in_pcbfree will remove the incpb from the list and release the rtentry
while the vnet is set, but the actual destruction will be deferred
until any threads in a (not yet used) epoch section, no longer potentially
have references.
This patch adds a new socket option, SO_REUSEPORT_LB, which allow multiple
programs or threads to bind to the same port and incoming connections will be
load balanced using a hash function.
Most of the code was copied from a similar patch for DragonflyBSD.
However, in DragonflyBSD, load balancing is a global on/off setting and can not
be set per socket. This patch allows for simultaneous use of both the current
SO_REUSEPORT and the new SO_REUSEPORT_LB options on the same system.
Required changes to structures
Globally change so_options from 16 to 32 bit value to allow for more options.
Add hashtable in pcbinfo to hold all SO_REUSEPORT_LB sockets.
Limitations
As DragonflyBSD, a load balance group is limited to 256 pcbs
(256 programs or threads sharing the same socket).
Submitted by: Johannes Lundberg <johanlun0@gmail.com>
Sponsored by: Limelight Networks
Differential Revision: https://reviews.freebsd.org/D11003
It is the forerunner/foundational work of bringing in both Rack and BBR
which use hpts for pacing out packets. The feature is optional and requires
the TCPHPTS option to be enabled before the feature will be active. TCP
modules that use it must assure that the base component is compile in
the kernel in which they are loaded.
MFC after: Never
Sponsored by: Netflix Inc.
Differential Revision: https://reviews.freebsd.org/D15020
Mainly focus on files that use BSD 3-Clause license.
The Software Package Data Exchange (SPDX) group provides a specification
to make it easier for automated tools to detect and summarize well known
opensource licenses. We are gradually adopting the specification, noting
that the tags are considered only advisory and do not, in any way,
superceed or replace the license texts.
Special thanks to Wind River for providing access to "The Duke of
Highlander" tool: an older (2014) run over FreeBSD tree was useful as a
starting point.
transmit queues aswell as non-ratelimited ones.
Add the required structure bits in order to support a backpressure
indication with ratelimited connections aswell as non-ratelimited
ones. The backpressure indicator is a value between zero and 65535
inclusivly, indicating if the destination transmit queue is empty or
full respectivly. Applications can use this value as a decision point
for when to stop transmitting data to avoid endless ENOBUFS error
codes upon transmitting an mbuf. This indicator is also useful to
reduce the latency for ratelimited queues.
Reviewed by: gallatin, kib, gnn
Differential Revision: https://reviews.freebsd.org/D11518
Sponsored by: Mellanox Technologies
considering cache line hits and misses. Put the lock and hash list
glue into the first cache line, put inp_refcount inp_flags inp_socket
into the second cache line.
o On allocation zero out entire structure except the lock and list entries,
including inp_route inp_lle inp_gencnt. When inp_route and inp_lle were
introduced, they were added below inp_zero_size, resulting on not being
cleared after free/alloc. This definitely was a source of bugs with route
caching. Could be that r315956 has just fixed one of them.
The inp_gencnt is reinitialized on every alloc, so it is safe to clear it.
This has been proved to improve TCP performance at Netflix.
Obtained from: rrs
Differential Revision: D10686
function (they used to say UMA_ZONE_NOFREE), so flag parameter goes away.
The zone_fini parameter also goes away. Previously no protocols (except
divert) supplied zone_fini function, so inpcb locks were leaked with slabs.
This was okay while zones were allocated with UMA_ZONE_NOFREE flag, but now
this is a leak. Fix that by suppling inpcb_fini() function as fini method
for all inpcb zones.
This is a painful change, but it is needed. On the one hand, we avoid
modifying them, and this slows down some ideas, on the other hand we still
eventually modify them and tools like netstat(1) never work on next version of
FreeBSD. We maintain a ton of spares in them, and we already got some ifdef
hell at the end of tcpcb.
Details:
- Hide struct inpcb, struct tcpcb under _KERNEL || _WANT_FOO.
- Make struct xinpcb, struct xtcpcb pure API structures, not including
kernel structures inpcb and tcpcb inside. Export into these structures
the fields from inpcb and tcpcb that are known to be used, and put there
a ton of spare space.
- Make kernel and userland utilities compilable after these changes.
- Bump __FreeBSD_version.
Reviewed by: rrs, gnn
Differential Revision: D10018
Unfortunately they will have different integer value due to Linux value being already assigned in FreeBSD.
The patch is similar to IP_RECVDSTADDR but also provides the destination port value to the application.
This allows/improves implementation of transparent proxies on UDP sockets due to having the whole information on forwarded packets.
Reviewed by: adrian, aw
Approved by: ae (mentor)
Sponsored by: rsync.net
Differential Revision: D9235
Renumber cluase 4 to 3, per what everybody else did when BSD granted
them permission to remove clause 3. My insistance on keeping the same
numbering for legal reasons is too pedantic, so give up on that point.
Submitted by: Jan Schaumann <jschauma@stevens.edu>
Pull Request: https://github.com/freebsd/freebsd/pull/96
Unfortunately they will have different integer value due to Linux value being already assigned in FreeBSD.
The patch is similar to IP_RECVDSTADDR but also provides the destination port value to the application.
This allows/improves implementation of transparent proxies on UDP sockets due to having the whole information on forwarded packets.
Sponsored-by: rsync.net
Differential Revision: D9235
Reviewed-by: adrian
- 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
use. Update comments regarding the spare fields in struct inpcb.
Bump __FreeBSD_version for the changes to the size of the structures.
Reviewed by: gnn@
Approved by: re@ (gjb@)
Sponsored by: Chelsio Communications
route caching for TCP, with some improvements. In particular, invalidate
the route cache if a new route is added, which might be a better match.
The cache is automatically invalidated if the old route is deleted.
Submitted by: Mike Karels
Reviewed by: gnn
Differential Revision: https://reviews.freebsd.org/D4306
o Unlike xor, in Jenkins hash every bit of input affects virtually
every bit of output, thus salting the hash actually works. With
xor salting only provides a false sense of security, since if
hash(x) collides with hash(y), then of course, hash(x) ^ salt
would also collide with hash(y) ^ salt. [1]
o Jenkins provides much better distribution than xor, very close to
ideal.
TCP connection setup/teardown benchmark has shown a 10% increase
with default hash size, and with bigger hashes that still provide
possibility for collisions. With enormous hash size, when dataset is
by an order of magnitude smaller than hash size, the benchmark has
shown 4% decrease in performance decrease, which is expected and
acceptable.
Noticed by: Jeffrey Knockel <jeffk cs.unm.edu> [1]
Benchmarks by: jch
Reviewed by: jch, pkelsey, delphij
Security: strengthens protection against hash collision DoS
Sponsored by: Nginx, Inc.
Avoid too strict INP_INFO_RLOCK_ASSERT checks due to
tcp_notify() being called from in6_pcbnotify().
Reported by: Larry Rosenman <ler@lerctr.org>
Submitted by: markj, jch
- The existing TCP INP_INFO lock continues to protect the global inpcb list
stability during full list traversal (e.g. tcp_pcblist()).
- A new INP_LIST lock protects inpcb list actual modifications (inp allocation
and free) and inpcb global counters.
It allows to use TCP INP_INFO_RLOCK lock in critical paths (e.g. tcp_input())
and INP_INFO_WLOCK only in occasional operations that walk all connections.
PR: 183659
Differential Revision: https://reviews.freebsd.org/D2599
Reviewed by: jhb, adrian
Tested by: adrian, nitroboost-gmail.com
Sponsored by: Verisign, Inc.
from the FreeBSD network code. The flag is still kept around in the
"sys/mbuf.h" header file, but does no longer have any users. Instead
the "m_pkthdr.rsstype" field in the mbuf structure is now used to
decide the meaning of the "m_pkthdr.flowid" field. To modify the
"m_pkthdr.rsstype" field please use the existing "M_HASHTYPE_XXX"
macros as defined in the "sys/mbuf.h" header file.
This patch introduces new behaviour in the transmit direction.
Previously network drivers checked if "M_FLOWID" was set in "m_flags"
before using the "m_pkthdr.flowid" field. This check has now now been
replaced by checking if "M_HASHTYPE_GET(m)" is different from
"M_HASHTYPE_NONE". In the future more hashtypes will be added, for
example hashtypes for hardware dedicated flows.
"M_HASHTYPE_OPAQUE" indicates that the "m_pkthdr.flowid" value is
valid and has no particular type. This change removes the need for an
"if" statement in TCP transmit code checking for the presence of a
valid flowid value. The "if" statement mentioned above is now a direct
variable assignment which is then later checked by the respective
network drivers like before.
Additional notes:
- The SCTP code changes will be committed as a separate patch.
- Removal of the "M_FLOWID" flag will also be done separately.
- The FreeBSD version has been bumped.
MFC after: 1 month
Sponsored by: Mellanox Technologies
have chosen different (and more traditional) stateless/statuful
NAT64 as translation mechanism. Last non-trivial commits to both
faith(4) and faithd(8) happened more than 12 years ago, so I assume
it is time to drop RFC3142 in FreeBSD.
No objections from: net@
A non-global IPv6 address can be used in more than one zone of the same
scope. This zone index is used to identify to which zone a non-global
address belongs.
Also we can have many foreign hosts with equal non-global addresses,
but from different zones. So, they can have different metrics in the
host cache.
Obtained from: Yandex LLC
Sponsored by: Yandex LLC
information as part of recvmsg().
This is primarily used for debugging/verification of the various
processing paths in the IP, PCB and driver layers.
Unfortunately the current implementation of the control message path
results in a ~10% or so drop in UDP frame throughput when it's used.
Differential Revision: https://reviews.freebsd.org/D527
Reviewed by: grehan
awareness.
* Introduce IP_BINDMULTI - indicating that it's okay to bind multiple
sockets on the same bind details.
Although the PCB code has been taught about this (see below) this patch
doesn't introduce the rest of the PCB changes necessary to distribute
lookups among multiple PCB entries in the global wildcard table.
* Introduce IP_RSS_LISTEN_BUCKET - placing an listen socket into the
given RSS bucket (and thus a single PCBGROUP hash.)
* Modify the PCB add path to be aware of IP_BINDMULTI:
+ Only allow further PCB entries to be added if the owner credentials
and IP_BINDMULTI has been specified. Ie, only allow further
IP_BINDMULTI sockets to appear if the first bind() was IP_BINDMULTI.
* Teach the PCBGROUP code about IP_RSS_LISTE_BUCKET marked PCB entries.
Instead of using the wildcard logic and hashing, these sockets are
simply placed into the PCBGROUP and _not_ in the wildcard hash.
* When doing a PCBGROUP lookup, also do a wildcard match as well.
This allows for an RSS bucket PCB entry to appear in a PCBGROUP
rather than having to exist in the wildcard list.
Tested:
* TCP IPv4 server testing with igb(4)
* TCP IPv4 server testing with ix(4)
TODO:
* The pcbgroup lookup code duplicated the wildcard and wildcard-PCB
logic. This could be refactored into a single function.
* This doesn't yet work for IPv6 (The PCBGROUP code in netinet6/ doesn't
yet know about this); nor does it yet fully work for UDP.
dereferencing, when checking for SO_REUSEPORT option (and SO_REUSEADDR
for multicast), INP_REUSEPORT flag was introduced to cache the socket
option. It was decided then that one flag would be enough to cache
both SO_REUSEPORT and SO_REUSEADDR: when processing SO_REUSEADDR
setsockopt(2), it was checked if it was called for a multicast address
and INP_REUSEPORT was set accordingly.
Unfortunately that approach does not work when setsockopt(2) is called
before binding to a multicast address: the multicast check fails and
INP_REUSEPORT is not set.
Fix this by adding INP_REUSEADDR flag to unconditionally cache
SO_REUSEADDR.
PR: 179901
Submitted by: Michael Gmelin freebsd grem.de (initial version)
Reviewed by: rwatson
MFC after: 1 week
reside on their own cache line to prevent false sharing with other
nearby structures, especially for those in the .bss segment.
NB: Those mutexes and rwlocks with variables next to them that get
changed on every invocation do not benefit from their own cache line.
Actually it may be net negative because two cache misses would be
incurred in those cases.
Both functions need to obtain lock on the found PCB, and they can't do
classic inter-lock with the PCB hash lock, due to lock order reversal.
To keep the PCB stable, these functions put a reference on it and after PCB
lock is acquired drop it. If the reference was the last one, this means
we've raced with in_pcbfree() and the PCB is no longer valid.
This approach works okay only if we are acquiring writer-lock on the PCB.
In case of reader-lock, the following scenario can happen:
- 2 threads locate pcb, and do in_pcbref() on it.
- These 2 threads drop the inp hash lock.
- Another thread comes to delete pcb via in_pcbfree(), it obtains hash lock,
does in_pcbremlists(), drops hash lock, and runs in_pcbrele_wlocked(), which
doesn't free the pcb due to two references on it. Then it unlocks the pcb.
- 2 aforementioned threads acquire reader lock on the pcb and run
in_pcbrele_rlocked(). One gets 1 from in_pcbrele_rlocked() and continues,
second gets 0 and considers pcb freed, returns.
- The thread that got 1 continutes working with detached pcb, which later
leads to panic in the underlying protocol level.
To plumb that problem an additional INPCB flag introduced - INP_FREED. We
check for that flag in the in_pcbrele_rlocked() and if it is set, we pretend
that that was the last reference.
Discussed with: rwatson, jhb
Reported by: Vladimir Medvedkin <medved rambler-co.ru>
packets a cmsg of type IP_RECVTOS which contains the TOS byte.
Much like IP_RECVTTL does for TTL. This allows to implement a
protocol on top of UDP and implementing ECN.
MFC after: 3 days
inp_socket->so_options dereference when we may not acquire the lock on
the inpcb.
This fixes the crash due to NULL pointer dereference in
in_pcbbind_setup() when inp_socket->so_options in a pcb returned by
in_pcblookup_local() was checked.
Reported by: dave jones <s.dave.jones@gmail.com>, Arnaud Lacombe <lacombar@gmail.com>
Suggested by: rwatson
Glanced by: rwatson
Tested by: dave jones <s.dave.jones@gmail.com>
- TCP keep* timers
- TCP UTO (adjust from what was there already)
- netmap
- route caching
- user cookie (temporary to allow for the real fix)
Slightly re-shuffle struct ifnet moving fields out of the middle
of spares and to better align.
Discussed with: rwatson (slightly earlier version)
Rather than including lock.h in in_pcbgroup.c in right order, fix it
for all consumers of in_pcb.h by further header file pollution under
#ifdef KERNEL.
Reported by: Pan Tsu (inyaoo gmail.com)
struct inpcbgroup. pcbgroups, or "connection groups", supplement the
existing inpcbinfo connection hash table, which when pcbgroups are
enabled, might now be thought of more usefully as a per-protocol
4-tuple reservation table.
Connections are assigned to connection groups base on a hash of their
4-tuple; wildcard sockets require special handling, and are members
of all connection groups. During a connection lookup, a
per-connection group lock is employed rather than the global pcbinfo
lock. By aligning connection groups with input path processing,
connection groups take on an effective CPU affinity, especially when
aligned with RSS work placement (see a forthcoming commit for
details). This eliminates cache line migration associated with
global, protocol-layer data structures in steady state TCP and UDP
processing (with the exception of protocol-layer statistics; further
commit to follow).
Elements of this approach were inspired by Willman, Rixner, and Cox's
2006 USENIX paper, "An Evaluation of Network Stack Parallelization
Strategies in Modern Operating Systems". However, there are also
significant differences: we maintain the inpcb lock, rather than using
the connection group lock for per-connection state.
Likewise, the focus of this implementation is alignment with NIC
packet distribution strategies such as RSS, rather than pure software
strategies. Despite that focus, software distribution is supported
through the parallel netisr implementation, and works well in
configurations where the number of hardware threads is greater than
the number of NIC input queues, such as in the RMI XLR threaded MIPS
architecture.
Another important difference is the continued maintenance of existing
hash tables as "reservation tables" -- these are useful both to
distinguish the resource allocation aspect of protocol name management
and the more common-case lookup aspect. In configurations where
connection tables are aligned with hardware hashes, it is desirable to
use the traditional lookup tables for loopback or encapsulated traffic
rather than take the expense of hardware hashes that are hard to
implement efficiently in software (such as RSS Toeplitz).
Connection group support is enabled by compiling "options PCBGROUP"
into your kernel configuration; for the time being, this is an
experimental feature, and hence is not enabled by default.
Subject to the limited MFCability of change dependencies in inpcb,
and its change to the inpcbinfo init function signature, this change
in principle could be merged to FreeBSD 8.x.
Reviewed by: bz
Sponsored by: Juniper Networks, Inc.
hash install, etc. For now, these are arguments are unused, but as we add
RSS support, we will want to use hashes extracted from mbufs, rather than
manually calculated hashes of header fields, due to the expensive of the
software version of Toeplitz (and similar hashes).
Add notes that it would be nice to be able to pass mbufs into lookup
routines in pf(4), optimising firewall lookup in the same way, but the
code structure there doesn't facilitate that currently.
(In principle there is no reason this couldn't be MFCed -- the change
extends rather than modifies the KBI. However, it won't be useful without
other previous possibly less MFCable changes.)
Reviewed by: bz
Sponsored by: Juniper Networks, Inc.
- The existing ipi_lock continues to protect the global inpcb list and
inpcb counter. This lock is now relegated to a small number of
allocation and free operations, and occasional operations that walk
all connections (including, awkwardly, certain UDP multicast receive
operations -- something to revisit).
- A new ipi_hash_lock protects the two inpcbinfo hash tables for
looking up connections and bound sockets, manipulated using new
INP_HASH_*() macros. This lock, combined with inpcb locks, protects
the 4-tuple address space.
Unlike the current ipi_lock, ipi_hash_lock follows the individual inpcb
connection locks, so may be acquired while manipulating a connection on
which a lock is already held, avoiding the need to acquire the inpcbinfo
lock preemptively when a binding change might later be required. As a
result, however, lookup operations necessarily go through a reference
acquire while holding the lookup lock, later acquiring an inpcb lock --
if required.
A new function in_pcblookup() looks up connections, and accepts flags
indicating how to return the inpcb. Due to lock order changes, callers
no longer need acquire locks before performing a lookup: the lookup
routine will acquire the ipi_hash_lock as needed. In the future, it will
also be able to use alternative lookup and locking strategies
transparently to callers, such as pcbgroup lookup. New lookup flags are,
supplementing the existing INPLOOKUP_WILDCARD flag:
INPLOOKUP_RLOCKPCB - Acquire a read lock on the returned inpcb
INPLOOKUP_WLOCKPCB - Acquire a write lock on the returned inpcb
Callers must pass exactly one of these flags (for the time being).
Some notes:
- All protocols are updated to work within the new regime; especially,
TCP, UDPv4, and UDPv6. pcbinfo ipi_lock acquisitions are largely
eliminated, and global hash lock hold times are dramatically reduced
compared to previous locking.
- The TCP syncache still relies on the pcbinfo lock, something that we
may want to revisit.
- Support for reverting to the FreeBSD 7.x locking strategy in TCP input
is no longer available -- hash lookup locks are now held only very
briefly during inpcb lookup, rather than for potentially extended
periods. However, the pcbinfo ipi_lock will still be acquired if a
connection state might change such that a connection is added or
removed.
- Raw IP sockets continue to use the pcbinfo ipi_lock for protection,
due to maintaining their own hash tables.
- The interface in6_pcblookup_hash_locked() is maintained, which allows
callers to acquire hash locks and perform one or more lookups atomically
with 4-tuple allocation: this is required only for TCPv6, as there is no
in6_pcbconnect_setup(), which there should be.
- UDPv6 locking remains significantly more conservative than UDPv4
locking, which relates to source address selection. This needs
attention, as it likely significantly reduces parallelism in this code
for multithreaded socket use (such as in BIND).
- In the UDPv4 and UDPv6 multicast cases, we need to revisit locking
somewhat, as they relied on ipi_lock to stablise 4-tuple matches, which
is no longer sufficient. A second check once the inpcb lock is held
should do the trick, keeping the general case from requiring the inpcb
lock for every inpcb visited.
- This work reminds us that we need to revisit locking of the v4/v6 flags,
which may be accessed lock-free both before and after this change.
- Right now, a single lock name is used for the pcbhash lock -- this is
undesirable, and probably another argument is required to take care of
this (or a char array name field in the pcbinfo?).
This is not an MFC candidate for 8.x due to its impact on lookup and
locking semantics. It's possible some of these issues could be worked
around with compatibility wrappers, if necessary.
Reviewed by: bz
Sponsored by: Juniper Networks, Inc.
reworking of inpcbinfo locking:
(1) Convert inpcb reference counting from manually manipulated integers to
the refcount(9) KPI. This allows the refcount to be managed atomically
with an inpcb read lock rather than write lock, or even with no inpcb
lock at all. As a result, in_pcbref() also no longer requires an inpcb
lock, so can be performed solely using the lock used to look up an
inpcb.
(2) Shift more inpcb freeing activity from the in_pcbrele() context (via
in_pcbfree_internal) to the explicit in_pcbfree() context. This means
that the inpcb refcount is increasingly used only to maintain memory
stability, not actually defer the clean up of inpcb protocol parts.
This is desirable as many of those protocol parts required the pcbinfo
lock, which we'd like not to acquire in in_pcbrele() contexts. Document
this in comments better.
(3) Introduce new read-locked and write-locked in_pcbrele() variations,
in_pcbrele_rlocked() and in_pcbrele_wlocked(), which allow the inpcb to
be properly unlocked as needed. in_pcbrele() is a wrapper around the
latter, and should probably go away at some point. This makes it
easier to use this weak reference model when holding only a read lock,
as will happen in the future.
This may well be safe to MFC, but some more KBI analysis is required.
Reviewed by: bz
MFC after: 3 weeks
Sponsored by: Juniper Networks, Inc.
(1) Add a locking guide for inpcbinfo.
(2) Annotate inpcbinfo fields with synchronisation information; not all
annotations are 100% satisfactory.
(3) Reorder inpcbinfo fields so that the lock is at the head of the
structure, and close to fields it protects.
(4) Sort fields that will eventually be hashlock/pcbgroup-related together
even though they remain locked by ipi_lock for now.
Reviewed by: bz
Sponsored by: Juniper Networks
X-MFC after: KBI analysis required
Move the ipport_tick_callout and related functions from ip_input.c
to in_pcb.c. The random source port allocation code has been merged
and is now local to in_pcb.c only.
Use a SYSINIT to get the callout started and no longer depend on
initialization from the inet code, which would not work in an IPv6
only setup.
Reviewed by: gnn
Sponsored by: The FreeBSD Foundation
Sponsored by: iXsystems
MFC after: 4 days
