static int vlan_setflag(struct ifnet *ifp, int flag, int status,
int (*func)(struct ifnet *, int));
static int vlan_setflags(struct ifnet *ifp, int status);
static int vlan_setmulti(struct ifnet *ifp);
static int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
static int vlan_output(struct ifnet *ifp, struct mbuf *m,
const struct sockaddr *dst, struct route *ro);
static void vlan_unconfig(struct ifnet *ifp);
static void vlan_unconfig_locked(struct ifnet *ifp, int departing);
static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag,
uint16_t proto);
static void vlan_link_state(struct ifnet *ifp);
2006-01-30 13:45:15 +00:00
static void vlan_capabilities(struct ifvlan *ifv);
static void vlan_trunk_capabilities(struct ifnet *ifp);
static struct ifnet *vlan_clone_match_ethervid(const char *, int *);
static int vlan_clone_match(struct if_clone *, const char *);
static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
static int vlan_clone_destroy(struct if_clone *, struct ifnet *);
static void vlan_ifdetach(void *arg, struct ifnet *ifp);
static void vlan_iflladdr(void *arg, struct ifnet *ifp);
static void vlan_lladdr_fn(void *arg, int pending);
static struct if_clone *vlan_cloner;
#ifdef VIMAGE
VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
#define V_vlan_cloner VNET(vlan_cloner)
#endif
static void
vlan_mc_free(struct epoch_context *ctx)
{
struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx);
free(mc, M_VLAN);
}
#ifndef VLAN_ARRAY
#define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
2006-01-30 13:45:15 +00:00
static void
vlan_inithash(struct ifvlantrunk *trunk)
{
int i, n;
2006-01-30 13:45:15 +00:00
/*
* The trunk must not be locked here since we call malloc(M_WAITOK).
* It is OK in case this function is called before the trunk struct
* gets hooked up and becomes visible from other threads.
*/
KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
("%s: hash already initialized", __func__));
trunk->hwidth = VLAN_DEF_HWIDTH;
n = 1 << trunk->hwidth;
trunk->hmask = n - 1;
trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
for (i = 0; i < n; i++)
CK_SLIST_INIT(&trunk->hash[i]);
2006-01-30 13:45:15 +00:00
}
static void
vlan_freehash(struct ifvlantrunk *trunk)
{
#ifdef INVARIANTS
int i;
KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
for (i = 0; i < (1 << trunk->hwidth); i++)
KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
2006-01-30 13:45:15 +00:00
("%s: hash table not empty", __func__));
#endif
free(trunk->hash, M_VLAN);
trunk->hash = NULL;
trunk->hwidth = trunk->hmask = 0;
}
static int
vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
{
int i, b;
struct ifvlan *ifv2;
VLAN_XLOCK_ASSERT();
2006-01-30 13:45:15 +00:00
KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
b = 1 << trunk->hwidth;
i = HASH(ifv->ifv_vid, trunk->hmask);
CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
if (ifv->ifv_vid == ifv2->ifv_vid)
2006-01-30 13:45:15 +00:00
return (EEXIST);
/*
* Grow the hash when the number of vlans exceeds half of the number of
* hash buckets squared. This will make the average linked-list length
* buckets/2.
*/
if (trunk->refcnt > (b * b) / 2) {
vlan_growhash(trunk, 1);
i = HASH(ifv->ifv_vid, trunk->hmask);
2006-01-30 13:45:15 +00:00
}
CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
2006-01-30 13:45:15 +00:00
trunk->refcnt++;
return (0);
}
static int
vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
{
int i, b;
struct ifvlan *ifv2;
VLAN_XLOCK_ASSERT();
2006-01-30 13:45:15 +00:00
KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
2006-01-30 13:45:15 +00:00
b = 1 << trunk->hwidth;
i = HASH(ifv->ifv_vid, trunk->hmask);
CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
2006-01-30 13:45:15 +00:00
if (ifv2 == ifv) {
trunk->refcnt--;
CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
2006-01-30 13:45:15 +00:00
if (trunk->refcnt < (b * b) / 2)
vlan_growhash(trunk, -1);
return (0);
}
panic("%s: vlan not found\n", __func__);
return (ENOENT); /*NOTREACHED*/
}
/*
* Grow the hash larger or smaller if memory permits.
*/
static void
vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
{
struct ifvlan *ifv;
struct ifvlanhead *hash2;
int hwidth2, i, j, n, n2;
VLAN_XLOCK_ASSERT();
2006-01-30 13:45:15 +00:00
KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
if (howmuch == 0) {
/* Harmless yet obvious coding error */
printf("%s: howmuch is 0\n", __func__);
return;
}
hwidth2 = trunk->hwidth + howmuch;
n = 1 << trunk->hwidth;
n2 = 1 << hwidth2;
/* Do not shrink the table below the default */
if (hwidth2 < VLAN_DEF_HWIDTH)
return;
hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
2006-01-30 13:45:15 +00:00
if (hash2 == NULL) {
printf("%s: out of memory -- hash size not changed\n",
__func__);
return; /* We can live with the old hash table */
}
for (j = 0; j < n2; j++)
CK_SLIST_INIT(&hash2[j]);
2006-01-30 13:45:15 +00:00
for (i = 0; i < n; i++)
while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
j = HASH(ifv->ifv_vid, n2 - 1);
CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
2006-01-30 13:45:15 +00:00
}
NET_EPOCH_WAIT();
2006-01-30 13:45:15 +00:00
free(trunk->hash, M_VLAN);
trunk->hash = hash2;
trunk->hwidth = hwidth2;
trunk->hmask = n2 - 1;
if (bootverbose)
if_printf(trunk->parent,
"VLAN hash table resized from %d to %d buckets\n", n, n2);
2006-01-30 13:45:15 +00:00
}
static __inline struct ifvlan *
vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
2006-01-30 13:45:15 +00:00
{
struct ifvlan *ifv;
NET_EPOCH_ASSERT();
2006-01-30 13:45:15 +00:00
CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
if (ifv->ifv_vid == vid)
2006-01-30 13:45:15 +00:00
return (ifv);
return (NULL);
}
#if 0
/* Debugging code to view the hashtables. */
static void
vlan_dumphash(struct ifvlantrunk *trunk)
{
int i;
struct ifvlan *ifv;
for (i = 0; i < (1 << trunk->hwidth); i++) {
printf("%d: ", i);
CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
2006-01-30 13:45:15 +00:00
printf("%s ", ifv->ifv_ifp->if_xname);
printf("\n");
}
}
#endif /* 0 */
#else
static __inline struct ifvlan *
vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
{
return trunk->vlans[vid];
}
static __inline int
vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
{
if (trunk->vlans[ifv->ifv_vid] != NULL)
return EEXIST;
trunk->vlans[ifv->ifv_vid] = ifv;
trunk->refcnt++;
return (0);
}
static __inline int
vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
{
trunk->vlans[ifv->ifv_vid] = NULL;
trunk->refcnt--;
return (0);
}
static __inline void
vlan_freehash(struct ifvlantrunk *trunk)
{
}
static __inline void
vlan_inithash(struct ifvlantrunk *trunk)
{
}
2006-01-30 13:45:15 +00:00
#endif /* !VLAN_ARRAY */
static void
trunk_destroy(struct ifvlantrunk *trunk)
{
VLAN_XLOCK_ASSERT();
2006-01-30 13:45:15 +00:00
vlan_freehash(trunk);
trunk->parent->if_vlantrunk = NULL;
TRUNK_LOCK_DESTROY(trunk);
if_rele(trunk->parent);
2006-01-30 13:45:15 +00:00
free(trunk, M_VLAN);
}
/*
* Program our multicast filter. What we're actually doing is
* programming the multicast filter of the parent. This has the
* side effect of causing the parent interface to receive multicast
* traffic that it doesn't really want, which ends up being discarded
* later by the upper protocol layers. Unfortunately, there's no way
* to avoid this: there really is only one physical interface.
*/
static int
vlan_setmulti(struct ifnet *ifp)
{
struct ifnet *ifp_p;
struct ifmultiaddr *ifma;
struct ifvlan *sc;
struct vlan_mc_entry *mc;
int error;
VLAN_XLOCK_ASSERT();
/* Find the parent. */
sc = ifp->if_softc;
2006-01-30 13:45:15 +00:00
ifp_p = PARENT(sc);
CURVNET_SET_QUIET(ifp_p->if_vnet);
/* First, remove any existing filter entries. */
while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
(void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
}
/* Now program new ones. */
IF_ADDR_WLOCK(ifp);
ifnet: Replace if_addr_lock rwlock with epoch + mutex Run on LLNW canaries and tested by pho@ gallatin: Using a 14-core, 28-HTT single socket E5-2697 v3 with a 40GbE MLX5 based ConnectX 4-LX NIC, I see an almost 12% improvement in received packet rate, and a larger improvement in bytes delivered all the way to userspace. When the host receiving 64 streams of netperf -H $DUT -t UDP_STREAM -- -m 1, I see, using nstat -I mce0 1 before the patch: InMpps OMpps InGbs OGbs err TCP Est %CPU syscalls csw irq GBfree 4.98 0.00 4.42 0.00 4235592 33 83.80 4720653 2149771 1235 247.32 4.73 0.00 4.20 0.00 4025260 33 82.99 4724900 2139833 1204 247.32 4.72 0.00 4.20 0.00 4035252 33 82.14 4719162 2132023 1264 247.32 4.71 0.00 4.21 0.00 4073206 33 83.68 4744973 2123317 1347 247.32 4.72 0.00 4.21 0.00 4061118 33 80.82 4713615 2188091 1490 247.32 4.72 0.00 4.21 0.00 4051675 33 85.29 4727399 2109011 1205 247.32 4.73 0.00 4.21 0.00 4039056 33 84.65 4724735 2102603 1053 247.32 After the patch InMpps OMpps InGbs OGbs err TCP Est %CPU syscalls csw irq GBfree 5.43 0.00 4.20 0.00 3313143 33 84.96 5434214 1900162 2656 245.51 5.43 0.00 4.20 0.00 3308527 33 85.24 5439695 1809382 2521 245.51 5.42 0.00 4.19 0.00 3316778 33 87.54 5416028 1805835 2256 245.51 5.42 0.00 4.19 0.00 3317673 33 90.44 5426044 1763056 2332 245.51 5.42 0.00 4.19 0.00 3314839 33 88.11 5435732 1792218 2499 245.52 5.44 0.00 4.19 0.00 3293228 33 91.84 5426301 1668597 2121 245.52 Similarly, netperf reports 230Mb/s before the patch, and 270Mb/s after the patch Reviewed by: gallatin Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D15366
2018-05-18 20:13:34 +00:00
CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
if (mc == NULL) {
IF_ADDR_WUNLOCK(ifp);
return (ENOMEM);
}
bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
mc->mc_addr.sdl_index = ifp_p->if_index;
CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
}
IF_ADDR_WUNLOCK(ifp);
CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
NULL);
if (error)
return (error);
}
CURVNET_RESTORE();
return (0);
}
/*
* A handler for parent interface link layer address changes.
* If the parent interface link layer address is changed we
* should also change it on all children vlans.
*/
static void
vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
{
struct epoch_tracker et;
struct ifvlan *ifv;
struct ifnet *ifv_ifp;
struct ifvlantrunk *trunk;
struct sockaddr_dl *sdl;
2019-01-10 00:37:14 +00:00
/* Need the epoch since this is run on taskqueue_swi. */
NET_EPOCH_ENTER(et);
trunk = ifp->if_vlantrunk;
if (trunk == NULL) {
NET_EPOCH_EXIT(et);
return;
}
/*
* OK, it's a trunk. Loop over and change all vlan's lladdrs on it.
* We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
* ioctl calls on the parent garbling the lladdr of the child vlan.
*/
TRUNK_WLOCK(trunk);
VLAN_FOREACH(ifv, trunk) {
/*
* Copy new new lladdr into the ifv_ifp, enqueue a task
* to actually call if_setlladdr. if_setlladdr needs to
* be deferred to a taskqueue because it will call into
* the if_vlan ioctl path and try to acquire the global
* lock.
*/
ifv_ifp = ifv->ifv_ifp;
bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
ifp->if_addrlen);
sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
sdl->sdl_alen = ifp->if_addrlen;
taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
}
TRUNK_WUNLOCK(trunk);
NET_EPOCH_EXIT(et);
}
/*
* A handler for network interface departure events.
* Track departure of trunks here so that we don't access invalid
* pointers or whatever if a trunk is ripped from under us, e.g.,
* by ejecting its hot-plug card. However, if an ifnet is simply
* being renamed, then there's no need to tear down the state.
*/
static void
vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
{
struct ifvlan *ifv;
struct ifvlantrunk *trunk;
/* If the ifnet is just being renamed, don't do anything. */
if (ifp->if_flags & IFF_RENAMING)
return;
VLAN_XLOCK();
trunk = ifp->if_vlantrunk;
if (trunk == NULL) {
VLAN_XUNLOCK();
return;
}
/*
* OK, it's a trunk. Loop over and detach all vlan's on it.
* Check trunk pointer after each vlan_unconfig() as it will
* free it and set to NULL after the last vlan was detached.
*/
VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
ifp->if_vlantrunk == NULL)
vlan_unconfig_locked(ifv->ifv_ifp, 1);
/* Trunk should have been destroyed in vlan_unconfig(). */
KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
VLAN_XUNLOCK();
}
/*
* Return the trunk device for a virtual interface.
*/
static struct ifnet *
vlan_trunkdev(struct ifnet *ifp)
{
struct ifvlan *ifv;
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
NET_EPOCH_ASSERT();
if (ifp->if_type != IFT_L2VLAN)
return (NULL);
ifv = ifp->if_softc;
ifp = NULL;
if (ifv->ifv_trunk)
ifp = PARENT(ifv);
return (ifp);
}
/*
* Return the 12-bit VLAN VID for this interface, for use by external
* components such as Infiniband.
*
* XXXRW: Note that the function name here is historical; it should be named
* vlan_vid().
*/
static int
vlan_tag(struct ifnet *ifp, uint16_t *vidp)
{
struct ifvlan *ifv;
if (ifp->if_type != IFT_L2VLAN)
return (EINVAL);
ifv = ifp->if_softc;
*vidp = ifv->ifv_vid;
return (0);
}
static int
vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
{
struct ifvlan *ifv;
if (ifp->if_type != IFT_L2VLAN)
return (EINVAL);
ifv = ifp->if_softc;
*pcpp = ifv->ifv_pcp;
return (0);
}
/*
* Return a driver specific cookie for this interface. Synchronization
* with setcookie must be provided by the driver.
*/
static void *
vlan_cookie(struct ifnet *ifp)
{
struct ifvlan *ifv;
if (ifp->if_type != IFT_L2VLAN)
return (NULL);
ifv = ifp->if_softc;
return (ifv->ifv_cookie);
}
/*
* Store a cookie in our softc that drivers can use to store driver
* private per-instance data in.
*/
static int
vlan_setcookie(struct ifnet *ifp, void *cookie)
{
struct ifvlan *ifv;
if (ifp->if_type != IFT_L2VLAN)
return (EINVAL);
ifv = ifp->if_softc;
ifv->ifv_cookie = cookie;
return (0);
}
/*
* Return the vlan device present at the specific VID.
*/
static struct ifnet *
vlan_devat(struct ifnet *ifp, uint16_t vid)
{
struct ifvlantrunk *trunk;
struct ifvlan *ifv;
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
NET_EPOCH_ASSERT();
trunk = ifp->if_vlantrunk;
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
if (trunk == NULL)
return (NULL);
ifp = NULL;
ifv = vlan_gethash(trunk, vid);
if (ifv)
ifp = ifv->ifv_ifp;
return (ifp);
}
/*
* VLAN support can be loaded as a module. The only place in the
* system that's intimately aware of this is ether_input. We hook
* into this code through vlan_input_p which is defined there and
* set here. No one else in the system should be aware of this so
* we use an explicit reference here.
*/
extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
/* For if_link_state_change() eyes only... */
extern void (*vlan_link_state_p)(struct ifnet *);
static int
vlan_modevent(module_t mod, int type, void *data)
{
switch (type) {
case MOD_LOAD:
ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
if (ifdetach_tag == NULL)
return (ENOMEM);
iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
if (iflladdr_tag == NULL)
return (ENOMEM);
VLAN_LOCKING_INIT();
vlan_input_p = vlan_input;
vlan_link_state_p = vlan_link_state;
2006-01-30 13:45:15 +00:00
vlan_trunk_cap_p = vlan_trunk_capabilities;
vlan_trunkdev_p = vlan_trunkdev;
vlan_cookie_p = vlan_cookie;
vlan_setcookie_p = vlan_setcookie;
vlan_tag_p = vlan_tag;
vlan_pcp_p = vlan_pcp;
vlan_devat_p = vlan_devat;
#ifndef VIMAGE
vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
vlan_clone_create, vlan_clone_destroy);
#endif
if (bootverbose)
printf("vlan: initialized, using "
#ifdef VLAN_ARRAY
"full-size arrays"
#else
"hash tables with chaining"
#endif
"\n");
break;
case MOD_UNLOAD:
#ifndef VIMAGE
if_clone_detach(vlan_cloner);
#endif
EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
vlan_input_p = NULL;
vlan_link_state_p = NULL;
2006-01-30 13:45:15 +00:00
vlan_trunk_cap_p = NULL;
vlan_trunkdev_p = NULL;
vlan_tag_p = NULL;
vlan_cookie_p = NULL;
vlan_setcookie_p = NULL;
vlan_devat_p = NULL;
VLAN_LOCKING_DESTROY();
if (bootverbose)
printf("vlan: unloaded\n");
break;
default:
return (EOPNOTSUPP);
}
return (0);
}
static moduledata_t vlan_mod = {
"if_vlan",
vlan_modevent,
0
};
DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
MODULE_VERSION(if_vlan, 3);
#ifdef VIMAGE
static void
vnet_vlan_init(const void *unused __unused)
{
vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
vlan_clone_create, vlan_clone_destroy);
V_vlan_cloner = vlan_cloner;
}
VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
vnet_vlan_init, NULL);
static void
vnet_vlan_uninit(const void *unused __unused)
{
if_clone_detach(V_vlan_cloner);
}
VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
vnet_vlan_uninit, NULL);
#endif
/*
* Check for <etherif>.<vlan>[.<vlan> ...] style interface names.
*/
static struct ifnet *
vlan_clone_match_ethervid(const char *name, int *vidp)
{
char ifname[IFNAMSIZ];
char *cp;
struct ifnet *ifp;
int vid;
strlcpy(ifname, name, IFNAMSIZ);
if ((cp = strrchr(ifname, '.')) == NULL)
return (NULL);
*cp = '\0';
if ((ifp = ifunit_ref(ifname)) == NULL)
return (NULL);
/* Parse VID. */
if (*++cp == '\0') {
if_rele(ifp);
return (NULL);
}
vid = 0;
for(; *cp >= '0' && *cp <= '9'; cp++)
vid = (vid * 10) + (*cp - '0');
if (*cp != '\0') {
if_rele(ifp);
return (NULL);
}
if (vidp != NULL)
*vidp = vid;
return (ifp);
}
static int
vlan_clone_match(struct if_clone *ifc, const char *name)
{
struct ifnet *ifp;
const char *cp;
ifp = vlan_clone_match_ethervid(name, NULL);
if (ifp != NULL) {
if_rele(ifp);
return (1);
}
if (strncmp(vlanname, name, strlen(vlanname)) != 0)
return (0);
for (cp = name + 4; *cp != '\0'; cp++) {
if (*cp < '0' || *cp > '9')
return (0);
}
return (1);
}
static int
vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
{
char *dp;
bool wildcard = false;
bool subinterface = false;
int unit;
int error;
int vid = 0;
uint16_t proto = ETHERTYPE_VLAN;
struct ifvlan *ifv;
struct ifnet *ifp;
struct ifnet *p = NULL;
struct ifaddr *ifa;
struct sockaddr_dl *sdl;
struct vlanreq vlr;
static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
/*
* There are three ways to specify the cloned device:
* o pass a parameter block with the clone request.
* o specify parameters in the text of the clone device name
* o specify no parameters and get an unattached device that
* must be configured separately.
* The first technique is preferred; the latter two are supported
* for backwards compatibility.
*
* XXXRW: Note historic use of the word "tag" here. New ioctls may be
* called for.
*/
if (params) {
error = copyin(params, &vlr, sizeof(vlr));
if (error)
return error;
vid = vlr.vlr_tag;
proto = vlr.vlr_proto;
#ifdef COMPAT_FREEBSD12
if (proto == 0)
proto = ETHERTYPE_VLAN;
#endif
p = ifunit_ref(vlr.vlr_parent);
if (p == NULL)
return (ENXIO);
}
if ((error = ifc_name2unit(name, &unit)) == 0) {
/*
* vlanX interface. Set wildcard to true if the unit number
* is not fixed (-1)
*/
wildcard = (unit < 0);
} else {
struct ifnet *p_tmp = vlan_clone_match_ethervid(name, &vid);
if (p_tmp != NULL) {
error = 0;
subinterface = true;
unit = IF_DUNIT_NONE;
wildcard = false;
if (p != NULL) {
if_rele(p_tmp);
if (p != p_tmp)
error = EINVAL;
} else
p = p_tmp;
} else
error = ENXIO;
}
if (error != 0) {
if (p != NULL)
if_rele(p);
return (error);
}
if (!subinterface) {
/* vlanX interface, mark X as busy or allocate new unit # */
error = ifc_alloc_unit(ifc, &unit);
if (error != 0) {
if (p != NULL)
if_rele(p);
return (error);
}
}
/* In the wildcard case, we need to update the name. */
if (wildcard) {
for (dp = name; *dp != '\0'; dp++);
if (snprintf(dp, len - (dp-name), "%d", unit) >
len - (dp-name) - 1) {
panic("%s: interface name too long", __func__);
}
}
ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
if (ifp == NULL) {
if (!subinterface)
ifc_free_unit(ifc, unit);
free(ifv, M_VLAN);
if (p != NULL)
if_rele(p);
return (ENOSPC);
}
CK_SLIST_INIT(&ifv->vlan_mc_listhead);
ifp->if_softc = ifv;
/*
2005-01-24 15:48:00 +00:00
* Set the name manually rather than using if_initname because
* we don't conform to the default naming convention for interfaces.
*/
strlcpy(ifp->if_xname, name, IFNAMSIZ);
ifp->if_dname = vlanname;
ifp->if_dunit = unit;
ifp->if_init = vlan_init;
ifp->if_transmit = vlan_transmit;
ifp->if_qflush = vlan_qflush;
ifp->if_ioctl = vlan_ioctl;
Add kernel-side support for in-kernel TLS. KTLS adds support for in-kernel framing and encryption of Transport Layer Security (1.0-1.2) data on TCP sockets. KTLS only supports offload of TLS for transmitted data. Key negotation must still be performed in userland. Once completed, transmit session keys for a connection are provided to the kernel via a new TCP_TXTLS_ENABLE socket option. All subsequent data transmitted on the socket is placed into TLS frames and encrypted using the supplied keys. Any data written to a KTLS-enabled socket via write(2), aio_write(2), or sendfile(2) is assumed to be application data and is encoded in TLS frames with an application data type. Individual records can be sent with a custom type (e.g. handshake messages) via sendmsg(2) with a new control message (TLS_SET_RECORD_TYPE) specifying the record type. At present, rekeying is not supported though the in-kernel framework should support rekeying. KTLS makes use of the recently added unmapped mbufs to store TLS frames in the socket buffer. Each TLS frame is described by a single ext_pgs mbuf. The ext_pgs structure contains the header of the TLS record (and trailer for encrypted records) as well as references to the associated TLS session. KTLS supports two primary methods of encrypting TLS frames: software TLS and ifnet TLS. Software TLS marks mbufs holding socket data as not ready via M_NOTREADY similar to sendfile(2) when TLS framing information is added to an unmapped mbuf in ktls_frame(). ktls_enqueue() is then called to schedule TLS frames for encryption. In the case of sendfile_iodone() calls ktls_enqueue() instead of pru_ready() leaving the mbufs marked M_NOTREADY until encryption is completed. For other writes (vn_sendfile when pages are available, write(2), etc.), the PRUS_NOTREADY is set when invoking pru_send() along with invoking ktls_enqueue(). A pool of worker threads (the "KTLS" kernel process) encrypts TLS frames queued via ktls_enqueue(). Each TLS frame is temporarily mapped using the direct map and passed to a software encryption backend to perform the actual encryption. (Note: The use of PHYS_TO_DMAP could be replaced with sf_bufs if someone wished to make this work on architectures without a direct map.) KTLS supports pluggable software encryption backends. Internally, Netflix uses proprietary pure-software backends. This commit includes a simple backend in a new ktls_ocf.ko module that uses the kernel's OpenCrypto framework to provide AES-GCM encryption of TLS frames. As a result, software TLS is now a bit of a misnomer as it can make use of hardware crypto accelerators. Once software encryption has finished, the TLS frame mbufs are marked ready via pru_ready(). At this point, the encrypted data appears as regular payload to the TCP stack stored in unmapped mbufs. ifnet TLS permits a NIC to offload the TLS encryption and TCP segmentation. In this mode, a new send tag type (IF_SND_TAG_TYPE_TLS) is allocated on the interface a socket is routed over and associated with a TLS session. TLS records for a TLS session using ifnet TLS are not marked M_NOTREADY but are passed down the stack unencrypted. The ip_output_send() and ip6_output_send() helper functions that apply send tags to outbound IP packets verify that the send tag of the TLS record matches the outbound interface. If so, the packet is tagged with the TLS send tag and sent to the interface. The NIC device driver must recognize packets with the TLS send tag and schedule them for TLS encryption and TCP segmentation. If the the outbound interface does not match the interface in the TLS send tag, the packet is dropped. In addition, a task is scheduled to refresh the TLS send tag for the TLS session. If a new TLS send tag cannot be allocated, the connection is dropped. If a new TLS send tag is allocated, however, subsequent packets will be tagged with the correct TLS send tag. (This latter case has been tested by configuring both ports of a Chelsio T6 in a lagg and failing over from one port to another. As the connections migrated to the new port, new TLS send tags were allocated for the new port and connections resumed without being dropped.) ifnet TLS can be enabled and disabled on supported network interfaces via new '[-]txtls[46]' options to ifconfig(8). ifnet TLS is supported across both vlan devices and lagg interfaces using failover, lacp with flowid enabled, or lacp with flowid enabled. Applications may request the current KTLS mode of a connection via a new TCP_TXTLS_MODE socket option. They can also use this socket option to toggle between software and ifnet TLS modes. In addition, a testing tool is available in tools/tools/switch_tls. This is modeled on tcpdrop and uses similar syntax. However, instead of dropping connections, -s is used to force KTLS connections to switch to software TLS and -i is used to switch to ifnet TLS. Various sysctls and counters are available under the kern.ipc.tls sysctl node. The kern.ipc.tls.enable node must be set to true to enable KTLS (it is off by default). The use of unmapped mbufs must also be enabled via kern.ipc.mb_use_ext_pgs to enable KTLS. KTLS is enabled via the KERN_TLS kernel option. This patch is the culmination of years of work by several folks including Scott Long and Randall Stewart for the original design and implementation; Drew Gallatin for several optimizations including the use of ext_pgs mbufs, the M_NOTREADY mechanism for TLS records awaiting software encryption, and pluggable software crypto backends; and John Baldwin for modifications to support hardware TLS offload. Reviewed by: gallatin, hselasky, rrs Obtained from: Netflix Sponsored by: Netflix, Chelsio Communications Differential Revision: https://reviews.freebsd.org/D21277
2019-08-27 00:01:56 +00:00
#if defined(KERN_TLS) || defined(RATELIMIT)
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
ifp->if_snd_tag_modify = vlan_snd_tag_modify;
ifp->if_snd_tag_query = vlan_snd_tag_query;
ifp->if_snd_tag_free = vlan_snd_tag_free;
ifp->if_next_snd_tag = vlan_next_snd_tag;
ifp->if_ratelimit_query = vlan_ratelimit_query;
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
#endif
ifp->if_flags = VLAN_IFFLAGS;
ether_ifattach(ifp, eaddr);
/* Now undo some of the damage... */
ifp->if_baudrate = 0;
ifp->if_type = IFT_L2VLAN;
ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
ifa = ifp->if_addr;
sdl = (struct sockaddr_dl *)ifa->ifa_addr;
sdl->sdl_type = IFT_L2VLAN;
if (p != NULL) {
error = vlan_config(ifv, p, vid, proto);
if_rele(p);
if (error != 0) {
/*
* Since we've partially failed, we need to back
* out all the way, otherwise userland could get
* confused. Thus, we destroy the interface.
*/
ether_ifdetach(ifp);
vlan_unconfig(ifp);
if_free(ifp);
if (!subinterface)
ifc_free_unit(ifc, unit);
free(ifv, M_VLAN);
return (error);
}
}
return (0);
}
static int
vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
{
struct ifvlan *ifv = ifp->if_softc;
int unit = ifp->if_dunit;
if (ifp->if_vlantrunk)
return (EBUSY);
ether_ifdetach(ifp); /* first, remove it from system-wide lists */
vlan_unconfig(ifp); /* now it can be unconfigured and freed */
/*
* We should have the only reference to the ifv now, so we can now
* drain any remaining lladdr task before freeing the ifnet and the
* ifvlan.
*/
taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
NET_EPOCH_WAIT();
if_free(ifp);
free(ifv, M_VLAN);
if (unit != IF_DUNIT_NONE)
ifc_free_unit(ifc, unit);
return (0);
}
/*
* The ifp->if_init entry point for vlan(4) is a no-op.
*/
static void
vlan_init(void *foo __unused)
{
}
/*
* The if_transmit method for vlan(4) interface.
*/
static int
vlan_transmit(struct ifnet *ifp, struct mbuf *m)
{
struct ifvlan *ifv;
struct ifnet *p;
int error, len, mcast;
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
NET_EPOCH_ASSERT();
ifv = ifp->if_softc;
if (TRUNK(ifv) == NULL) {
if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
m_freem(m);
return (ENETDOWN);
}
2006-01-30 13:45:15 +00:00
p = PARENT(ifv);
len = m->m_pkthdr.len;
mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
BPF_MTAP(ifp, m);
Add kernel-side support for in-kernel TLS. KTLS adds support for in-kernel framing and encryption of Transport Layer Security (1.0-1.2) data on TCP sockets. KTLS only supports offload of TLS for transmitted data. Key negotation must still be performed in userland. Once completed, transmit session keys for a connection are provided to the kernel via a new TCP_TXTLS_ENABLE socket option. All subsequent data transmitted on the socket is placed into TLS frames and encrypted using the supplied keys. Any data written to a KTLS-enabled socket via write(2), aio_write(2), or sendfile(2) is assumed to be application data and is encoded in TLS frames with an application data type. Individual records can be sent with a custom type (e.g. handshake messages) via sendmsg(2) with a new control message (TLS_SET_RECORD_TYPE) specifying the record type. At present, rekeying is not supported though the in-kernel framework should support rekeying. KTLS makes use of the recently added unmapped mbufs to store TLS frames in the socket buffer. Each TLS frame is described by a single ext_pgs mbuf. The ext_pgs structure contains the header of the TLS record (and trailer for encrypted records) as well as references to the associated TLS session. KTLS supports two primary methods of encrypting TLS frames: software TLS and ifnet TLS. Software TLS marks mbufs holding socket data as not ready via M_NOTREADY similar to sendfile(2) when TLS framing information is added to an unmapped mbuf in ktls_frame(). ktls_enqueue() is then called to schedule TLS frames for encryption. In the case of sendfile_iodone() calls ktls_enqueue() instead of pru_ready() leaving the mbufs marked M_NOTREADY until encryption is completed. For other writes (vn_sendfile when pages are available, write(2), etc.), the PRUS_NOTREADY is set when invoking pru_send() along with invoking ktls_enqueue(). A pool of worker threads (the "KTLS" kernel process) encrypts TLS frames queued via ktls_enqueue(). Each TLS frame is temporarily mapped using the direct map and passed to a software encryption backend to perform the actual encryption. (Note: The use of PHYS_TO_DMAP could be replaced with sf_bufs if someone wished to make this work on architectures without a direct map.) KTLS supports pluggable software encryption backends. Internally, Netflix uses proprietary pure-software backends. This commit includes a simple backend in a new ktls_ocf.ko module that uses the kernel's OpenCrypto framework to provide AES-GCM encryption of TLS frames. As a result, software TLS is now a bit of a misnomer as it can make use of hardware crypto accelerators. Once software encryption has finished, the TLS frame mbufs are marked ready via pru_ready(). At this point, the encrypted data appears as regular payload to the TCP stack stored in unmapped mbufs. ifnet TLS permits a NIC to offload the TLS encryption and TCP segmentation. In this mode, a new send tag type (IF_SND_TAG_TYPE_TLS) is allocated on the interface a socket is routed over and associated with a TLS session. TLS records for a TLS session using ifnet TLS are not marked M_NOTREADY but are passed down the stack unencrypted. The ip_output_send() and ip6_output_send() helper functions that apply send tags to outbound IP packets verify that the send tag of the TLS record matches the outbound interface. If so, the packet is tagged with the TLS send tag and sent to the interface. The NIC device driver must recognize packets with the TLS send tag and schedule them for TLS encryption and TCP segmentation. If the the outbound interface does not match the interface in the TLS send tag, the packet is dropped. In addition, a task is scheduled to refresh the TLS send tag for the TLS session. If a new TLS send tag cannot be allocated, the connection is dropped. If a new TLS send tag is allocated, however, subsequent packets will be tagged with the correct TLS send tag. (This latter case has been tested by configuring both ports of a Chelsio T6 in a lagg and failing over from one port to another. As the connections migrated to the new port, new TLS send tags were allocated for the new port and connections resumed without being dropped.) ifnet TLS can be enabled and disabled on supported network interfaces via new '[-]txtls[46]' options to ifconfig(8). ifnet TLS is supported across both vlan devices and lagg interfaces using failover, lacp with flowid enabled, or lacp with flowid enabled. Applications may request the current KTLS mode of a connection via a new TCP_TXTLS_MODE socket option. They can also use this socket option to toggle between software and ifnet TLS modes. In addition, a testing tool is available in tools/tools/switch_tls. This is modeled on tcpdrop and uses similar syntax. However, instead of dropping connections, -s is used to force KTLS connections to switch to software TLS and -i is used to switch to ifnet TLS. Various sysctls and counters are available under the kern.ipc.tls sysctl node. The kern.ipc.tls.enable node must be set to true to enable KTLS (it is off by default). The use of unmapped mbufs must also be enabled via kern.ipc.mb_use_ext_pgs to enable KTLS. KTLS is enabled via the KERN_TLS kernel option. This patch is the culmination of years of work by several folks including Scott Long and Randall Stewart for the original design and implementation; Drew Gallatin for several optimizations including the use of ext_pgs mbufs, the M_NOTREADY mechanism for TLS records awaiting software encryption, and pluggable software crypto backends; and John Baldwin for modifications to support hardware TLS offload. Reviewed by: gallatin, hselasky, rrs Obtained from: Netflix Sponsored by: Netflix, Chelsio Communications Differential Revision: https://reviews.freebsd.org/D21277
2019-08-27 00:01:56 +00:00
#if defined(KERN_TLS) || defined(RATELIMIT)
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
struct vlan_snd_tag *vst;
struct m_snd_tag *mst;
MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
mst = m->m_pkthdr.snd_tag;
vst = mst_to_vst(mst);
if (vst->tag->ifp != p) {
if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
m_freem(m);
return (EAGAIN);
}
m->m_pkthdr.snd_tag = m_snd_tag_ref(vst->tag);
m_snd_tag_rele(mst);
}
#endif
/*
* Do not run parent's if_transmit() if the parent is not up,
* or parent's driver will cause a system crash.
*/
if (!UP_AND_RUNNING(p)) {
if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
m_freem(m);
return (ENETDOWN);
}
if (!ether_8021q_frame(&m, ifp, p, &ifv->ifv_qtag)) {
if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
return (0);
}
/*
* Send it, precisely as ether_output() would have.
*/
error = (p->if_transmit)(p, m);
if (error == 0) {
if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
} else
if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
return (error);
}
static int
vlan_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
struct route *ro)
{
struct ifvlan *ifv;
struct ifnet *p;
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
NET_EPOCH_ASSERT();
/*
* Find the first non-VLAN parent interface.
*/
ifv = ifp->if_softc;
do {
if (TRUNK(ifv) == NULL) {
m_freem(m);
return (ENETDOWN);
}
p = PARENT(ifv);
ifv = p->if_softc;
} while (p->if_type == IFT_L2VLAN);
return p->if_output(ifp, m, dst, ro);
}
/*
* The ifp->if_qflush entry point for vlan(4) is a no-op.
*/
static void
vlan_qflush(struct ifnet *ifp __unused)
{
}
static void
vlan_input(struct ifnet *ifp, struct mbuf *m)
{
struct ifvlantrunk *trunk;
struct ifvlan *ifv;
struct m_tag *mtag;
uint16_t vid, tag;
2006-01-30 13:45:15 +00:00
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
NET_EPOCH_ASSERT();
trunk = ifp->if_vlantrunk;
if (trunk == NULL) {
m_freem(m);
return;
}
if (m->m_flags & M_VLANTAG) {
/*
* Packet is tagged, but m contains a normal
* Ethernet frame; the tag is stored out-of-band.
*/
tag = m->m_pkthdr.ether_vtag;
m->m_flags &= ~M_VLANTAG;
} else {
2006-01-30 13:45:15 +00:00
struct ether_vlan_header *evl;
/*
* Packet is tagged in-band as specified by 802.1q.
*/
switch (ifp->if_type) {
case IFT_ETHER:
if (m->m_len < sizeof(*evl) &&
(m = m_pullup(m, sizeof(*evl))) == NULL) {
if_printf(ifp, "cannot pullup VLAN header\n");
return;
}
evl = mtod(m, struct ether_vlan_header *);
tag = ntohs(evl->evl_tag);
/*
* Remove the 802.1q header by copying the Ethernet
* addresses over it and adjusting the beginning of
* the data in the mbuf. The encapsulated Ethernet
* type field is already in place.
*/
bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
ETHER_HDR_LEN - ETHER_TYPE_LEN);
m_adj(m, ETHER_VLAN_ENCAP_LEN);
break;
default:
#ifdef INVARIANTS
panic("%s: %s has unsupported if_type %u",
__func__, ifp->if_xname, ifp->if_type);
#endif
if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
m_freem(m);
return;
}
}
vid = EVL_VLANOFTAG(tag);
ifv = vlan_gethash(trunk, vid);
if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
m_freem(m);
2006-01-30 13:45:15 +00:00
return;
}
if (vlan_mtag_pcp) {
/*
* While uncommon, it is possible that we will find a 802.1q
* packet encapsulated inside another packet that also had an
* 802.1q header. For example, ethernet tunneled over IPSEC
* arriving over ethernet. In that case, we replace the
* existing 802.1q PCP m_tag value.
*/
mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
if (mtag == NULL) {
mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
sizeof(uint8_t), M_NOWAIT);
if (mtag == NULL) {
if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
m_freem(m);
return;
}
m_tag_prepend(m, mtag);
}
*(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
}
m->m_pkthdr.rcvif = ifv->ifv_ifp;
if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
/* Pass it back through the parent's input routine. */
(*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
}
static void
vlan_lladdr_fn(void *arg, int pending __unused)
{
struct ifvlan *ifv;
struct ifnet *ifp;
ifv = (struct ifvlan *)arg;
ifp = ifv->ifv_ifp;
CURVNET_SET(ifp->if_vnet);
/* The ifv_ifp already has the lladdr copied in. */
if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
CURVNET_RESTORE();
}
static int
vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid,
uint16_t proto)
{
struct epoch_tracker et;
2006-01-30 13:45:15 +00:00
struct ifvlantrunk *trunk;
struct ifnet *ifp;
2006-01-30 13:45:15 +00:00
int error = 0;
/*
* We can handle non-ethernet hardware types as long as
* they handle the tagging and headers themselves.
*/
if (p->if_type != IFT_ETHER &&
p->if_type != IFT_L2VLAN &&
(p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
return (EPROTONOSUPPORT);
if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
return (EPROTONOSUPPORT);
/*
* Don't let the caller set up a VLAN VID with
* anything except VLID bits.
* VID numbers 0x0 and 0xFFF are reserved.
*/
if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
return (EINVAL);
2006-01-30 13:45:15 +00:00
if (ifv->ifv_trunk)
return (EBUSY);
VLAN_XLOCK();
2006-01-30 13:45:15 +00:00
if (p->if_vlantrunk == NULL) {
trunk = malloc(sizeof(struct ifvlantrunk),
M_VLAN, M_WAITOK | M_ZERO);
vlan_inithash(trunk);
2006-01-30 13:45:15 +00:00
TRUNK_LOCK_INIT(trunk);
TRUNK_WLOCK(trunk);
2006-01-30 13:45:15 +00:00
p->if_vlantrunk = trunk;
trunk->parent = p;
if_ref(trunk->parent);
TRUNK_WUNLOCK(trunk);
2006-01-30 13:45:15 +00:00
} else {
trunk = p->if_vlantrunk;
}
ifv->ifv_vid = vid; /* must set this before vlan_inshash() */
ifv->ifv_pcp = 0; /* Default: best effort delivery. */
2006-01-30 13:45:15 +00:00
error = vlan_inshash(trunk, ifv);
if (error)
goto done;
ifv->ifv_proto = proto;
ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
ifv->ifv_mintu = ETHERMIN;
ifv->ifv_pflags = 0;
ifv->ifv_capenable = -1;
/*
* If the parent supports the VLAN_MTU capability,
* i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
* use it.
*/
if (p->if_capenable & IFCAP_VLAN_MTU) {
/*
* No need to fudge the MTU since the parent can
* handle extended frames.
*/
ifv->ifv_mtufudge = 0;
} else {
/*
* Fudge the MTU by the encapsulation size. This
* makes us incompatible with strictly compliant
* 802.1Q implementations, but allows us to use
* the feature with other NetBSD implementations,
* which might still be useful.
*/
ifv->ifv_mtufudge = ifv->ifv_encaplen;
}
2006-01-30 13:45:15 +00:00
ifv->ifv_trunk = trunk;
ifp = ifv->ifv_ifp;
/*
* Initialize fields from our parent. This duplicates some
* work with ether_ifattach() but allows for non-ethernet
* interfaces to also work.
*/
ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
2006-01-30 13:45:15 +00:00
ifp->if_baudrate = p->if_baudrate;
ifp->if_input = p->if_input;
ifp->if_resolvemulti = p->if_resolvemulti;
ifp->if_addrlen = p->if_addrlen;
ifp->if_broadcastaddr = p->if_broadcastaddr;
ifp->if_pcp = ifv->ifv_pcp;
/*
* We wrap the parent's if_output using vlan_output to ensure that it
* can't become stale.
*/
ifp->if_output = vlan_output;
/*
* Copy only a selected subset of flags from the parent.
* Other flags are none of our business.
*/
#define VLAN_COPY_FLAGS (IFF_SIMPLEX)
ifp->if_flags &= ~VLAN_COPY_FLAGS;
ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
#undef VLAN_COPY_FLAGS
ifp->if_link_state = p->if_link_state;
NET_EPOCH_ENTER(et);
2006-01-30 13:45:15 +00:00
vlan_capabilities(ifv);
NET_EPOCH_EXIT(et);
/*
* Set up our interface address to reflect the underlying
* physical interface's.
*/
TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
p->if_addrlen;
/*
* Do not schedule link address update if it was the same
* as previous parent's. This helps avoid updating for each
* associated llentry.
*/
if (memcmp(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen) != 0) {
bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
}
/* We are ready for operation now. */
ifp->if_drv_flags |= IFF_DRV_RUNNING;
/* Update flags on the parent, if necessary. */
vlan_setflags(ifp, 1);
/*
* Configure multicast addresses that may already be
* joined on the vlan device.
*/
(void)vlan_setmulti(ifp);
done:
if (error == 0)
EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
VLAN_XUNLOCK();
2006-01-30 13:45:15 +00:00
return (error);
}
static void
vlan_unconfig(struct ifnet *ifp)
{
VLAN_XLOCK();
vlan_unconfig_locked(ifp, 0);
VLAN_XUNLOCK();
}
static void
vlan_unconfig_locked(struct ifnet *ifp, int departing)
{
2006-01-30 13:45:15 +00:00
struct ifvlantrunk *trunk;
struct vlan_mc_entry *mc;
struct ifvlan *ifv;
struct ifnet *parent;
int error;
VLAN_XLOCK_ASSERT();
ifv = ifp->if_softc;
2006-01-30 13:45:15 +00:00
trunk = ifv->ifv_trunk;
parent = NULL;
if (trunk != NULL) {
parent = trunk->parent;
/*
* Since the interface is being unconfigured, we need to
* empty the list of multicast groups that we may have joined
* while we were alive from the parent's list.
*/
while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
/*
* If the parent interface is being detached,
* all its multicast addresses have already
* been removed. Warn about errors if
* if_delmulti() does fail, but don't abort as
* all callers expect vlan destruction to
* succeed.
*/
if (!departing) {
error = if_delmulti(parent,
(struct sockaddr *)&mc->mc_addr);
if (error)
if_printf(ifp,
"Failed to delete multicast address from parent: %d\n",
error);
}
CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
}
vlan_setflags(ifp, 0); /* clear special flags on parent */
2006-01-30 13:45:15 +00:00
vlan_remhash(trunk, ifv);
ifv->ifv_trunk = NULL;
/*
* Check if we were the last.
*/
if (trunk->refcnt == 0) {
parent->if_vlantrunk = NULL;
NET_EPOCH_WAIT();
2006-01-30 13:45:15 +00:00
trunk_destroy(trunk);
}
}
/* Disconnect from parent. */
if (ifv->ifv_pflags)
if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
ifp->if_mtu = ETHERMTU;
ifp->if_link_state = LINK_STATE_UNKNOWN;
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
/*
* Only dispatch an event if vlan was
* attached, otherwise there is nothing
* to cleanup anyway.
*/
if (parent != NULL)
EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
}
/* Handle a reference counted flag that should be set on the parent as well */
static int
vlan_setflag(struct ifnet *ifp, int flag, int status,
int (*func)(struct ifnet *, int))
{
struct ifvlan *ifv;
int error;
VLAN_SXLOCK_ASSERT();
ifv = ifp->if_softc;
status = status ? (ifp->if_flags & flag) : 0;
/* Now "status" contains the flag value or 0 */
/*
* See if recorded parent's status is different from what
* we want it to be. If it is, flip it. We record parent's
* status in ifv_pflags so that we won't clear parent's flag
* we haven't set. In fact, we don't clear or set parent's
* flags directly, but get or release references to them.
* That's why we can be sure that recorded flags still are
* in accord with actual parent's flags.
*/
if (status != (ifv->ifv_pflags & flag)) {
2006-01-30 13:45:15 +00:00
error = (*func)(PARENT(ifv), status);
if (error)
return (error);
ifv->ifv_pflags &= ~flag;
ifv->ifv_pflags |= status;
}
return (0);
}
/*
* Handle IFF_* flags that require certain changes on the parent:
* if "status" is true, update parent's flags respective to our if_flags;
* if "status" is false, forcedly clear the flags set on parent.
*/
static int
vlan_setflags(struct ifnet *ifp, int status)
{
int error, i;
for (i = 0; vlan_pflags[i].flag; i++) {
error = vlan_setflag(ifp, vlan_pflags[i].flag,
status, vlan_pflags[i].func);
if (error)
return (error);
}
return (0);
}
/* Inform all vlans that their parent has changed link state */
static void
vlan_link_state(struct ifnet *ifp)
{
struct epoch_tracker et;
struct ifvlantrunk *trunk;
struct ifvlan *ifv;
2006-01-30 13:45:15 +00:00
NET_EPOCH_ENTER(et);
trunk = ifp->if_vlantrunk;
if (trunk == NULL) {
NET_EPOCH_EXIT(et);
return;
}
TRUNK_WLOCK(trunk);
VLAN_FOREACH(ifv, trunk) {
ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
if_link_state_change(ifv->ifv_ifp,
trunk->parent->if_link_state);
}
TRUNK_WUNLOCK(trunk);
NET_EPOCH_EXIT(et);
2006-01-30 13:45:15 +00:00
}
static void
vlan_capabilities(struct ifvlan *ifv)
{
struct ifnet *p;
struct ifnet *ifp;
struct ifnet_hw_tsomax hw_tsomax;
int cap = 0, ena = 0, mena;
u_long hwa = 0;
2006-01-30 13:45:15 +00:00
NET_EPOCH_ASSERT();
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
VLAN_SXLOCK_ASSERT();
p = PARENT(ifv);
ifp = ifv->ifv_ifp;
2006-01-30 13:45:15 +00:00
/* Mask parent interface enabled capabilities disabled by user. */
mena = p->if_capenable & ifv->ifv_capenable;
2006-01-30 13:45:15 +00:00
/*
* If the parent interface can do checksum offloading
* on VLANs, then propagate its hardware-assisted
* checksumming flags. Also assert that checksum
* offloading requires hardware VLAN tagging.
*/
if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
2006-01-30 13:45:15 +00:00
if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
p->if_capenable & IFCAP_VLAN_HWTAGGING) {
ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
if (ena & IFCAP_TXCSUM)
hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
CSUM_UDP | CSUM_SCTP);
if (ena & IFCAP_TXCSUM_IPV6)
hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
}
/*
* If the parent interface can do TSO on VLANs then
* propagate the hardware-assisted flag. TSO on VLANs
* does not necessarily require hardware VLAN tagging.
*/
memset(&hw_tsomax, 0, sizeof(hw_tsomax));
if_hw_tsomax_common(p, &hw_tsomax);
if_hw_tsomax_update(ifp, &hw_tsomax);
if (p->if_capabilities & IFCAP_VLAN_HWTSO)
cap |= p->if_capabilities & IFCAP_TSO;
if (p->if_capenable & IFCAP_VLAN_HWTSO) {
ena |= mena & IFCAP_TSO;
if (ena & IFCAP_TSO)
hwa |= p->if_hwassist & CSUM_TSO;
}
/*
* If the parent interface can do LRO and checksum offloading on
* VLANs, then guess it may do LRO on VLANs. False positive here
* cost nothing, while false negative may lead to some confusions.
*/
if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
cap |= p->if_capabilities & IFCAP_LRO;
if (p->if_capenable & IFCAP_VLAN_HWCSUM)
ena |= p->if_capenable & IFCAP_LRO;
/*
* If the parent interface can offload TCP connections over VLANs then
* propagate its TOE capability to the VLAN interface.
*
* All TOE drivers in the tree today can deal with VLANs. If this
* changes then IFCAP_VLAN_TOE should be promoted to a full capability
* with its own bit.
*/
#define IFCAP_VLAN_TOE IFCAP_TOE
if (p->if_capabilities & IFCAP_VLAN_TOE)
cap |= p->if_capabilities & IFCAP_TOE;
if (p->if_capenable & IFCAP_VLAN_TOE) {
TOEDEV(ifp) = TOEDEV(p);
ena |= mena & IFCAP_TOE;
}
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
/*
* If the parent interface supports dynamic link state, so does the
* VLAN interface.
*/
cap |= (p->if_capabilities & IFCAP_LINKSTATE);
ena |= (mena & IFCAP_LINKSTATE);
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
#ifdef RATELIMIT
/*
* If the parent interface supports ratelimiting, so does the
* VLAN interface.
*/
cap |= (p->if_capabilities & IFCAP_TXRTLMT);
ena |= (mena & IFCAP_TXRTLMT);
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
#endif
/*
* If the parent interface supports unmapped mbufs, so does
* the VLAN interface. Note that this should be fine even for
* interfaces that don't support hardware tagging as headers
* are prepended in normal mbufs to unmapped mbufs holding
* payload data.
*/
cap |= (p->if_capabilities & IFCAP_MEXTPG);
ena |= (mena & IFCAP_MEXTPG);
Add kernel-side support for in-kernel TLS. KTLS adds support for in-kernel framing and encryption of Transport Layer Security (1.0-1.2) data on TCP sockets. KTLS only supports offload of TLS for transmitted data. Key negotation must still be performed in userland. Once completed, transmit session keys for a connection are provided to the kernel via a new TCP_TXTLS_ENABLE socket option. All subsequent data transmitted on the socket is placed into TLS frames and encrypted using the supplied keys. Any data written to a KTLS-enabled socket via write(2), aio_write(2), or sendfile(2) is assumed to be application data and is encoded in TLS frames with an application data type. Individual records can be sent with a custom type (e.g. handshake messages) via sendmsg(2) with a new control message (TLS_SET_RECORD_TYPE) specifying the record type. At present, rekeying is not supported though the in-kernel framework should support rekeying. KTLS makes use of the recently added unmapped mbufs to store TLS frames in the socket buffer. Each TLS frame is described by a single ext_pgs mbuf. The ext_pgs structure contains the header of the TLS record (and trailer for encrypted records) as well as references to the associated TLS session. KTLS supports two primary methods of encrypting TLS frames: software TLS and ifnet TLS. Software TLS marks mbufs holding socket data as not ready via M_NOTREADY similar to sendfile(2) when TLS framing information is added to an unmapped mbuf in ktls_frame(). ktls_enqueue() is then called to schedule TLS frames for encryption. In the case of sendfile_iodone() calls ktls_enqueue() instead of pru_ready() leaving the mbufs marked M_NOTREADY until encryption is completed. For other writes (vn_sendfile when pages are available, write(2), etc.), the PRUS_NOTREADY is set when invoking pru_send() along with invoking ktls_enqueue(). A pool of worker threads (the "KTLS" kernel process) encrypts TLS frames queued via ktls_enqueue(). Each TLS frame is temporarily mapped using the direct map and passed to a software encryption backend to perform the actual encryption. (Note: The use of PHYS_TO_DMAP could be replaced with sf_bufs if someone wished to make this work on architectures without a direct map.) KTLS supports pluggable software encryption backends. Internally, Netflix uses proprietary pure-software backends. This commit includes a simple backend in a new ktls_ocf.ko module that uses the kernel's OpenCrypto framework to provide AES-GCM encryption of TLS frames. As a result, software TLS is now a bit of a misnomer as it can make use of hardware crypto accelerators. Once software encryption has finished, the TLS frame mbufs are marked ready via pru_ready(). At this point, the encrypted data appears as regular payload to the TCP stack stored in unmapped mbufs. ifnet TLS permits a NIC to offload the TLS encryption and TCP segmentation. In this mode, a new send tag type (IF_SND_TAG_TYPE_TLS) is allocated on the interface a socket is routed over and associated with a TLS session. TLS records for a TLS session using ifnet TLS are not marked M_NOTREADY but are passed down the stack unencrypted. The ip_output_send() and ip6_output_send() helper functions that apply send tags to outbound IP packets verify that the send tag of the TLS record matches the outbound interface. If so, the packet is tagged with the TLS send tag and sent to the interface. The NIC device driver must recognize packets with the TLS send tag and schedule them for TLS encryption and TCP segmentation. If the the outbound interface does not match the interface in the TLS send tag, the packet is dropped. In addition, a task is scheduled to refresh the TLS send tag for the TLS session. If a new TLS send tag cannot be allocated, the connection is dropped. If a new TLS send tag is allocated, however, subsequent packets will be tagged with the correct TLS send tag. (This latter case has been tested by configuring both ports of a Chelsio T6 in a lagg and failing over from one port to another. As the connections migrated to the new port, new TLS send tags were allocated for the new port and connections resumed without being dropped.) ifnet TLS can be enabled and disabled on supported network interfaces via new '[-]txtls[46]' options to ifconfig(8). ifnet TLS is supported across both vlan devices and lagg interfaces using failover, lacp with flowid enabled, or lacp with flowid enabled. Applications may request the current KTLS mode of a connection via a new TCP_TXTLS_MODE socket option. They can also use this socket option to toggle between software and ifnet TLS modes. In addition, a testing tool is available in tools/tools/switch_tls. This is modeled on tcpdrop and uses similar syntax. However, instead of dropping connections, -s is used to force KTLS connections to switch to software TLS and -i is used to switch to ifnet TLS. Various sysctls and counters are available under the kern.ipc.tls sysctl node. The kern.ipc.tls.enable node must be set to true to enable KTLS (it is off by default). The use of unmapped mbufs must also be enabled via kern.ipc.mb_use_ext_pgs to enable KTLS. KTLS is enabled via the KERN_TLS kernel option. This patch is the culmination of years of work by several folks including Scott Long and Randall Stewart for the original design and implementation; Drew Gallatin for several optimizations including the use of ext_pgs mbufs, the M_NOTREADY mechanism for TLS records awaiting software encryption, and pluggable software crypto backends; and John Baldwin for modifications to support hardware TLS offload. Reviewed by: gallatin, hselasky, rrs Obtained from: Netflix Sponsored by: Netflix, Chelsio Communications Differential Revision: https://reviews.freebsd.org/D21277
2019-08-27 00:01:56 +00:00
/*
* If the parent interface can offload encryption and segmentation
* of TLS records over TCP, propagate it's capability to the VLAN
* interface.
*
* All TLS drivers in the tree today can deal with VLANs. If
* this ever changes, then a new IFCAP_VLAN_TXTLS can be
* defined.
*/
Support hardware rate limiting (pacing) with TLS offload. - Add a new send tag type for a send tag that supports both rate limiting (packet pacing) and TLS offload (mostly similar to D22669 but adds a separate structure when allocating the new tag type). - When allocating a send tag for TLS offload, check to see if the connection already has a pacing rate. If so, allocate a tag that supports both rate limiting and TLS offload rather than a plain TLS offload tag. - When setting an initial rate on an existing ifnet KTLS connection, set the rate in the TCP control block inp and then reset the TLS send tag (via ktls_output_eagain) to reallocate a TLS + ratelimit send tag. This allocates the TLS send tag asynchronously from a task queue, so the TLS rate limit tag alloc is always sleepable. - When modifying a rate on a connection using KTLS, look for a TLS send tag. If the send tag is only a plain TLS send tag, assume we failed to allocate a TLS ratelimit tag (either during the TCP_TXTLS_ENABLE socket option, or during the send tag reset triggered by ktls_output_eagain) and ignore the new rate. If the send tag is a ratelimit TLS send tag, change the rate on the TLS tag and leave the inp tag alone. - Lock the inp lock when setting sb_tls_info for a socket send buffer so that the routines in tcp_ratelimit can safely dereference the pointer without needing to grab the socket buffer lock. - Add an IFCAP_TXTLS_RTLMT capability flag and associated administrative controls in ifconfig(8). TLS rate limit tags are only allocated if this capability is enabled. Note that TLS offload (whether unlimited or rate limited) always requires IFCAP_TXTLS[46]. Reviewed by: gallatin, hselasky Relnotes: yes Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D26691
2020-10-29 00:23:16 +00:00
if (p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
cap |= p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
if (p->if_capenable & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
ena |= mena & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
Add kernel-side support for in-kernel TLS. KTLS adds support for in-kernel framing and encryption of Transport Layer Security (1.0-1.2) data on TCP sockets. KTLS only supports offload of TLS for transmitted data. Key negotation must still be performed in userland. Once completed, transmit session keys for a connection are provided to the kernel via a new TCP_TXTLS_ENABLE socket option. All subsequent data transmitted on the socket is placed into TLS frames and encrypted using the supplied keys. Any data written to a KTLS-enabled socket via write(2), aio_write(2), or sendfile(2) is assumed to be application data and is encoded in TLS frames with an application data type. Individual records can be sent with a custom type (e.g. handshake messages) via sendmsg(2) with a new control message (TLS_SET_RECORD_TYPE) specifying the record type. At present, rekeying is not supported though the in-kernel framework should support rekeying. KTLS makes use of the recently added unmapped mbufs to store TLS frames in the socket buffer. Each TLS frame is described by a single ext_pgs mbuf. The ext_pgs structure contains the header of the TLS record (and trailer for encrypted records) as well as references to the associated TLS session. KTLS supports two primary methods of encrypting TLS frames: software TLS and ifnet TLS. Software TLS marks mbufs holding socket data as not ready via M_NOTREADY similar to sendfile(2) when TLS framing information is added to an unmapped mbuf in ktls_frame(). ktls_enqueue() is then called to schedule TLS frames for encryption. In the case of sendfile_iodone() calls ktls_enqueue() instead of pru_ready() leaving the mbufs marked M_NOTREADY until encryption is completed. For other writes (vn_sendfile when pages are available, write(2), etc.), the PRUS_NOTREADY is set when invoking pru_send() along with invoking ktls_enqueue(). A pool of worker threads (the "KTLS" kernel process) encrypts TLS frames queued via ktls_enqueue(). Each TLS frame is temporarily mapped using the direct map and passed to a software encryption backend to perform the actual encryption. (Note: The use of PHYS_TO_DMAP could be replaced with sf_bufs if someone wished to make this work on architectures without a direct map.) KTLS supports pluggable software encryption backends. Internally, Netflix uses proprietary pure-software backends. This commit includes a simple backend in a new ktls_ocf.ko module that uses the kernel's OpenCrypto framework to provide AES-GCM encryption of TLS frames. As a result, software TLS is now a bit of a misnomer as it can make use of hardware crypto accelerators. Once software encryption has finished, the TLS frame mbufs are marked ready via pru_ready(). At this point, the encrypted data appears as regular payload to the TCP stack stored in unmapped mbufs. ifnet TLS permits a NIC to offload the TLS encryption and TCP segmentation. In this mode, a new send tag type (IF_SND_TAG_TYPE_TLS) is allocated on the interface a socket is routed over and associated with a TLS session. TLS records for a TLS session using ifnet TLS are not marked M_NOTREADY but are passed down the stack unencrypted. The ip_output_send() and ip6_output_send() helper functions that apply send tags to outbound IP packets verify that the send tag of the TLS record matches the outbound interface. If so, the packet is tagged with the TLS send tag and sent to the interface. The NIC device driver must recognize packets with the TLS send tag and schedule them for TLS encryption and TCP segmentation. If the the outbound interface does not match the interface in the TLS send tag, the packet is dropped. In addition, a task is scheduled to refresh the TLS send tag for the TLS session. If a new TLS send tag cannot be allocated, the connection is dropped. If a new TLS send tag is allocated, however, subsequent packets will be tagged with the correct TLS send tag. (This latter case has been tested by configuring both ports of a Chelsio T6 in a lagg and failing over from one port to another. As the connections migrated to the new port, new TLS send tags were allocated for the new port and connections resumed without being dropped.) ifnet TLS can be enabled and disabled on supported network interfaces via new '[-]txtls[46]' options to ifconfig(8). ifnet TLS is supported across both vlan devices and lagg interfaces using failover, lacp with flowid enabled, or lacp with flowid enabled. Applications may request the current KTLS mode of a connection via a new TCP_TXTLS_MODE socket option. They can also use this socket option to toggle between software and ifnet TLS modes. In addition, a testing tool is available in tools/tools/switch_tls. This is modeled on tcpdrop and uses similar syntax. However, instead of dropping connections, -s is used to force KTLS connections to switch to software TLS and -i is used to switch to ifnet TLS. Various sysctls and counters are available under the kern.ipc.tls sysctl node. The kern.ipc.tls.enable node must be set to true to enable KTLS (it is off by default). The use of unmapped mbufs must also be enabled via kern.ipc.mb_use_ext_pgs to enable KTLS. KTLS is enabled via the KERN_TLS kernel option. This patch is the culmination of years of work by several folks including Scott Long and Randall Stewart for the original design and implementation; Drew Gallatin for several optimizations including the use of ext_pgs mbufs, the M_NOTREADY mechanism for TLS records awaiting software encryption, and pluggable software crypto backends; and John Baldwin for modifications to support hardware TLS offload. Reviewed by: gallatin, hselasky, rrs Obtained from: Netflix Sponsored by: Netflix, Chelsio Communications Differential Revision: https://reviews.freebsd.org/D21277
2019-08-27 00:01:56 +00:00
ifp->if_capabilities = cap;
ifp->if_capenable = ena;
ifp->if_hwassist = hwa;
2006-01-30 13:45:15 +00:00
}
static void
vlan_trunk_capabilities(struct ifnet *ifp)
{
struct epoch_tracker et;
struct ifvlantrunk *trunk;
2006-01-30 13:45:15 +00:00
struct ifvlan *ifv;
VLAN_SLOCK();
trunk = ifp->if_vlantrunk;
if (trunk == NULL) {
VLAN_SUNLOCK();
return;
}
NET_EPOCH_ENTER(et);
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
VLAN_FOREACH(ifv, trunk)
vlan_capabilities(ifv);
NET_EPOCH_EXIT(et);
VLAN_SUNLOCK();
}
static int
vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
{
struct ifnet *p;
struct ifreq *ifr;
struct ifaddr *ifa;
struct ifvlan *ifv;
struct ifvlantrunk *trunk;
struct vlanreq vlr;
int error = 0, oldmtu;
ifr = (struct ifreq *)data;
ifa = (struct ifaddr *) data;
ifv = ifp->if_softc;
switch (cmd) {
case SIOCSIFADDR:
ifp->if_flags |= IFF_UP;
#ifdef INET
if (ifa->ifa_addr->sa_family == AF_INET)
arp_ifinit(ifp, ifa);
#endif
break;
case SIOCGIFADDR:
bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
ifp->if_addrlen);
break;
case SIOCGIFMEDIA:
VLAN_SLOCK();
2006-01-30 13:45:15 +00:00
if (TRUNK(ifv) != NULL) {
p = PARENT(ifv);
if_ref(p);
error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
if_rele(p);
/* Limit the result to the parent's current config. */
if (error == 0) {
struct ifmediareq *ifmr;
ifmr = (struct ifmediareq *)data;
if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
ifmr->ifm_count = 1;
error = copyout(&ifmr->ifm_current,
ifmr->ifm_ulist,
sizeof(int));
}
}
} else {
error = EINVAL;
}
VLAN_SUNLOCK();
break;
case SIOCSIFMEDIA:
error = EINVAL;
break;
case SIOCSIFMTU:
/*
* Set the interface MTU.
*/
VLAN_SLOCK();
trunk = TRUNK(ifv);
if (trunk != NULL) {
TRUNK_WLOCK(trunk);
if (ifr->ifr_mtu >
2006-01-30 13:45:15 +00:00
(PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
ifr->ifr_mtu <
(ifv->ifv_mintu - ifv->ifv_mtufudge))
error = EINVAL;
else
ifp->if_mtu = ifr->ifr_mtu;
TRUNK_WUNLOCK(trunk);
} else
error = EINVAL;
VLAN_SUNLOCK();
break;
case SIOCSETVLAN:
#ifdef VIMAGE
/*
* XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
* interface to be delegated to a jail without allowing the
* jail to change what underlying interface/VID it is
* associated with. We are not entirely convinced that this
* is the right way to accomplish that policy goal.
*/
if (ifp->if_vnet != ifp->if_home_vnet) {
error = EPERM;
break;
}
#endif
error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
if (error)
break;
if (vlr.vlr_parent[0] == '\0') {
vlan_unconfig(ifp);
break;
}
p = ifunit_ref(vlr.vlr_parent);
2009-09-09 03:36:43 +00:00
if (p == NULL) {
error = ENOENT;
break;
}
#ifdef COMPAT_FREEBSD12
if (vlr.vlr_proto == 0)
vlr.vlr_proto = ETHERTYPE_VLAN;
#endif
oldmtu = ifp->if_mtu;
error = vlan_config(ifv, p, vlr.vlr_tag, vlr.vlr_proto);
if_rele(p);
/*
* VLAN MTU may change during addition of the vlandev.
* If it did, do network layer specific procedure.
*/
if (ifp->if_mtu != oldmtu) {
#ifdef INET6
nd6_setmtu(ifp);
#endif
rt_updatemtu(ifp);
}
break;
case SIOCGETVLAN:
#ifdef VIMAGE
if (ifp->if_vnet != ifp->if_home_vnet) {
error = EPERM;
break;
}
#endif
bzero(&vlr, sizeof(vlr));
VLAN_SLOCK();
2006-01-30 13:45:15 +00:00
if (TRUNK(ifv) != NULL) {
strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
sizeof(vlr.vlr_parent));
vlr.vlr_tag = ifv->ifv_vid;
vlr.vlr_proto = ifv->ifv_proto;
}
VLAN_SUNLOCK();
error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
break;
case SIOCSIFFLAGS:
/*
* We should propagate selected flags to the parent,
* e.g., promiscuous mode.
*/
VLAN_XLOCK();
2006-01-30 13:45:15 +00:00
if (TRUNK(ifv) != NULL)
error = vlan_setflags(ifp, 1);
VLAN_XUNLOCK();
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
2006-01-30 13:45:15 +00:00
/*
* If we don't have a parent, just remember the membership for
* when we do.
*
* XXX We need the rmlock here to avoid sleeping while
* holding in6_multi_mtx.
2006-01-30 13:45:15 +00:00
*/
VLAN_XLOCK();
trunk = TRUNK(ifv);
if (trunk != NULL)
2006-01-30 13:45:15 +00:00
error = vlan_setmulti(ifp);
VLAN_XUNLOCK();
2006-01-30 13:45:15 +00:00
break;
case SIOCGVLANPCP:
#ifdef VIMAGE
if (ifp->if_vnet != ifp->if_home_vnet) {
error = EPERM;
break;
}
#endif
ifr->ifr_vlan_pcp = ifv->ifv_pcp;
break;
case SIOCSVLANPCP:
#ifdef VIMAGE
if (ifp->if_vnet != ifp->if_home_vnet) {
error = EPERM;
break;
}
#endif
error = priv_check(curthread, PRIV_NET_SETVLANPCP);
if (error)
break;
if (ifr->ifr_vlan_pcp > 7) {
error = EINVAL;
break;
}
ifv->ifv_pcp = ifr->ifr_vlan_pcp;
ifp->if_pcp = ifv->ifv_pcp;
/* broadcast event about PCP change */
EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
break;
case SIOCSIFCAP:
VLAN_SLOCK();
ifv->ifv_capenable = ifr->ifr_reqcap;
trunk = TRUNK(ifv);
if (trunk != NULL) {
struct epoch_tracker et;
NET_EPOCH_ENTER(et);
vlan_capabilities(ifv);
NET_EPOCH_EXIT(et);
}
VLAN_SUNLOCK();
break;
default:
error = EINVAL;
break;
}
return (error);
}
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
Add kernel-side support for in-kernel TLS. KTLS adds support for in-kernel framing and encryption of Transport Layer Security (1.0-1.2) data on TCP sockets. KTLS only supports offload of TLS for transmitted data. Key negotation must still be performed in userland. Once completed, transmit session keys for a connection are provided to the kernel via a new TCP_TXTLS_ENABLE socket option. All subsequent data transmitted on the socket is placed into TLS frames and encrypted using the supplied keys. Any data written to a KTLS-enabled socket via write(2), aio_write(2), or sendfile(2) is assumed to be application data and is encoded in TLS frames with an application data type. Individual records can be sent with a custom type (e.g. handshake messages) via sendmsg(2) with a new control message (TLS_SET_RECORD_TYPE) specifying the record type. At present, rekeying is not supported though the in-kernel framework should support rekeying. KTLS makes use of the recently added unmapped mbufs to store TLS frames in the socket buffer. Each TLS frame is described by a single ext_pgs mbuf. The ext_pgs structure contains the header of the TLS record (and trailer for encrypted records) as well as references to the associated TLS session. KTLS supports two primary methods of encrypting TLS frames: software TLS and ifnet TLS. Software TLS marks mbufs holding socket data as not ready via M_NOTREADY similar to sendfile(2) when TLS framing information is added to an unmapped mbuf in ktls_frame(). ktls_enqueue() is then called to schedule TLS frames for encryption. In the case of sendfile_iodone() calls ktls_enqueue() instead of pru_ready() leaving the mbufs marked M_NOTREADY until encryption is completed. For other writes (vn_sendfile when pages are available, write(2), etc.), the PRUS_NOTREADY is set when invoking pru_send() along with invoking ktls_enqueue(). A pool of worker threads (the "KTLS" kernel process) encrypts TLS frames queued via ktls_enqueue(). Each TLS frame is temporarily mapped using the direct map and passed to a software encryption backend to perform the actual encryption. (Note: The use of PHYS_TO_DMAP could be replaced with sf_bufs if someone wished to make this work on architectures without a direct map.) KTLS supports pluggable software encryption backends. Internally, Netflix uses proprietary pure-software backends. This commit includes a simple backend in a new ktls_ocf.ko module that uses the kernel's OpenCrypto framework to provide AES-GCM encryption of TLS frames. As a result, software TLS is now a bit of a misnomer as it can make use of hardware crypto accelerators. Once software encryption has finished, the TLS frame mbufs are marked ready via pru_ready(). At this point, the encrypted data appears as regular payload to the TCP stack stored in unmapped mbufs. ifnet TLS permits a NIC to offload the TLS encryption and TCP segmentation. In this mode, a new send tag type (IF_SND_TAG_TYPE_TLS) is allocated on the interface a socket is routed over and associated with a TLS session. TLS records for a TLS session using ifnet TLS are not marked M_NOTREADY but are passed down the stack unencrypted. The ip_output_send() and ip6_output_send() helper functions that apply send tags to outbound IP packets verify that the send tag of the TLS record matches the outbound interface. If so, the packet is tagged with the TLS send tag and sent to the interface. The NIC device driver must recognize packets with the TLS send tag and schedule them for TLS encryption and TCP segmentation. If the the outbound interface does not match the interface in the TLS send tag, the packet is dropped. In addition, a task is scheduled to refresh the TLS send tag for the TLS session. If a new TLS send tag cannot be allocated, the connection is dropped. If a new TLS send tag is allocated, however, subsequent packets will be tagged with the correct TLS send tag. (This latter case has been tested by configuring both ports of a Chelsio T6 in a lagg and failing over from one port to another. As the connections migrated to the new port, new TLS send tags were allocated for the new port and connections resumed without being dropped.) ifnet TLS can be enabled and disabled on supported network interfaces via new '[-]txtls[46]' options to ifconfig(8). ifnet TLS is supported across both vlan devices and lagg interfaces using failover, lacp with flowid enabled, or lacp with flowid enabled. Applications may request the current KTLS mode of a connection via a new TCP_TXTLS_MODE socket option. They can also use this socket option to toggle between software and ifnet TLS modes. In addition, a testing tool is available in tools/tools/switch_tls. This is modeled on tcpdrop and uses similar syntax. However, instead of dropping connections, -s is used to force KTLS connections to switch to software TLS and -i is used to switch to ifnet TLS. Various sysctls and counters are available under the kern.ipc.tls sysctl node. The kern.ipc.tls.enable node must be set to true to enable KTLS (it is off by default). The use of unmapped mbufs must also be enabled via kern.ipc.mb_use_ext_pgs to enable KTLS. KTLS is enabled via the KERN_TLS kernel option. This patch is the culmination of years of work by several folks including Scott Long and Randall Stewart for the original design and implementation; Drew Gallatin for several optimizations including the use of ext_pgs mbufs, the M_NOTREADY mechanism for TLS records awaiting software encryption, and pluggable software crypto backends; and John Baldwin for modifications to support hardware TLS offload. Reviewed by: gallatin, hselasky, rrs Obtained from: Netflix Sponsored by: Netflix, Chelsio Communications Differential Revision: https://reviews.freebsd.org/D21277
2019-08-27 00:01:56 +00:00
#if defined(KERN_TLS) || defined(RATELIMIT)
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
static int
vlan_snd_tag_alloc(struct ifnet *ifp,
union if_snd_tag_alloc_params *params,
struct m_snd_tag **ppmt)
{
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
struct epoch_tracker et;
struct vlan_snd_tag *vst;
struct ifvlan *ifv;
struct ifnet *parent;
int error;
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
NET_EPOCH_ENTER(et);
ifv = ifp->if_softc;
if (ifv->ifv_trunk != NULL)
parent = PARENT(ifv);
else
parent = NULL;
if (parent == NULL) {
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
NET_EPOCH_EXIT(et);
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
return (EOPNOTSUPP);
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
}
if_ref(parent);
NET_EPOCH_EXIT(et);
vst = malloc(sizeof(*vst), M_VLAN, M_NOWAIT);
if (vst == NULL) {
if_rele(parent);
return (ENOMEM);
}
error = m_snd_tag_alloc(parent, params, &vst->tag);
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
if_rele(parent);
if (error) {
free(vst, M_VLAN);
return (error);
}
m_snd_tag_init(&vst->com, ifp, vst->tag->type);
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
*ppmt = &vst->com;
return (0);
}
static struct m_snd_tag *
vlan_next_snd_tag(struct m_snd_tag *mst)
{
struct vlan_snd_tag *vst;
vst = mst_to_vst(mst);
return (vst->tag);
}
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
static int
vlan_snd_tag_modify(struct m_snd_tag *mst,
union if_snd_tag_modify_params *params)
{
struct vlan_snd_tag *vst;
vst = mst_to_vst(mst);
return (vst->tag->ifp->if_snd_tag_modify(vst->tag, params));
}
static int
vlan_snd_tag_query(struct m_snd_tag *mst,
union if_snd_tag_query_params *params)
{
struct vlan_snd_tag *vst;
vst = mst_to_vst(mst);
return (vst->tag->ifp->if_snd_tag_query(vst->tag, params));
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
}
static void
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
vlan_snd_tag_free(struct m_snd_tag *mst)
{
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
struct vlan_snd_tag *vst;
vst = mst_to_vst(mst);
m_snd_tag_rele(vst->tag);
free(vst, M_VLAN);
}
static void
vlan_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
{
/*
* For vlan, we have an indirect
* interface. The caller needs to
* get a ratelimit tag on the actual
* interface the flow will go on.
*/
q->rate_table = NULL;
q->flags = RT_IS_INDIRECT;
q->max_flows = 0;
q->number_of_rates = 0;
}
Implement kernel support for hardware rate limited sockets. - Add RATELIMIT kernel configuration keyword which must be set to enable the new functionality. - Add support for hardware driven, Receive Side Scaling, RSS aware, rate limited sendqueues and expose the functionality through the already established SO_MAX_PACING_RATE setsockopt(). The API support rates in the range from 1 to 4Gbytes/s which are suitable for regular TCP and UDP streams. The setsockopt(2) manual page has been updated. - Add rate limit function callback API to "struct ifnet" which supports the following operations: if_snd_tag_alloc(), if_snd_tag_modify(), if_snd_tag_query() and if_snd_tag_free(). - Add support to ifconfig to view, set and clear the IFCAP_TXRTLMT flag, which tells if a network driver supports rate limiting or not. - This patch also adds support for rate limiting through VLAN and LAGG intermediate network devices. - How rate limiting works: 1) The userspace application calls setsockopt() after accepting or making a new connection to set the rate which is then stored in the socket structure in the kernel. Later on when packets are transmitted a check is made in the transmit path for rate changes. A rate change implies a non-blocking ifp->if_snd_tag_alloc() call will be made to the destination network interface, which then sets up a custom sendqueue with the given rate limitation parameter. A "struct m_snd_tag" pointer is returned which serves as a "snd_tag" hint in the m_pkthdr for the subsequently transmitted mbufs. 2) When the network driver sees the "m->m_pkthdr.snd_tag" different from NULL, it will move the packets into a designated rate limited sendqueue given by the snd_tag pointer. It is up to the individual drivers how the rate limited traffic will be rate limited. 3) Route changes are detected by the NIC drivers in the ifp->if_transmit() routine when the ifnet pointer in the incoming snd_tag mismatches the one of the network interface. The network adapter frees the mbuf and returns EAGAIN which causes the ip_output() to release and clear the send tag. Upon next ip_output() a new "snd_tag" will be tried allocated. 4) When the PCB is detached the custom sendqueue will be released by a non-blocking ifp->if_snd_tag_free() call to the currently bound network interface. Reviewed by: wblock (manpages), adrian, gallatin, scottl (network) Differential Revision: https://reviews.freebsd.org/D3687 Sponsored by: Mellanox Technologies MFC after: 3 months
2017-01-18 13:31:17 +00:00
#endif