Pablo Neira Ayuso says:
====================
Netfilter updates for net-next
A final pull request, I know it's very late but this time I think it's worth a
bit of rush.
The following patchset contains Netfilter/nf_tables updates for net-next, more
specifically concatenation support and dynamic stateful expression
instantiation.
This also comes with a couple of small patches. One to fix the ebtables.h
userspace header and another to get rid of an obsolete example file in tree
that describes a nf_tables expression.
This time, I decided to paste the original descriptions. This will result in a
rather large commit description, but I think these bytes to keep.
Patrick McHardy says:
====================
netfilter: nf_tables: concatenation support
The following patches add support for concatenations, which allow multi
dimensional exact matches in O(1).
The basic idea is to split the data registers, currently consisting of
4 registers of 16 bytes each, into smaller units, 16 registers of 4
bytes each, and making sure each register store always leaves the
full 32 bit in a well defined state, meaning smaller stores will
zero the remaining bits.
Based on that, we can load multiple adjacent registers with different
values, thereby building a concatenated bigger value, and use that
value for set lookups.
Sets are changed to use variable sized extensions for their key and
data values, removing the fixed limit of 16 bytes while saving memory
if less space is needed.
As a side effect, these patches will allow some nice optimizations in
the future, like using jhash2 in nft_hash, removing the masking in
nft_cmp_fast, optimized data comparison using 32 bit word size etc.
These are not done so far however.
The patches are split up as follows:
* the first five patches add length validation to register loads and
stores to make sure we stay within bounds and prepare the validation
functions for the new addressing mode
* the next patches prepare for changing to 32 bit addressing by
introducing a struct nft_regs, which holds the verdict register as
well as the data registers. The verdict members are moved to a new
struct nft_verdict to allow to pull struct nft_data out of the stack.
* the next patches contain preparatory conversions of expressions and
sets to use 32 bit addressing
* the next patch introduces so far unused register conversion helpers
for parsing and dumping register numbers over netlink
* following is the real conversion to 32 bit addressing, consisting of
replacing struct nft_data in struct nft_regs by an array of u32s and
actually translating and validating the new register numbers.
* the final two patches add support for variable sized data items and
variable sized keys / data in set elements
The patches have been verified to work correctly with nft binaries using
both old and new addressing.
====================
Patrick McHardy says:
====================
netfilter: nf_tables: dynamic stateful expression instantiation
The following patches are the grand finale of my nf_tables set work,
using all the building blocks put in place by the previous patches
to support something like iptables hashlimit, but a lot more powerful.
Sets are extended to allow attaching expressions to set elements.
The dynset expression dynamically instantiates these expressions
based on a template when creating new set elements and evaluates
them for all new or updated set members.
In combination with concatenations this effectively creates state
tables for arbitrary combinations of keys, using the existing
expression types to maintain that state. Regular set GC takes care
of purging expired states.
We currently support two different stateful expressions, counter
and limit. Using limit as a template we can express the functionality
of hashlimit, but completely unrestricted in the combination of keys.
Using counter we can perform accounting for arbitrary flows.
The following examples from patch 5/5 show some possibilities.
Userspace syntax is still WIP, especially the listing of state
tables will most likely be seperated from normal set listings
and use a more structured format:
1. Limit the rate of new SSH connections per host, similar to iptables
hashlimit:
flow ip saddr timeout 60s \
limit 10/second \
accept
2. Account network traffic between each set of /24 networks:
flow ip saddr & 255.255.255.0 . ip daddr & 255.255.255.0 \
counter
3. Account traffic to each host per user:
flow skuid . ip daddr \
counter
4. Account traffic for each combination of source address and TCP flags:
flow ip saddr . tcp flags \
counter
The resulting set content after a Xmas-scan look like this:
{
192.168.122.1 . fin | psh | urg : counter packets 1001 bytes 40040,
192.168.122.1 . ack : counter packets 74 bytes 3848,
192.168.122.1 . psh | ack : counter packets 35 bytes 3144
}
In the future the "expressions attached to elements" will be extended
to also support user created non-stateful expressions to allow to
efficiently select beween a set of parameter sets, f.i. a set of log
statements with different prefixes based on the interface, which currently
require one rule each. This will most likely have to wait until the next
kernel version though.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The dwmac-socfpga.c conflict was a case of a bug fix overlapping
changes in net-next to handle an error pointer differently.
Signed-off-by: David S. Miller <davem@davemloft.net>
Hariprasad Shenai says:
====================
cxgb4: Misc. fixes for sge
Increases value of MAX_IMM_TX_PKT_LEN to improve latency, fill freelist
starving threshold based on adapter type, add comments for tx flits and sge
length code and don't call t4_slow_intr_handler when we are not master PF.
This patch series has been created against net-next tree and includes patches on
cxgb4 driver
We have included all the maintainers of respective drivers. Kindly review the
change and let us know in case of any review comments.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Add comment for tx filt and sge length calucaltion code, also remove
a hardcoded value
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
fl_starv_thres could be different from adapter to adapter, don't use
hardcoded values
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This allows a significant latency drop for packets of sizes between 128 and 192
bytes
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ring size is always known at compile time, so make the code a bit
more efficient
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver needs to inform the hardware about the first invalid (not yet
filled) rx slot, by writing its DMA descriptor pointer offset to the
BGMAC_DMA_RX_INDEX register.
This register was set to a value exceeding the rx ring size, effectively
allowing the hardware constant access to the full ring, regardless of
which slots are initialized.
To fix this issue, always mark the last filled rx slot as invalid.
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of allocating buffers at device init time and initializing
descriptors at device open, do both at the same time (during open).
Free all buffers when closing the device.
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Acked-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Limiting it to 511 looks like a failed attempt at leaving one descriptor
empty to allow the hardware to stop processing a buffer that has not
been prepared yet. However, this doesn't work because this affects the
total ring size as well
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In very rare cases, the MAC can catch an internal buffer that is bigger
than it's supposed to be. Instead of crashing the kernel, simply pass
the buffer back to the hardware
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allocate a new buffer before processing the completed one. If allocation
fails, reuse the old buffer.
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Acked-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
A packet buffer offset of 30 bytes is inefficient, because the first 2
bytes end up in a different cacheline.
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Always poll rx and tx during NAPI poll instead of relying on the status
of the first interrupt. This prevents bgmac_poll from leaving unfinished
work around until the next IRQ.
In my tests this makes bridging/routing throughput under heavy load more
stable and ensures that no new IRQs arrive as long as bgmac_poll uses up
the entire budget.
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Keep incrementing ring->start and ring->end instead of pointing it to
the actual ring slot entry. This simplifies the calculation of the
number of free slots.
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Acked-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If remaining space in a send buffer slot is too small for the whole message,
we only copy the RNDIS header and PPI data into send buffer, so we can batch
one more packet each time. It reduces the vmbus per-message overhead.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Al Viro says:
====================
netdev-related stuff in vfs.git
There are several commits sitting in vfs.git that probably ought to go in
via net-next.git. First of all, there's merge with vfs.git#iocb - that's
Christoph's aio rework, which has triggered conflicts with the ->sendmsg()
and ->recvmsg() patches a while ago. It's not so much Christoph's stuff
that ought to be in net-next, as (pretty simple) conflict resolution on merge.
The next chunk is switch to {compat_,}import_iovec/import_single_range - new
safer primitives for initializing iov_iter. The primitives themselves come
from vfs/git#iov_iter (and they are used quite a lot in vfs part of queue),
conversion of net/socket.c syscalls belongs in net-next, IMO. Next there's
afs and rxrpc stuff from dhowells. And then there's sanitizing kernel_sendmsg
et.al. + missing inlined helper for "how much data is left in msg->msg_iter" -
this stuff is used in e.g. cifs stuff, but it belongs in net-next.
That pile is pullable from
git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git for-davem
I'll post the individual patches in there in followups; could you take a look
and tell if everything in there is OK with you?
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Using a timer wheel for timewait sockets was nice ~15 years ago when
memory was expensive and machines had a single processor.
This does not scale, code is ugly and source of huge latencies
(Typically 30 ms have been seen, cpus spinning on death_lock spinlock.)
We can afford to use an extra 64 bytes per timewait sock and spread
timewait load to all cpus to have better behavior.
Tested:
On following test, /proc/sys/net/ipv4/tcp_tw_recycle is set to 1
on the target (lpaa24)
Before patch :
lpaa23:~# ./super_netperf 200 -H lpaa24 -t TCP_CC -l 60 -- -p0,0
419594
lpaa23:~# ./super_netperf 200 -H lpaa24 -t TCP_CC -l 60 -- -p0,0
437171
While test is running, we can observe 25 or even 33 ms latencies.
lpaa24:~# ping -c 1000 -i 0.02 -qn lpaa23
...
1000 packets transmitted, 1000 received, 0% packet loss, time 20601ms
rtt min/avg/max/mdev = 0.020/0.217/25.771/1.535 ms, pipe 2
lpaa24:~# ping -c 1000 -i 0.02 -qn lpaa23
...
1000 packets transmitted, 1000 received, 0% packet loss, time 20702ms
rtt min/avg/max/mdev = 0.019/0.183/33.761/1.441 ms, pipe 2
After patch :
About 90% increase of throughput :
lpaa23:~# ./super_netperf 200 -H lpaa24 -t TCP_CC -l 60 -- -p0,0
810442
lpaa23:~# ./super_netperf 200 -H lpaa24 -t TCP_CC -l 60 -- -p0,0
800992
And latencies are kept to minimal values during this load, even
if network utilization is 90% higher :
lpaa24:~# ping -c 1000 -i 0.02 -qn lpaa23
...
1000 packets transmitted, 1000 received, 0% packet loss, time 19991ms
rtt min/avg/max/mdev = 0.023/0.064/0.360/0.042 ms
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The printed values are all of type unsigned integer, therefore use
%u instead of %d. Otherwise an user can face negative values.
Signed-off-by: Richard Weinberger <richard@nod.at>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The printed values are all of type unsigned integer, therefore use
%u instead of %d. Otherwise an user can face negative values.
Fixes:
$ cat /proc/net/netfilter/nfnetlink_queue
0 29508 278 2 65531 0 2004213241 -2129885586 1
1 -27747 0 2 65531 0 0 0 1
2 -27748 0 2 65531 0 0 0 1
Signed-off-by: Richard Weinberger <richard@nod.at>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The netlink portid is an unsigned integer, use this type
also in netfilter.
Signed-off-by: Richard Weinberger <richard@nod.at>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
portid is an unsigned integer. Fix urelease_work to
match all other portid user in the kernel.
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: David S. Miller <davem@davemloft.net>
portid is an unsigned integer. Fix netlink_notify to
match all other portid user in the kernel.
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: David S. Miller <davem@davemloft.net>
There's an example net/netfilter/nft_expr_template.c example file in tree that
got out of sync along time, remove it.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Acked-by: Patrick McHardy <kaber@trash.net>
Support instantiating stateful expressions based on a template that
are associated with dynamically created set entries. The expressions
are evaluated when adding or updating the set element.
This allows to maintain per flow state using the existing set
infrastructure and expression types, with arbitrary definitions of
a flow.
Usage is currently restricted to anonymous sets, meaning only a single
binding can exist, since the desired semantics of multiple independant
bindings haven't been defined so far.
Examples (userspace syntax is still WIP):
1. Limit the rate of new SSH connections per host, similar to iptables
hashlimit:
flow ip saddr timeout 60s \
limit 10/second \
accept
2. Account network traffic between each set of /24 networks:
flow ip saddr & 255.255.255.0 . ip daddr & 255.255.255.0 \
counter
3. Account traffic to each host per user:
flow skuid . ip daddr \
counter
4. Account traffic for each combination of source address and TCP flags:
flow ip saddr . tcp flags \
counter
The resulting set content after a Xmas-scan look like this:
{
192.168.122.1 . fin | psh | urg : counter packets 1001 bytes 40040,
192.168.122.1 . ack : counter packets 74 bytes 3848,
192.168.122.1 . psh | ack : counter packets 35 bytes 3144
}
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Add a set flag to indicate that the set is used as a state table and
contains expressions for evaluation. This operation is mutually
exclusive with the mapping operation, so sets specifying both are
rejected. The lookup expression also rejects binding to state tables
since it only deals with loopup and map operations.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Add a flag to mark stateful expressions.
This is used for dynamic expression instanstiation to limit the usable
expressions. Strictly speaking only the dynset expression can not be
used in order to avoid recursion, but since dynamically instantiating
non-stateful expressions will simply create an identical copy, which
behaves no differently than the original, this limits to expressions
where it actually makes sense to dynamically instantiate them.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Preparation to attach expressions to set elements: add a set extension
type to hold an expression and dump the expression information with the
set element.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Add helper functions for initializing, cloning, dumping and destroying
a single expression that is not part of a rule.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
linux/if.h creates conflicts in userspace with net/if.h
By using it here we force userspace to use linux/if.h while
net/if.h may be needed.
Note that:
include/linux/netfilter_ipv4/ip_tables.h and
include/linux/netfilter_ipv6/ip6_tables.h
don't include linux/if.h and they also refer to IFNAMSIZ, so they are
expecting userspace to include use net/if.h from the client program.
Signed-off-by: Arturo Borrero Gonzalez <arturo.borrero.glez@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Since retransmitted segments are not used for RTT estimation, previously
SACKed segments present in the rtx queue are used. This estimation can be
several times larger than the actual RTT. When a cumulative ack covers both
previously SACKed and retransmitted segments, CC may thus get a bogus RTT.
Such segments previously had an RTT estimation in tcp_sacktag_one(), so it
seems reasonable to not reuse them in tcp_clean_rtx_queue() at all.
Afaik, this has had no effect on SRTT/RTO because of Karn's check.
Signed-off-by: Kenneth Klette Jonassen <kennetkl@ifi.uio.no>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Even if we make use of classifier and actions from the egress
path, we're going into handle_ing() executing additional code
on a per-packet cost for ingress qdisc, just to realize that
nothing is attached on ingress.
Instead, this can just be blinded out as a no-op entirely with
the use of a static key. On input fast-path, we already make
use of static keys in various places, e.g. skb time stamping,
in RPS, etc. It makes sense to not waste time when we're assured
that no ingress qdisc is attached anywhere.
Enabling/disabling of that code path is being done via two
helpers, namely net_{inc,dec}_ingress_queue(), that are being
invoked under RTNL mutex when a ingress qdisc is being either
initialized or destructed.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Thomas Graf says:
====================
Bring sizeof(net_device) down to < 2K bytes
The size of struct net_device crossed the 2K boundary a while ago which
is a waste in combination with many net namespaces. This series brings
the size of struct net_device down to well below 2K in total size with
a typical configuration. Some reserves a several holes leave room for
further expansion.
Before:
/* size: 2176, cachelines: 34, members: 121 */
After:
/* size: 1984, cachelines: 31, members: 120 */
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Some trivial reorders while preserving the RX/TX cache lines
split to fill a couple of holes.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
e1000e is the only driver requiring pm_qos_req, instead of causing
every device to waste up to 240 bytes. Allocate it for the specific
driver.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When a new rtnl or xfrm command is added, this part of the code is frequently
missing. Let's help the developer with a build time test.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch changes sets to support variable sized set element keys / data
up to 64 bytes each by using variable sized set extensions. This allows
to use concatenations with bigger data items suchs as IPv6 addresses.
As a side effect, small keys/data now don't require the full 16 bytes
of struct nft_data anymore but just the space they need.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Add a size argument to nft_data_init() and pass in the available space.
This will be used by the following patches to support variable sized
set element data.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Switch the nf_tables registers from 128 bit addressing to 32 bit
addressing to support so called concatenations, where multiple values
can be concatenated over multiple registers for O(1) exact matches of
multiple dimensions using sets.
The old register values are mapped to areas of 128 bits for compatibility.
When dumping register numbers, values are expressed using the old values
if they refer to the beginning of a 128 bit area for compatibility.
To support concatenations, register loads of less than a full 32 bit
value need to be padded. This mainly affects the payload and exthdr
expressions, which both unconditionally zero the last word before
copying the data.
Userspace fully passes the testsuite using both old and new register
addressing.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Add helper functions to parse and dump register values in netlink attributes.
These helpers will later be changed to take care of translation between the
old 128 bit and the new 32 bit register numbers.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Simple conversion to use u32 pointers to the beginning of the data
area to keep follow up patches smaller.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Only needlessly complicates things due to requiring specific argument
types. Use memcmp directly.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Simple conversion to use u32 pointers to the beginning of the registers
to keep follow up patches smaller.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Replace the array of registers passed to expressions by a struct nft_regs,
containing the verdict as a seperate member, which aliases to the
NFT_REG_VERDICT register.
This is needed to seperate the verdict from the data registers completely,
so their size can be changed.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Change nft_validate_input_register() to not only validate the input
register number, but also the length of the load, and rename it to
nft_validate_register_load() to reflect that change.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
All users of nft_validate_register_store() first invoke
nft_validate_output_register(). There is in fact no use for using it
on its own, so simplify the code by folding the functionality into
nft_validate_register_store() and kill it.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
