For each CPU vendor that implements CPU microcode patching, there will
be a minimum family for which this is implemented. Verify this
minimum level of support.
This can be done in the dispatch function or early in the application
functions. Doing the latter turned out to be somewhat awkward because
of the ineviable split between the BSP and the AP paths, and rather
than pushing deep into the application functions, do this in
the dispatch function.
Reported-by: "Bryan O'Donoghue" <bryan.odonoghue.lkml@nexus-software.ie>
Suggested-by: Borislav Petkov <bp@alien8.de>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Link: http://lkml.kernel.org/r/1366392183-4149-1-git-send-email-bryan.odonoghue.lkml@nexus-software.ie
If one does do something unfortunate and allow a
bad offload bug into the kernel, this the
skb_warn_bad_offload can effectively live-lock the
system, filling the logs with the same error over
and over.
Add rate limitation to this so that box remains otherwise
functional in this case.
Signed-off-by: Ben Greear <greearb@candelatech.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When building ax88796 on an ARM platform with 64-bit resource_size_t,
we currently get
drivers/net/ethernet/8390/ax88796.c:875: undefined reference to `__aeabi_uldivmod'
because we do a division on the length of the MMIO resource.
Since we know that this resource is very short, using an
"unsigned long" instead of "resource_size_t" is entirely
sufficient, and avoids this link-time error.
Cc: Ben Dooks <ben-linux@fluff.org>
Cc: netdev@vger.kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pablo Neira Ayuso says:
====================
The following patchset contains a small batch of Netfilter
updates for your net-next tree, they are:
* Three patches that provide more accurate error reporting to
user-space, instead of -EPERM, in IPv4/IPv6 netfilter re-routing
code and NAT, from Patrick McHardy.
* Update copyright statements in Netfilter filters of
Patrick McHardy, from himself.
* Add Kconfig dependency on the raw/mangle tables to the
rpfilter, from Florian Westphal.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Autoneg is supported on specific port types only. Fix the driver to advertise
autoneg based on the port type.
Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
o Fix the driver to drop error frames in the receive path
o Update error counter which was not getting incremented
Signed-off-by: Sritej Velaga <sritej.velaga@qlogic.com>
Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Bjørn Mork says:
====================
This series adds workarounds for 3 different firmware bugs, each
preventing the affected devices from working at all. I therefore
humbly request that these fixes go to stable-3.8 (if still
maintained) and 3.9 (either via net if still possible, or via
stable if not).
All 3 workarounds are applied to all devices supported by the driver.
Adding quirks for specific devices was considered as an alternative,
but was rejected because we have too little information about the
exact distribution of the buggy firmwares. All we know is that the
same bug shows up in devices from at least 3 different, and presumably
independent, vendors.
The workarounds have instead been designed to automatically apply
when necessary, and to have as little impact as possible on unaffected
devices. The series has been tested on a number of devices both with
and without these bugs.
The series should apply cleanly to net/master, net-next/master and
stable/linux-3.8.y
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
We normally trust and use the CDC functional descriptors provided by a
number of devices. But some of these will erroneously list the address
reserved for the device end of the link. Attempting to use this on
both the device and host side will naturally not work.
Work around this bug by ignoring the functional descriptor and assign a
random address instead in this case.
Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
Received packets are sometimes addressed to 00:a0:c6:00:00:00
instead of the address the device firmware should have learned
from the host:
321.224126 77.16.85.204 -> 148.122.171.134 ICMP 98 Echo (ping) request id=0x4025, seq=64/16384, ttl=64
0000 82 c0 82 c9 f1 67 82 c0 82 c9 f1 67 08 00 45 00 .....g.....g..E.
0010 00 54 00 00 40 00 40 01 57 cc 4d 10 55 cc 94 7a .T..@.@.W.M.U..z
0020 ab 86 08 00 62 fc 40 25 00 40 b2 bc 6e 51 00 00 ....b.@%.@..nQ..
0030 00 00 6b bd 09 00 00 00 00 00 10 11 12 13 14 15 ..k.............
0040 16 17 18 19 1a 1b 1c 1d 1e 1f 20 21 22 23 24 25 .......... !"#$%
0050 26 27 28 29 2a 2b 2c 2d 2e 2f 30 31 32 33 34 35 &'()*+,-./012345
0060 36 37 67
321.240607 148.122.171.134 -> 77.16.85.204 ICMP 98 Echo (ping) reply id=0x4025, seq=64/16384, ttl=55
0000 00 a0 c6 00 00 00 02 50 f3 00 00 00 08 00 45 00 .......P......E.
0010 00 54 00 56 00 00 37 01 a0 76 94 7a ab 86 4d 10 .T.V..7..v.z..M.
0020 55 cc 00 00 6a fc 40 25 00 40 b2 bc 6e 51 00 00 U...j.@%.@..nQ..
0030 00 00 6b bd 09 00 00 00 00 00 10 11 12 13 14 15 ..k.............
0040 16 17 18 19 1a 1b 1c 1d 1e 1f 20 21 22 23 24 25 .......... !"#$%
0050 26 27 28 29 2a 2b 2c 2d 2e 2f 30 31 32 33 34 35 &'()*+,-./012345
0060 36 37 67
The bogus address is always the same, and matches the address
suggested by many devices as a default address. It is likely a
hardcoded firmware default.
The circumstances where this bug has been observed indicates that
the trigger is related to timing or some other factor the host
cannot control. Repeating the exact same configuration sequence
that caused it to trigger once, will not necessarily cause it to
trigger the next time. Reproducing the bug is therefore difficult.
This opens up a possibility that the bug is more common than we can
confirm, because affected devices often will work properly again
after a reset. A procedure most users are likely to try out before
reporting a bug.
Unconditionally rewriting the destination address if the first digit
of the received packet is 0, is considered an acceptable compromise
since we already have to inspect this digit. The simplification will
cause unnecessary rewrites if the real address starts with 0, but this
is still better than adding additional tests for this particular case.
Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
A number of LTE devices from different vendors all suffer from the
same firmware bug: Most of the packets received from the device while
it is attached to a LTE network will not have an ethernet header. The
devices work as expected when attached to 2G or 3G networks, sending
an ethernet header with all packets.
This driver is not aware of which network the modem attached to, and
even if it were there are still some packet types which are always
received with the header intact.
All devices supported by this driver have severely limited
networking capabilities:
- can only transmit IPv4, IPv6 and possibly ARP
- can only support a single host hardware address at any time
- will only do point-to-point communcation with the host
Because of this, we are able to reliably identify any bogus raw IP
packets by simply looking at the 4 IP version bits. All we need to
do is to avoid 4 or 6 in the first digit of the mac address. This
workaround ensures this, and fix up the received packets as necessary.
Given the distribution of the bug, it is believed that the source is
the chipset vendor. The devices which are verified to be affected are:
Huawei E392u-12 (Qualcomm MDM9200)
Pantech UML290 (Qualcomm MDM9600)
Novatel USB551L (Qualcomm MDM9600)
Novatel E362 (Qualcomm MDM9600)
It is believed that the bug depend on firmware revision, which means
that possibly all devices based on the above mentioned chipset may be
affected if we consider all available firmware revisions.
The information about affected devices and versions is likely
incomplete. As the additional overhead for packets not needing this
fixup is very small, it is considered acceptable to apply the
workaround to all devices handled by this driver.
Reported-by: Dan Williams <dcbw@redhat.com>
Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
Nikolay Aleksandrov says:
====================
This patch-set fixes mainly bugs on enslave failure and one occasion
of a needed locking. The patches are:
1. On enslave failure mc addresses are not flushed from the slave
2. On enslave failure vlans are not cleaned up from the slave
3. On enslave failure the bond's primary and curr_active_slave
are not cleaned up (which might result in use of freed memory)
4. On enslave failure netpoll is not disabled which might result in
a memory leak
5. In bond_mc_swap() the bond's mc addr list is walked without
netif_addr_lock, since it can be called without rtnl, add it
v2: patch 01 - fix log message and remove unnecessary code move
====================
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use netif_addr_lock_bh() to acquire the appropriate lock before walking.
Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
slave_disable_netpoll() is not called upon enslave failure which would
lead to a memory leak. Call slave_disable_netpoll() after err_detach as
that's the first error path after enabling netpoll on that slave.
Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On enslave failure primary_slave can point to new_slave which is to be
freed, and the same applies to curr_active_slave. So check if this is
the case and clean up properly after err_detach because that's the first
error code path after they're set.
Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The main problem is with vid refcount which only gets bumped up.
Delete the vlans after err_detach as that's the first error path
after the vlans are added.
Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add bond_mc_list_flush() after err_detach as that's the first error path
after the addresses are added. The main issue is the mc addresses' refcount
which only gets bumped up.
v2: update log message and don't move code unnecessarily
Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix to return -EINVAL when tb[TCA_FW_MASK] is set and head->mask != 0xFFFFFFFF
instead of 0 (ifdef CONFIG_NET_CLS_IND and tb[TCA_FW_INDEV]), as done elsewhere
in this function.
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The "reason" can come from skb->data[] and it hasn't been capped so it
can be from 0-255 instead of just 0-6. For example in irlmp_state_dtr()
the code does:
reason = skb->data[3];
...
irlmp_disconnect_indication(self, reason, skb);
Also LMREASON has a couple other values which don't have entries in the
irlmp_reasons[] array. And 0xff is a valid reason as well which means
"unknown".
So far as I can see we don't actually care about "reason" except for in
the debug code.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As reported by Dave Kleikamp, when we emit cross calls to do batched
TLB flush processing we have a race because we do not synchronize on
the sibling cpus completing the cross call.
So meanwhile the TLB batch can be reset (tb->tlb_nr set to zero, etc.)
and either flushes are missed or flushes will flush the wrong
addresses.
Fix this by using generic infrastructure to synchonize on the
completion of the cross call.
This first required getting the flush_tlb_pending() call out from
switch_to() which operates with locks held and interrupts disabled.
The problem is that smp_call_function_many() cannot be invoked with
IRQs disabled and this is explicitly checked for with WARN_ON_ONCE().
We get the batch processing outside of locked IRQ disabled sections by
using some ideas from the powerpc port. Namely, we only batch inside
of arch_{enter,leave}_lazy_mmu_mode() calls. If we're not in such a
region, we flush TLBs synchronously.
1) Get rid of xcall_flush_tlb_pending and per-cpu type
implementations.
2) Do TLB batch cross calls instead via:
smp_call_function_many()
tlb_pending_func()
__flush_tlb_pending()
3) Batch only in lazy mmu sequences:
a) Add 'active' member to struct tlb_batch
b) Define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
c) Set 'active' in arch_enter_lazy_mmu_mode()
d) Run batch and clear 'active' in arch_leave_lazy_mmu_mode()
e) Check 'active' in tlb_batch_add_one() and do a synchronous
flush if it's clear.
4) Add infrastructure for synchronous TLB page flushes.
a) Implement __flush_tlb_page and per-cpu variants, patch
as needed.
b) Likewise for xcall_flush_tlb_page.
c) Implement smp_flush_tlb_page() to invoke the cross-call.
d) Wire up global_flush_tlb_page() to the right routine based
upon CONFIG_SMP
5) It turns out that singleton batches are very common, 2 out of every
3 batch flushes have only a single entry in them.
The batch flush waiting is very expensive, both because of the poll
on sibling cpu completeion, as well as because passing the tlb batch
pointer to the sibling cpus invokes a shared memory dereference.
Therefore, in flush_tlb_pending(), if there is only one entry in
the batch perform a completely asynchronous global_flush_tlb_page()
instead.
Reported-by: Dave Kleikamp <dave.kleikamp@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Dave Kleikamp <dave.kleikamp@oracle.com>
cyc_to_sched_clock() is called by sched_clock() and cyc_to_ns()
is called by cyc_to_sched_clock(). I suspect that some compilers
inline both of these functions into sched_clock() and so we've
been getting away without having a notrace marking. It seems that
my compiler isn't inlining cyc_to_sched_clock() though, so I'm
hitting a recursion bug when I enable the function graph tracer,
causing my system to crash. Marking these functions notrace fixes
it. Technically cyc_to_ns() doesn't need the notrace because it's
already marked inline, but let's just add it so that if we ever
remove inline from that function it doesn't blow up.
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This patch adds support for the get_settings ethtool op to the bonding
driver. This was motivated by users who wanted to get the speed of the
bond and compare that against throughput to understand utilization.
The behavior before this patch was added was problematic when computing
line utilization after trying to get link-speed and throughput via SNMP.
Output from ethtool looks like this for a round-robin bond:
Settings for bond0:
Supported ports: [ ]
Supported link modes: Not reported
Supported pause frame use: No
Supports auto-negotiation: No
Advertised link modes: Not reported
Advertised pause frame use: No
Advertised auto-negotiation: No
Speed: 11000Mb/s
Duplex: Full
Port: Other
PHYAD: 0
Transceiver: internal
Auto-negotiation: off
MDI-X: Unknown
Link detected: yes
I tested this and verified it works as expected. A test was also done
on a version backported to an older kernel and it worked well there.
v2: Switch to using ethtool_cmd_speed_set to set speed, added check to
SLAVE_IS_OK for each slave in bond, dropped mode-specific calculations
as they were not needed, and set port type to 'Other.'
v3: Fix useless assignment and checkpatch warning.
Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
Reviewed-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch introduces a small, internal helper function, that is used by
PF_PACKET. Based on the flags that are passed, it extracts the packet
timestamp in the receive path. This is merely a refactoring to remove
some duplicate code in tpacket_rcv(), to make it more readable, and to
enable others to use this function in PF_PACKET as well, e.g. for TX.
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, ktime2ts is a small helper function that is only used in
net/socket.c. Move this helper into the ktime API as a small inline
function, so that i) it's maintained together with ktime routines,
and ii) also other files can make use of it. The function is named
ktime_to_timespec_cond() and placed into the generic part of ktime,
since we internally make use of ktime_to_timespec(). ktime_to_timespec()
itself does not check the ktime variable for zero, hence, we name
this function ktime_to_timespec_cond() for only a conditional
conversion, and adapt its users to it.
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rajesh Borundia says:
====================
* "qlcnic: Change 82xx adapter VLAN id endian type".
- Adapter requires VLAN id in little endian. VLAN id was being
converted to __le16 and then passed as a parameter. Pass VLAN id
as u16 and then use cpu_to_le16 at appropriate places. It is
appropriate for net-next as SR-IOV patches have a dependency on it.
* "qlcnic: Fix loopback test for SR-IOV PF".
- It is appropriate for net-next as change is needed for SRIOV PF
only.
* Remaining patches add enhancements to SR-IOV functionality like
- FLR handling
- Adapter reset recovery handling
- iproute2 tool support for configuring MAC address, Tx rate and
VLAN id.
- Mailbox polling support for SR-IOV PF in case mailbox interrupts
are disabled.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
o When mailbox interrupt is disabled PF should be
able to process request from VF. Enable polling
for such cases.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
o Do not disable mailbox interrupts while running
loopback test through SR-IOV PF.
Signed-off-by: Manish Chopra <manish.chopra@qlogic.com>
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
o Add support for VLAN id configuration per VF using
iproute2 tool.
o VLAN id's 1-4094 are treated as PVID by the PF and
Guest VLAN tagging is not allowed by default.
o PVID is disabled when the VLAN id is set to 0
o Guest VLAN tagging is allowed when the VLAN id is set to 4095.
o Only one Guest VLAN id is supported.
o VLAN id can be changed only when the VF driver is not loaded.
Signed-off-by: Manish Chopra <manish.chopra@qlogic.com>
Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
o Add support for MAC address and Tx rate configuration
per VF via iproute2 tool.
o Tx rate change is allowed while the guest is running
and the VF driver is loaded.
o MAC address change is allowed only when VF driver
is not loaded.
Signed-off-by: Manish Chopra <manish.chopra@qlogic.com>
Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
o Implement recovery mechanism for VF to recover from
adapter resets.
Signed-off-by: Manish Chopra <manish.chopra@qlogic.com>
Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
o FLR from Hypervisor - When hypervisor issues a VF FLR request,
adapter notifies the parent PF driver of the FLR request for PF
driver to perform any cleanup on behalf of that VF.
o FLR from VF Driver - VF driver may initiate a VF FLR request,
if VF state needs to be cleaned up before a re-initialization.
VF re-initialization during kdump is an example.
o PF driver cleans up all resources allocated on behalf of a VF,
on VF FLR notifications from the adapter or from the VF driver.
Signed-off-by: Manish Chopra <manish.chopra@qlogic.com>
Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
o 82xx adapter requires VLAN id in little endian format.
Instead of passing vlan id parameter as __le16, pass the
parameter as u16 and use cpu_to_le16 at appropriate places.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Patrick McHardy says:
====================
The following patches contain an implementation of memory mapped I/O for
netlink. The implementation is modelled after AF_PACKET memory mapped I/O
with a few differences:
- In order to perform memory mapped I/O to userspace, the kernel allocates
skbs with the data area pointing to the data area of the mapped frames.
All netlink subsystems assume a linear data area, so for the sake of
simplicity, the mapped data area is not attached to the paged area but
to skb->data. This requires introduction of a special skb alloction
function that just allocates an skb head without the data area. Since this
is a quite rare use case, I introduced a new function based on __alloc_skb
instead of splitting it up into head and data alloction. The alternative
would be to introduce an __alloc_skb_head and __alloc_skb_data function,
which would actually be useful for a specific error case in memory mapped
netlink, but would require a couple of extra instructions for the common
skb allocation case, so it doesn't really seem worth it.
In order to get the destination memory area for skb->data before message
construction, memory mapped netlink I/O needs to look up the destination
socket during allocation instead of during transmission because the
ring is owned by the receiveing socket/process. A special skb allocation
function (netlink_alloc_skb) taking the destination pid as an argument is
used for this, all subsystems that want to support memory mapped I/O need
to use this function, automatic fallback to the receive queue happens
for unconverted subsystems. Dumps automatically use memory mapped I/O if
the receiving socket has enabled it.
The visible effect of looking up the destination socket during allocation
instead of transmission is that message ordering in userspace might
change in case allocation and transmission aren't performed atomically.
This usually doesn't matter since most subsystems have a BKL-like lock
like the rtnl mutex, to my knowledge the currently only existing case
where it might matter is nfnetlink_queue combined with the recently
introduced batched verdicts, but a) that subsystem already includes
sequence numbers which allow userspace to reorder messages in case it
cares to, also the reodering window is quite small and b) with memory
mapped transmission batching can be performed in a subsystem indepandant
manner.
- AF_NETLINK contains flow control for database dumps, with regular I/O
dump continuation are triggered based on the sockets receive queue space
and by recvmsg() calls. Since with memory mapped I/O there are no
recvmsg() calls under normal operation, this is done in netlink_poll(),
under the assumption that userspace has processed all pending frames
before invoking poll(), thus the ring is expected to have room for new
messages. Dumps currently don't benefit as much as they could from
memory mapped I/O because each single continuation requires a poll()
call. A more agressive approach seems like a good idea to me, especially
in case the socket is not subscribed to any multicast groups (IOW only
receiving explicitly requested data).
Besides that, the memory mapped netlink implementation extends the states
defined by AF_PACKET between userspace and the kernel by a SKIP status, this
is intended for the case that userspace wants to queue frames (specifically
when using nfnetlink_queue, an IDS and stream reassembly, requested by
Eric Leblond) for a longer period of time. The kernel skips over all frames
marked with SKIP when looking or unused frames and only fails when not finding
a free frame or when having skipped the entire ring.
Also noteworthy is memory mapped sendmsg: the kernel performs validation
of messages before accepting and processing them, in order to prevent
userspace from changing the messages contents after validation, the
kernel checks that the ring is only mapped once and the file descriptor
is not shared (in order to avoid having userspace set up another mapping
after the first mentioned check). If either of both is not true, the
message copied to an allocated skb and processed as with regular I/O.
I'd especially appreciate review of this part since I'm not really versed
in memory, file and process management,
The remaining interesting details are included in the changelogs of the
individual patches and the documentation, so I won't repeat them here.
As an example, nfnetlink_queue is convererted to support memory mapped
I/O. Other subsystems that would probably benefit are nfnetlink_log,
audit and maybe ISCSI, not sure.
Following are some numbers collected by Florian Westphal based on a
slightly older version, which included an experimental patch for the
nfnetlink_queue ordering issue.
===
Test hardware is a 12-core machine
Intel(R) Xeon(R) CPU E5-2620 0 @ 2.00GHz
ixgbe interfaces are used (i.e., multiqueue nics).
irqs are distributed across the cpus.
I've made several tests.
The simple one consists of 3GBit UDP traffic, packets are 1500 bytes
in size (i.e., no fragmentation), with a single nfqueue
and the test client programs in libmnl examples directory.
Packets are sent from one /24 net to another /24 net, i.e.
there are a few hundred flows active at any given time.
I've also tested with snort, but I disabled all rules.
6Gbit UDP traffic is generated in the snort case, and
6 nfqueues are used (i.e., 6 snorts run in parallel).
I've tested with 3 different kernels, all based on 3.7.1.
- 3.7.1, without the mmap patches
- 3.7.1, with Patricks mmap patches
- 3.7.1, with mmap patches and extended spinlock to ensure packet ids are
monotonically increasing and cannot be re-ordered. This is what we
currently ship in our product.
[ the spinlock that is extended is the per nfqueue spinlock, it will
be held from the time the netlink skb is allocated until the netlink
skb is sent to userspace:
http://1984.lsi.us.es/git/nf-next/commit/?h=mmap-netlink3&id=b8eb19c46650fef4e9e4fe53f367f99bbf72afc9
]
snort is normally used in "batch mode", i.e., after processing 25 packets
a single "batch verdict" is sent to accept the packets seen so far.
"mmap snort" means RX_RING + sendmsg(), i.e. TX_RING is not used at this
time (except where noted below).
One reason is that snort has a reload thread, so kernel needs to copy;
also in the snort case no payload rewrite takes place, so compared
to the rx path the tx path is cheap.
Results:
3.7.1, without mmap patches, i.e. recv()+sendmsg() for everyone
nfq-queue: 1.7 gbit out
snort-recv-batch-25 5.1 gbit out
snort-recv-no-batch 3.1 gbit out
3.7.1 + mmap + without extended spinlocked section
nfq-queue: 1.7 gbit out (recv/sendmsg)
nfq-queue-mmap: 2.4 gbit out
snort-mmap-batch-25 5.6 gbit out (warning: since ids can be
re-ordered, this version is "broken").
snort-recv-batch-25 5.1 gbit out
snort-mmap-no-batch 4.6 gbit out (i.e., one verdict per packet)
Kernel 3.7.1 + mmap + extended spinlock section:
nfq-queue: 1.4 gbit out
nfq-queue-mmap: 2.3 gbit out
snort: 5.6 gbit out
Conclusions:
- The "extended spinlocked section" hurts performance in the
single queue case; with 6 snorts there is no measureable slowdown.
- I tried to re-write the mmap-snort to work without batch verdicts, but
results were not very encouraging:
kernel 3.7.1 + mmap (without extended spinlocked section):
snort-mmap-batch-25 5.6 gbit out (what we currenlty ship)
snort-recv-batch-25 5.1 gbit out (without using mmap)
snort-mmap-batch-1 4.6 gbit out (with mmap but without batch verdicts)
snort-mmap-txring-25 5.2 gbit out (with mmap but without batch verdicts)
snort-mmap-txring-1 4.6 gbit out (with mmap but without batch verdicts)
The difference between the last two is that in the txring-25 case, we
put a verdict into the tx ring after every packet, but will only
invoke sendmsg(, NULL, 0) after processing 25 packets. So the only
difference is the number of sendmsg calls/context switches.
So, i.o.w, kernel 3.7.1 + mmap + the extra locking crap is faster
than 3.7.1 + mmap-without-extra-locking and single-verdict-per packet.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Get rid of the confusing mix of pid and portid and use portid consistently
for all netlink related socket identities.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add flow control for memory mapped RX. Since user-space usually doesn't
invoke recvmsg() when using memory mapped I/O, flow control is performed
in netlink_poll(). Dumps are allowed to continue if at least half of the
ring frames are unused.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for mmap'ed recvmsg(). To allow the kernel to construct messages
into the mapped area, a dataless skb is allocated and the data pointer is
set to point into the ring frame. This means frames will be delivered to
userspace in order of allocation instead of order of transmission. This
usually doesn't matter since the order is either not determinable by
userspace or message creation/transmission is serialized. The only case
where this can have a visible difference is nfnetlink_queue. Userspace
can't assume mmap'ed messages have ordered IDs anymore and needs to check
this if using batched verdicts.
For non-mapped sockets, nothing changes.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for mmap'ed sendmsg() to netlink. Since the kernel validates
received messages before processing them, the code makes sure userspace
can't modify the message contents after invoking sendmsg(). To do that
only a single mapping of the TX ring is allowed to exist and the socket
must not be shared. If either of these two conditions does not hold, it
falls back to copying.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add helper functions for looking up mmap'ed frame headers, reading and
writing their status, allocating skbs with mmap'ed data areas and a poll
function.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for mmap'ed RX and TX ring setup and teardown based on the
af_packet.c code. The following patches will use this to add the real
mmap'ed receive and transmit functionality.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
For mmap'ed I/O a netlink specific skb destructor needs to be invoked
after the final kfree_skb() to clean up state. This doesn't work currently
since the skb's ownership is transfered to the receiving socket using
skb_set_owner_r(), which orphans the skb, thereby invoking the destructor
prematurely.
Since netlink doesn't account skbs to the originating socket, there's no
need to orphan the skb. Add a netlink specific skb_set_owner_r() variant
that does not orphan the skb and use a netlink specific destructor to
call sock_rfree().
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Netlink doesn't account skbs to the sending socket, so the there's no
need to orphan the skb before trimming it.
Removing the skb_orphan() call is required for mmap'ed netlink, which uses
a netlink specific skb destructor that must not be invoked before the
final freeing of the skb.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a function to allocate a sk_buff head without any data. This will
be used by memory mapped netlink to attach data from the mmaped area
to the skb.
Additionally change skb_release_all() to check whether the skb has a
data area to allow the skb destructor to clear the data pointer in case
only a head has been allocated.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Memory mapped netlink needs to store the receiving userspace socket
when sending from the kernel to userspace. Rename 'ssk' to 'sk' to
avoid confusion.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>