Flow statistics need to take into account the TCP flags from the packet
currently being processed (in 'key'), not the TCP flags matched by the
flow found in the kernel flow table (in 'flow').
This bug made the Open vSwitch userspace fin_timeout action have no effect
in many cases.
This bug is introduced by commit 88d73f6c41 (openvswitch: Use
TCP flags in the flow key for stats.)
Reported-by: Len Gao <leng@vmware.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
For ovs_flow_stats_get() using ovsl_dereference() was wrong, since
flow dumps call this with RCU read lock.
ovs_flow_stats_clear() is always called with ovs_mutex, so can use
ovsl_dereference().
Also, make the ovs_flow_stats_get() 'flow' argument const to make
later patches cleaner.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Remove unnecessary locking from functions that are always called with
appropriate locking.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Thomas Graf <tgraf@redhat.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Minimize padding in sw_flow_key and move 'tp' top the main struct.
These changes simplify code when accessing the transport port numbers
and the tcp flags, and makes the sw_flow_key 8 bytes smaller on 64-bit
systems (128->120 bytes). These changes also make the keys for IPv4
packets to fit in one cache line.
There is a valid concern for safety of packing the struct
ovs_key_ipv4_tunnel, as it would be possible to take the address of
the tun_id member as a __be64 * which could result in unaligned access
in some systems. However:
- sw_flow_key itself is 64-bit aligned, so the tun_id within is
always
64-bit aligned.
- We never make arrays of ovs_key_ipv4_tunnel (which would force
every
second tun_key to be misaligned).
- We never take the address of the tun_id in to a __be64 *.
- Whereever we use struct ovs_key_ipv4_tunnel outside the
sw_flow_key,
it is in stack (on tunnel input functions), where compiler has full
control of the alignment.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
We already extract the TCP flags for the key, might as well use that
for stats.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Keep kernel flow stats for each NUMA node rather than each (logical)
CPU. This avoids using the per-CPU allocator and removes most of the
kernel-side OVS locking overhead otherwise on the top of perf reports
and allows OVS to scale better with higher number of threads.
With 9 handlers and 4 revalidators netperf TCP_CRR test flow setup
rate doubles on a server with two hyper-threaded physical CPUs (16
logical cores each) compared to the current OVS master. Tested with
non-trivial flow table with a TCP port match rule forcing all new
connections with unique port numbers to OVS userspace. The IP
addresses are still wildcarded, so the kernel flows are not considered
as exact match 5-tuple flows. This type of flows can be expected to
appear in large numbers as the result of more effective wildcarding
made possible by improvements in OVS userspace flow classifier.
Perf results for this test (master):
Events: 305K cycles
+ 8.43% ovs-vswitchd [kernel.kallsyms] [k] mutex_spin_on_owner
+ 5.64% ovs-vswitchd [kernel.kallsyms] [k] __ticket_spin_lock
+ 4.75% ovs-vswitchd ovs-vswitchd [.] find_match_wc
+ 3.32% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_lock
+ 2.61% ovs-vswitchd [kernel.kallsyms] [k] pcpu_alloc_area
+ 2.19% ovs-vswitchd ovs-vswitchd [.] flow_hash_in_minimask_range
+ 2.03% swapper [kernel.kallsyms] [k] intel_idle
+ 1.84% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_unlock
+ 1.64% ovs-vswitchd ovs-vswitchd [.] classifier_lookup
+ 1.58% ovs-vswitchd libc-2.15.so [.] 0x7f4e6
+ 1.07% ovs-vswitchd [kernel.kallsyms] [k] memset
+ 1.03% netperf [kernel.kallsyms] [k] __ticket_spin_lock
+ 0.92% swapper [kernel.kallsyms] [k] __ticket_spin_lock
...
And after this patch:
Events: 356K cycles
+ 6.85% ovs-vswitchd ovs-vswitchd [.] find_match_wc
+ 4.63% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_lock
+ 3.06% ovs-vswitchd [kernel.kallsyms] [k] __ticket_spin_lock
+ 2.81% ovs-vswitchd ovs-vswitchd [.] flow_hash_in_minimask_range
+ 2.51% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_unlock
+ 2.27% ovs-vswitchd ovs-vswitchd [.] classifier_lookup
+ 1.84% ovs-vswitchd libc-2.15.so [.] 0x15d30f
+ 1.74% ovs-vswitchd [kernel.kallsyms] [k] mutex_spin_on_owner
+ 1.47% swapper [kernel.kallsyms] [k] intel_idle
+ 1.34% ovs-vswitchd ovs-vswitchd [.] flow_hash_in_minimask
+ 1.33% ovs-vswitchd ovs-vswitchd [.] rule_actions_unref
+ 1.16% ovs-vswitchd ovs-vswitchd [.] hindex_node_with_hash
+ 1.16% ovs-vswitchd ovs-vswitchd [.] do_xlate_actions
+ 1.09% ovs-vswitchd ovs-vswitchd [.] ofproto_rule_ref
+ 1.01% netperf [kernel.kallsyms] [k] __ticket_spin_lock
...
There is a small increase in kernel spinlock overhead due to the same
spinlock being shared between multiple cores of the same physical CPU,
but that is barely visible in the netperf TCP_CRR test performance
(maybe ~1% performance drop, hard to tell exactly due to variance in
the test results), when testing for kernel module throughput (with no
userspace activity, handful of kernel flows).
On flow setup, a single stats instance is allocated (for the NUMA node
0). As CPUs from multiple NUMA nodes start updating stats, new
NUMA-node specific stats instances are allocated. This allocation on
the packet processing code path is made to never block or look for
emergency memory pools, minimizing the allocation latency. If the
allocation fails, the existing preallocated stats instance is used.
Also, if only CPUs from one NUMA-node are updating the preallocated
stats instance, no additional stats instances are allocated. This
eliminates the need to pre-allocate stats instances that will not be
used, also relieving the stats reader from the burden of reading stats
that are never used.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
The 5-tuple optimization becomes unnecessary with a later per-NUMA
node stats patch. Remove it first to make the changes easier to
grasp.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
The kernel starts out its "jiffies" timer as 5 minutes below zero, as
shown in include/linux/jiffies.h:
/*
* Have the 32 bit jiffies value wrap 5 minutes after boot
* so jiffies wrap bugs show up earlier.
*/
#define INITIAL_JIFFIES ((unsigned long)(unsigned int) (-300*HZ))
The loop in ovs_flow_stats_get() starts out with 'used' set to 0, then
takes any "later" time. This means that for the first five minutes after
boot, flows will always be reported as never used, since 0 is greater than
any time already seen.
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Only the first IP fragment can have a TCP header, check for this.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
With mega flow implementation ovs flow can be shared between
multiple CPUs which makes stats updates highly contended
operation. This patch uses per-CPU stats in cases where a flow
is likely to be shared (if there is a wildcard in the 5-tuple
and therefore likely to be spread by RSS). In other situations,
it uses the current strategy, saving memory and allocation time.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
tcp_flags=flags/mask
Bitwise match on TCP flags. The flags and mask are 16-bit num‐
bers written in decimal or in hexadecimal prefixed by 0x. Each
1-bit in mask requires that the corresponding bit in port must
match. Each 0-bit in mask causes the corresponding bit to be
ignored.
TCP protocol currently defines 9 flag bits, and additional 3
bits are reserved (must be transmitted as zero), see RFCs 793,
3168, and 3540. The flag bits are, numbering from the least
significant bit:
0: FIN No more data from sender.
1: SYN Synchronize sequence numbers.
2: RST Reset the connection.
3: PSH Push function.
4: ACK Acknowledgement field significant.
5: URG Urgent pointer field significant.
6: ECE ECN Echo.
7: CWR Congestion Windows Reduced.
8: NS Nonce Sum.
9-11: Reserved.
12-15: Not matchable, must be zero.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Widen TCP flags handling from 7 bits (uint8_t) to 12 bits (uint16_t).
The kernel interface remains at 8 bits, which makes no functional
difference now, as none of the higher bits is currently of interest
to the userspace.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Over the time datapath.c and flow.c has became pretty large files.
Following patch restructures functionality of component into three
different components:
flow.c: contains flow extract.
flow_netlink.c: netlink flow api.
flow_table.c: flow table api.
This patch restructures code without changing logic.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
In function __parse_flow_nlattrs(), we check for condition
(type > OVS_KEY_ATTR_MAX) and if true, print an error, but we do
not return from this function as in other checks. It seems this
has been forgotten, as otherwise, we could access beyond the
memory of ovs_key_lens, which is of ovs_key_lens[OVS_KEY_ATTR_MAX + 1].
Hence, a maliciously prepared nla_type from user space could access
beyond this upper limit.
Introduced by 03f0d916a ("openvswitch: Mega flow implementation").
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Cc: Andy Zhou <azhou@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sw_flow_key alignment was declared as " __aligned(__alignof__(long))".
However, this breaks on the m68k architecture where long is 32 bit in
size but 16 bit aligned by default. This aligns to the size of a long to
ensure that we can always do comparsions in full long-sized chunks. It
also adds an additional build check to catch any reduction in alignment.
CC: Andy Zhou <azhou@nicira.com>
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make sure the sw_flow_key structure and valid mask boundaries are always
machine word aligned. Optimize the flow compare and mask operations
using machine word size operations. This patch improves throughput on
average by 15% when CPU is the bottleneck of forwarding packets.
This patch is inspired by ideas and code from a patch submitted by Peter
Klausler titled "replace memcmp() with specialized comparator".
However, The original patch only optimizes for architectures
support unaligned machine word access. This patch optimizes for all
architectures.
Signed-off-by: Andy Zhou <azhou@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Key_end is a better name describing the ending boundary than key_len.
Rename those variables to make it less confusing.
Signed-off-by: Andy Zhou <azhou@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
This patch adds support for rewriting SCTP src,dst ports similar to the
functionality already available for TCP/UDP.
Rewriting SCTP ports is expensive due to double-recalculation of the
SCTP checksums; this is performed to ensure that packets traversing OVS
with invalid checksums will continue to the destination with any
checksum corruption intact.
Reviewed-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Joe Stringer <joe@wand.net.nz>
Signed-off-by: Ben Pfaff <blp@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Add wildcarded flow support in kernel datapath.
Wildcarded flow can improve OVS flow set up performance by avoid sending
matching new flows to the user space program. The exact performance boost
will largely dependent on wildcarded flow hit rate.
In case all new flows hits wildcard flows, the flow set up rate is
within 5% of that of linux bridge module.
Pravin has made significant contributions to this patch. Including API
clean ups and bug fixes.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Andy Zhou <azhou@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Flow table destroy is done in rcu call-back context. Therefore
there is no need to use rcu variant of hlist_del().
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Flex array is used to allocate hash buckets which is type struct
hlist_head, but we use `struct hlist_head *` to calculate
array size. Since hlist_head is of size pointer it works fine.
Following patch use correct type.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Following patch adds start offset for sw_flow-key, so that we can
skip tunneling information in key for non-tunnel flows.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ben Pfaff
Jarno Rajahalme
Joe Perches
Flavio Leitner
Pravin B Shelar
Daniel Borkmann
Jesse Gross
Andy Zhou
Joe Stringer