cascardo/ovs.git
10 years agoofproto-dpif-monitor: Acquire write lock in monitor_run().
Alex Wang [Fri, 13 Dec 2013 19:29:09 +0000 (11:29 -0800)]
ofproto-dpif-monitor: Acquire write lock in monitor_run().

Commit 307464a1 (ofproto-dpif-monitor: Use heap to order the mport
wakeup time.) re-heapifies the heap in monitor_run().  So the
monitor_run() should be protected by the write lock, rather than
the read lock.

This commit fixes the issue.

Signed-off-by: Alex Wang <alexw@nicira.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agodpif-linux: Do not call nl_sock_pid() on NULL sock pointer.
Alex Wang [Tue, 17 Dec 2013 22:37:10 +0000 (14:37 -0800)]
dpif-linux: Do not call nl_sock_pid() on NULL sock pointer.

This commit adds check of sock pointer in dpif_linux_port_get_pid().
If the pointer is NULL, do not call nl_sock_pid().

Signed-off-by: Alex Wang <alexw@nicira.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agodpif-linux: Fix the return type of dpif_linux_port_dump_next__().
Alex Wang [Tue, 17 Dec 2013 22:37:09 +0000 (14:37 -0800)]
dpif-linux: Fix the return type of dpif_linux_port_dump_next__().

Commit 222837 (dpif-linux: Factor out port dumping helper functions.)
introduced a bug by making dpif_linux_port_dump_next__() return 'bool'
instead of 'int' as defined in dpif-provider.h.  This bug causes ovs-
vswitchd failure with SEGFAULT when processing slow-path packet.

This commit fixes the bug by following the dpif-provider specification.

Signed-off-by: Alex Wang <alexw@nicira.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoRemove stream, vconn, and rconn functions to get local/remote IPs/ports.
Ben Pfaff [Tue, 17 Dec 2013 23:07:12 +0000 (15:07 -0800)]
Remove stream, vconn, and rconn functions to get local/remote IPs/ports.

These functions don't have any ultimate users.  The in-band control code
used to use them, but not anymore, so we might as well delete them all.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
10 years agodatapath: make functions local
Stephen Hemminger [Tue, 17 Dec 2013 22:57:46 +0000 (14:57 -0800)]
datapath: make functions local

Several functions and datastructures could be local
Found with 'make namespacecheck'

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agobfd: Set next_tx correctly when processing packets
Joe Stringer [Fri, 6 Dec 2013 20:22:07 +0000 (12:22 -0800)]
bfd: Set next_tx correctly when processing packets

In the case where we have not yet sent a control packet for a bfd
connection, and we receive a control packet from the remote host,
bfd->next_tx is updated to an unusual value. This causes the logging to
incorrectly report that there has been long delays (in the order of
weeks) since the last bfd transmission time.

This patch only modifies bfd->next_tx in this case if we are not
expecting to immediately send a control packet. This should mean that
bfd->next_tx is either 0 (immediate tx) or in the order of time_msec().

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agolacp: Give LACP a moment to initialize before testing its state, in tests.
Ben Pfaff [Tue, 17 Dec 2013 22:14:52 +0000 (14:14 -0800)]
lacp: Give LACP a moment to initialize before testing its state, in tests.

These tests configured LACP and then immediately dumped out its state.
Most of the time, this worked, but there was a brief race window in which
the "negotiated" flag could be missing because this took one pass through
the main loop.  This fixes the problem.

This race may be seen in the failures of tests 11 and 12 here:
https://launchpadlibrarian.net/151884888/buildlog_ubuntu-precise-amd64.openvswitch_2.0~201309300804-1ppa1~precise_FAILEDTOBUILD.txt.gz

Reported-by: Vasiliy Tolstov <v.tolstov@selfip.ru>
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Andy Zhou <azhou@nicira.com>
10 years agobridge: Let ofprotos run once before reporting configuration complete.
Ben Pfaff [Mon, 30 Sep 2013 20:07:35 +0000 (13:07 -0700)]
bridge: Let ofprotos run once before reporting configuration complete.

Occasionally in the unit tests the following race can happen:

   1. ovs-vsctl updates database
   2. ovs-vswitchd reconfigures, notifies ovs-vsctl that it is complete
   3. ovs-appctl ofproto/trace fails to see newly added port
   4. ovs-vswitchd main loop calls ofproto's ->type_run(), making the
      new port visible to translation.

This race may be seen in the failures of tests 5 and 624 here:
https://launchpadlibrarian.net/151884888/buildlog_ubuntu-precise-amd64.openvswitch_2.0~201309300804-1ppa1~precise_FAILEDTOBUILD.txt.gz

Reported-by: Vasiliy Tolstov <v.tolstov@selfip.ru>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agodpif-linux: fix the size of n_masks
Francesco Fusco [Tue, 17 Dec 2013 19:18:18 +0000 (20:18 +0100)]
dpif-linux: fix the size of n_masks

The command ovs-dpctl can wrongly output the masks even if the
datapath does not implement mega flows. In this case the output
will be similar to the following:

system@ovs-system:
lookups: hit:14 missed:41 lost:0
flows: 0
masks: hit:18446744073709551615 total:4294967295
hit/pkt:335395346794719104.00
port 0: ovs-system (internal)
port 1: gre_system (gre: df_default=false, ttl=0)
port 2: ots-br0 (internal)
port 3: int0 (internal)
port 4: vnet0
port 5: vnet1

The problem depends on the fact that n_masks stats is stored as a
uint32 in the struct ovs_dp_megaflow_stats and as a uint64 in the
struct dpif_dp_stats. UINT32_MAX instead of UINT64_MAX should be
used to detect if the datapath supports megaflows or not.

Signed-off-by: Francesco Fusco <ffusco@redhat.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoRename NOT_REACHED to OVS_NOT_REACHED
Harold Lim [Tue, 17 Dec 2013 18:32:12 +0000 (10:32 -0800)]
Rename NOT_REACHED to OVS_NOT_REACHED

This allows other libraries to use util.h that has already
defined NOT_REACHED.

Signed-off-by: Harold Lim <haroldl@vmware.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoUpdate openvswitch to allow linking from C++ projects
Harold Lim [Tue, 17 Dec 2013 18:32:11 +0000 (10:32 -0800)]
Update openvswitch to allow linking from C++ projects

The input variable of ovs_scan is changed from 'template' to
'format'. template is a keyword in C++.

Signed-off-by: Harold Lim <haroldl@vmware.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agovtep: add "Arp_sources" tables
Bruce Davie [Mon, 25 Nov 2013 16:19:50 +0000 (08:19 -0800)]
vtep: add "Arp_sources" tables

Add two new tables to the VTEP schema in support of distributed L3.
Each table contains MAC addresses to be used by VTEPs (both hardware
and software) when issuing ARP requests on behalf of a logical router.

Signed-off-by: Bruce Davie <bdavie@vmware.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoFAQ: Reference 2.0, not 1.12.
Justin Pettit [Tue, 17 Dec 2013 20:50:50 +0000 (12:50 -0800)]
FAQ: Reference 2.0, not 1.12.

Signed-off-by: Justin Pettit <jpettit@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoFAQ: Update to reflect that tunneling is now in upstream Linux.
Jesse Gross [Tue, 17 Dec 2013 18:49:26 +0000 (10:49 -0800)]
FAQ: Update to reflect that tunneling is now in upstream Linux.

Reported-by: Ben Pfaff <blp@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: pritesh <pritesh.kothari@cisco.com>
10 years agodpif-linux: Fix a bug.
Alex Wang [Tue, 17 Dec 2013 16:16:24 +0000 (16:16 +0000)]
dpif-linux: Fix a bug.

Commit da546e0 (dpif: Allow execute to modify the packet.) introduced
a bug by subtracting the zero-value ofpbuf size by "sizeof(struct
nlattr)" and assigning the result back to the ofpbuf size.  This bug
causes the ovs-assert failure in facet_push_stats().

This commit fixes the bug by assigning the right value to the ofpbuf
size.

Signed-off-by: Alex Wang <alexw@nicira.com>
Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
10 years agorconn: Update comments on is_admitted_msg().
Ben Pfaff [Thu, 21 Nov 2013 23:03:23 +0000 (15:03 -0800)]
rconn: Update comments on is_admitted_msg().

Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoFAQ: Explain how to add QoS features to Open vSwitch.
Ben Pfaff [Fri, 22 Nov 2013 01:03:13 +0000 (17:03 -0800)]
FAQ: Explain how to add QoS features to Open vSwitch.

Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoFAQ: Describe weak and strong ES models.
Ben Pfaff [Tue, 17 Dec 2013 06:19:08 +0000 (22:19 -0800)]
FAQ: Describe weak and strong ES models.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Andy Zhou <azhou@nicira.com>
10 years agodatapath: Handle different definitions of page in skb_frag_t.
Jesse Gross [Tue, 17 Dec 2013 05:03:09 +0000 (21:03 -0800)]
datapath: Handle different definitions of page in skb_frag_t.

Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: Add hash library to .gitignore.
Jesse Gross [Tue, 17 Dec 2013 02:10:40 +0000 (18:10 -0800)]
datapath: Add hash library to .gitignore.

Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: Backport __skb_fill_page_desc
Jesse Gross [Tue, 17 Dec 2013 02:07:16 +0000 (18:07 -0800)]
datapath: Backport __skb_fill_page_desc

Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: Backport skb_has_frag_list().
Jesse Gross [Tue, 17 Dec 2013 02:05:01 +0000 (18:05 -0800)]
datapath: Backport skb_has_frag_list().

Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: Backport skb_frag_ functions
Pravin B Shelar [Sun, 3 Mar 2013 07:53:52 +0000 (23:53 -0800)]
datapath: Backport skb_frag_ functions

Define accessors skb_frag_* functions whch were introduced in 3.2

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
10 years agodatapath: Add missing #include for skb page accessors.
Jesse Gross [Tue, 31 Jan 2012 21:15:30 +0000 (13:15 -0800)]
datapath: Add missing #include for skb page accessors.

Reported-by: Ben Pfaff <blp@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: Compute checksum in skb_gso_segment() if needed
Thomas Graf [Tue, 17 Dec 2013 01:03:45 +0000 (17:03 -0800)]
datapath: Compute checksum in skb_gso_segment() if needed

The copy & csum optimization is no longer present with zerocopy
enabled. Compute the checksum in skb_gso_segment() directly by
dropping the HW CSUM capability from the features passed in.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: Use skb_zerocopy() for upcall
Thomas Graf [Tue, 17 Dec 2013 00:56:03 +0000 (16:56 -0800)]
datapath: Use skb_zerocopy() for upcall

Use of skb_zerocopy() can avoid the expensive call to memcpy()
when copying the packet data into the Netlink skb. Completes
checksum through skb_checksum_help() if not already done in
GSO segmentation.

Zerocopy is only performed if user space supported unaligned
Netlink messages. memory mapped netlink i/o is preferred over
zerocopy if it is set up.

Cost of upcall is significantly reduced from:
+   7.48%       vhost-8471  [k] memcpy
+   5.57%     ovs-vswitchd  [k] memcpy
+   2.81%       vhost-8471  [k] csum_partial_copy_generic

to:
+   5.72%     ovs-vswitchd  [k] memcpy
+   3.32%       vhost-5153  [k] memcpy
+   0.68%       vhost-5153  [k] skb_zerocopy

(megaflows disabled)

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: Pass datapath into userspace queue functions
Thomas Graf [Tue, 17 Dec 2013 00:51:07 +0000 (16:51 -0800)]
datapath: Pass datapath into userspace queue functions

Allows removing the net and dp_ifindex argument and simplify the
code.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: Drop user features if old user space attempted to create datapath
Thomas Graf [Wed, 4 Dec 2013 04:56:22 +0000 (20:56 -0800)]
datapath: Drop user features if old user space attempted to create datapath

Drop user features if an outdated user space instance that does not
understand the concept of user_features attempted to create a new
datapath.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Reviewed-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: Allow user space to announce ability to accept unaligned Netlink messages
Thomas Graf [Wed, 4 Dec 2013 04:47:24 +0000 (20:47 -0800)]
datapath: Allow user space to announce ability to accept unaligned Netlink messages

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Reviewed-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: Backport skb_zerocopy() functions.
Jesse Gross [Wed, 4 Dec 2013 04:41:15 +0000 (20:41 -0800)]
datapath: Backport skb_zerocopy() functions.

These functions will be factored out and exported upstream. On
kernels 3.5 and newer the backport will provide zero copy support
but older kernels will work as before (due to lack of skb->head_frag).

Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agoofproto/trace: Fix memory leak on error path.
Jarno Rajahalme [Mon, 16 Dec 2013 23:03:38 +0000 (15:03 -0800)]
ofproto/trace: Fix memory leak on error path.

Propagate error response from parse_ofp_exact_flow() to the caller and
properly free() it afterward.

Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agodatapath: use CRC32 accelerated flow hash if available
Francesco Fusco [Mon, 16 Dec 2013 09:22:43 +0000 (01:22 -0800)]
datapath: use CRC32 accelerated flow hash if available

Currently OVS uses jhash2() for calculating flow hashes in its
internal flow_hash() function. The performance of the flow_hash()
function is critical, as the input data can be hundreds of bytes
long.

OVS is largely deployed in x86_64 based datacenters.  Therefore,
we argue that the performance critical fast path of OVS should
exploit underlying CPU features in order to reduce the per packet
processing costs. We replace jhash2 with the hash implementation
provided by the kernel hash lib, which exploits the crc32l
instruction to achieve high performance

Our patch greatly reduces the hash footprint from ~200 cycles of
jhash2() to around ~90 cycles in case of ovs_flow_hash_crc()
(measured with rdtsc over maximum length flow keys on an i7 Intel
CPU).

Additionally, we wrote a microbenchmark to stress the flow table
performance. The benchmark inserts random flows into the flow
hash and then performs lookups. Our hash deployed on a CRC32
capable CPU reduces the lookup for 1000 flows, 100 masks from
~10,100us to ~6,700us, for example.

Thus, simply use the newly introduced arch_fast_hash2() as a
drop-in replacement.

Signed-off-by: Francesco Fusco <ffusco@redhat.com>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: Thomas Graf <tgraf@redhat.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: Backport for architecture hash library.
Jesse Gross [Mon, 16 Dec 2013 08:44:05 +0000 (00:44 -0800)]
datapath: Backport for architecture hash library.

The architecture-specific hash library automatically selects either
jhash or CRC32 if it is available on the current processor.

Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agodatapath: remove duplicated include from flow_table.c
Wei Yongjun [Mon, 16 Dec 2013 08:46:04 +0000 (00:46 -0800)]
datapath: remove duplicated include from flow_table.c

Remove duplicated include.

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Jesse Gross <jesse@nicira.com>
10 years agoconnmgr: Fix attempt to take mutex recursively when exiting fail-open.
Ben Pfaff [Mon, 16 Dec 2013 18:58:29 +0000 (10:58 -0800)]
connmgr: Fix attempt to take mutex recursively when exiting fail-open.

If one configured a controller which does not exist, waited for the switch
to enter fail-open mode, and then deleted the controller, then
ofproto_set_controllers() would take ofproto_mutex and call
update_fail_open(), which would call fail_open_destroy(), which would call
fail_open_recover(), which would call ofproto_delete_flow(), which requires
ofproto_mutex not to be held since it eventually try to take it.  This
caused OVS to abort.

This fixes the problem by releasing ofproto_mutex earlier, since nothing
seems to require it being held so long (a comment in
connmgr_set_controllers() says that this is likely to be the case).

Better annotations would have found this problem at compile time.  A later
patch adds them.

Reported-by: ZhengLingyun <konghuarukhr@163.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
10 years agobridge: Remove superfluous "inline" from ofp12_controller_role_to_str().
Ben Pfaff [Mon, 16 Dec 2013 17:20:12 +0000 (09:20 -0800)]
bridge: Remove superfluous "inline" from ofp12_controller_role_to_str().

Putting "static inline" on a function definition in a .c file does not help
the compiler and does suppress warnings for unused functions that one would
want, should the function ever become unused.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Joe Stringer <joestringer@nicira.com>
10 years agodpif: Allow execute to modify the packet.
Jarno Rajahalme [Mon, 16 Dec 2013 16:14:52 +0000 (08:14 -0800)]
dpif: Allow execute to modify the packet.

Allowing the packet to be modified by execution allows less data
copying for userspace action execution.  Some users of the
dpif_execute already expect that the packet may be modified.  This
patch makes this behavior uniform and makes the userspace datapath and
the execution helpers modify the packet as it is being executed.
Userspace action now steals the packet if given permission, as the
packet is normally not needed after it.  The only exception is the
sample action, and this is accounted for my keeping track of any
actions that could be following the userspace action.

The packet in dpif_upcall is changed from a pointer to a struct,
allowing the packet to be honest about it's headroom.  After this
change the packet can safely be pushed on over the precarious 4 byte
limit earlier allowed by the netlink data preceding the packet.

Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agomonitor: Replace monitor_seq with periodic wakeup
Joe Stringer [Sat, 14 Dec 2013 01:24:10 +0000 (17:24 -0800)]
monitor: Replace monitor_seq with periodic wakeup

Currently, every time a monitoring port is added or reconfigured, the
main thread notifies the monitoring thread to wake up immediately using
monitor_seq. When adding a large number of ports at once, this causes
contention as the threads fight over access to the monitor heap---one
thread adding new ports, the other popping and processing the new ports.

This patch removes this mechanism in favour of a simple periodic wakeup
in the monitor thread. When configuring batches of 500 tunnels at once,
up to a total of 5000 tunnels, this reduces configuration time by up to
35%. New port monitor port configuration may not take effect for as long
as ~100ms, although it may be less if the timer has not recently timed
out, or if a monitoring port is due to send a packet.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
10 years agofedora-build: Remove %{build_number} from the configure line
Kyle Mestery [Sat, 14 Dec 2013 02:53:25 +0000 (21:53 -0500)]
fedora-build: Remove %{build_number} from the configure line

Fixes the OVS builds on Fedora, which are currently broken upstream.

Signed-off-by: Kyle Mestery <kmestery@cisco.com>
Signed-off-by: Gurucharan Shetty <gshetty@nicira.com>
10 years agoovs-thread: Fix crash by making count_cpu_count() return type a signed int.
Ben Pfaff [Fri, 13 Dec 2013 22:31:27 +0000 (14:31 -0800)]
ovs-thread: Fix crash by making count_cpu_count() return type a signed int.

ofproto_set_threads() uses the calculation MAX(count_cpu_cores() - 2, 1)
to decide on the default thread count.  However, count_cpu_cores() returns
0 if it can't count the number of cores, or 1 if there's only one core,
and that causes the calculation to come out as UINT_MAX-2 or UINT_MAX-1,
respectively, which causes a memory allocation failure later.

There are other ways to fix this problem, too, of course.

Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agovconn: Reply with OFPBRC_BAD_VERSION for bad version.
Ben Pfaff [Thu, 7 Nov 2013 17:39:03 +0000 (09:39 -0800)]
vconn: Reply with OFPBRC_BAD_VERSION for bad version.

Testers keep wanting to know why this doesn't work.  I think it's a silly
test, but it's easy enough to make them happy.

Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agodpif-linux: Fix fd leak for vports that disappear from the datapath.
Ben Pfaff [Thu, 2 May 2013 00:13:14 +0000 (17:13 -0700)]
dpif-linux: Fix fd leak for vports that disappear from the datapath.

When ovs-vswitchd deletes a port with dpif_linux_port_del(), that function
uses del_channel() to delete the corresponding channel, including closing
its Netlink socket fd.  However, if the vport gets removed by some other
process (e.g. "ip link delete" for veths) then this function never gets
called and thus the channel never gets deleted.

This commit fixes the problem by resynchronizing channels with the
datapath whenever a vport gets deleted.

Bug #16784.
Reported-by: Paul Ingram <paul@nicira.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agodpif-linux: Generalize dpif_linux_recv_set() to synchronize channels.
Ben Pfaff [Wed, 4 Dec 2013 21:37:31 +0000 (13:37 -0800)]
dpif-linux: Generalize dpif_linux_recv_set() to synchronize channels.

Until now, the code in dpif_linux_recv_set() has assumed it is starting
from a clean slate with no channels at all.  This commit generalizes it to
compare the existing set of channels against the ones that should exist
given the vports that the kernel has, adding any new ones that are missing
and removing any that should not exist.  A followup commit will make good
use of this functionality.

Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agodpif-linux: Factor out port dumping helper functions.
Ben Pfaff [Wed, 1 May 2013 21:40:38 +0000 (14:40 -0700)]
dpif-linux: Factor out port dumping helper functions.

These helpers simplify new code to be added in an upcoming commit.

Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agojsonrpc-server: Add test for disconnecting connections with too long queues.
Ben Pfaff [Wed, 27 Mar 2013 21:39:05 +0000 (14:39 -0700)]
jsonrpc-server: Add test for disconnecting connections with too long queues.

Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoovsdb-client: Make "monitor" command able to monitor all tables.
Ben Pfaff [Wed, 27 Mar 2013 21:26:21 +0000 (14:26 -0700)]
ovsdb-client: Make "monitor" command able to monitor all tables.

Users may find this feature useful, but the reason to add this feature is
to allow a test to use it in an upcoming commit.

Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agocfm: Refine the cfm_extended manual.
Alex Wang [Sat, 26 Oct 2013 00:18:02 +0000 (17:18 -0700)]
cfm: Refine the cfm_extended manual.

Commit 348f01e3e3 (cfm: Eight byte MPIDs in extended mode.) allows
eight byte MPIDs when running in extended mode.  This commit explains
this change in the vswitch.conf.db.

Signed-off-by: Alex Wang <alexw@nicira.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoovs-thread: Break recursion for coverage counters.
Gurucharan Shetty [Wed, 11 Dec 2013 23:55:03 +0000 (15:55 -0800)]
ovs-thread: Break recursion for coverage counters.

For systems that do not have either HAVE_THREAD_LOCAL or
HAVE___THREAD (ex: windows using MSVC), a COVERAGE_INC() calls xmalloc
which inturn calls COVERAGE_INC() creating a recursion that causes a
stack overflow.

This commit breaks the recursion by calling malloc() instead of
xmalloc().

Signed-off-by: Gurucharan Shetty <gshetty@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agocfm: Notify connectivity_seq on cfm_set_fault
Joe Stringer [Sat, 14 Dec 2013 02:18:13 +0000 (02:18 +0000)]
cfm: Notify connectivity_seq on cfm_set_fault

Commit f23d157c ("ofproto-dpif: Don't poll ports when nothing changes"
changed how cfm information is propagated to ovsdb, but did not ensure
that this happened when users made "appctl cfm/set_fault" calls. This
patch fixes this case.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
10 years agocoverage: Use size_t for coverage counter array lengths.
Ethan Jackson [Fri, 13 Dec 2013 17:57:38 +0000 (09:57 -0800)]
coverage:  Use size_t for coverage counter array lengths.

Fixes the following warning on my system.

../lib/coverage.c:54:40: error: incompatible pointer types passing
'unsigned int *' to parameter of type 'size_t *' (aka 'unsigned long
*')

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agostring: Use workaround for #include_next when it is not available.
Ben Pfaff [Fri, 13 Dec 2013 17:54:55 +0000 (09:54 -0800)]
string: Use workaround for #include_next when it is not available.

Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoconfigure: Add macros to check for #include_next and add a workaround.
Ben Pfaff [Fri, 13 Dec 2013 17:46:44 +0000 (09:46 -0800)]
configure: Add macros to check for #include_next and add a workaround.

This will be used for the #include_next in string.h in the following
commit.

Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto: Simplify thread creation API.
Ethan Jackson [Fri, 8 Nov 2013 21:47:49 +0000 (13:47 -0800)]
ofproto: Simplify thread creation API.

There's no particular reason for the function controlling the number
of threads to be bound up with dpif_recv_set().  This patch breaks
them up, but as a side effect means threads will run doing nothing
when datapath upcall receiving is disabled.  By doing this, the udpif
thread creation API becomes a bit easier to reason about once there
are multiple types of thread introduced in future patches.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoovs-ofctl: Replace TODO with XXX.
Ethan Jackson [Sat, 7 Dec 2013 23:07:01 +0000 (15:07 -0800)]
ovs-ofctl: Replace TODO with XXX.

In accordance with the style guidelines.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoflow: New function flow_unwildcard_tp_ports().
Ethan Jackson [Sat, 7 Dec 2013 02:53:12 +0000 (18:53 -0800)]
flow: New function flow_unwildcard_tp_ports().

This patch adds a new function flow_unildcard_tp_ports() which doesn't
unwildcard the upper half of tp_src and tp_dst with ICMP packets.
Unfortunately, this matters in future patches when we compare masks
carefully to determine if flows should be evicted from the datapath.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoRemove unused USE_LINKER_SECTIONS and OVS_LINK2_IFELSE
Helmut Schaa [Fri, 13 Dec 2013 13:05:03 +0000 (14:05 +0100)]
Remove unused USE_LINKER_SECTIONS and OVS_LINK2_IFELSE

They are unused and can be removed.

Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agocoverage: Use OVS_CONSTRUCTOR to initialize the coverage counter array
Helmut Schaa [Fri, 13 Dec 2013 13:05:02 +0000 (14:05 +0100)]
coverage: Use OVS_CONSTRUCTOR to initialize the coverage counter array

Use a static array in coverage.c that gets initialized by constructor
functions per coverage definition.

Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agovlog: Use OVS_CONSTRUCTOR for vlog initialization
Helmut Schaa [Fri, 13 Dec 2013 13:05:01 +0000 (14:05 +0100)]
vlog: Use OVS_CONSTRUCTOR for vlog initialization

This allows to get rid of some special segment handling to allow
distributed registering of vlog modules.

Instead use a global list and vlog module constructor functions to
build up the list. That means vlog modules reside within the
compilation unit they are defined in and can be iterated upon
by using the global list vlog_modules.

Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agocompiler: Add OVS_CONSTRUCTOR to mark functions as init functions
Helmut Schaa [Fri, 13 Dec 2013 13:05:00 +0000 (14:05 +0100)]
compiler: Add OVS_CONSTRUCTOR to mark functions as init functions

Functions marked with OVS_CONSTRUCTOR are called unconditionally
before main.

Tested with GCC. Untested with MSVC.

Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agolib/util: Make some functions in util.c inline
Helmut Schaa [Fri, 13 Dec 2013 13:17:17 +0000 (14:17 +0100)]
lib/util: Make some functions in util.c inline

str_to_uint, str_to_ulong and str_to_ullong are just wrappers
around the corresponding signed functions. Move these to util.h
and make them inline saving some library exports and letting
the compiler do some more magic.

Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agolib/mac-learning: Add missing semicolon
Helmut Schaa [Fri, 13 Dec 2013 13:17:15 +0000 (14:17 +0100)]
lib/mac-learning: Add missing semicolon

Does not matter as long as ovs_assert is a simple macro but if ovs_assert
is converted to a function this will fail.

Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agolib/random: Make some functions in random.c inline
Helmut Schaa [Fri, 13 Dec 2013 13:17:14 +0000 (14:17 +0100)]
lib/random: Make some functions in random.c inline

random_range, random_uint8 and random_uint16 are basically just wrappers
around random_uint32. Hence, don't export these functions so the compiler
can be smarter when it comes to optimizations.

Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agolib/ofp: Consolidate some struct field orders
Helmut Schaa [Fri, 13 Dec 2013 13:17:13 +0000 (14:17 +0100)]
lib/ofp: Consolidate some struct field orders

Avoid "holes" in structs by moving "small" members together.

Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto-dpif: Remove ovs-appctl dpif/del-flows.
Ethan Jackson [Sat, 7 Dec 2013 23:04:09 +0000 (15:04 -0800)]
ofproto-dpif: Remove ovs-appctl dpif/del-flows.

It's not entirely clear why someone would want to delete the flows
associated with only one bridge and not be fine with simply using
ovs-dpctl del-flows to get all of them.  Additionally, this will be
hard to implement cleanly in the future, so this patch removes it.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto-dpif: Remove avg_subfacet_life.
Ethan Jackson [Thu, 21 Nov 2013 00:18:24 +0000 (16:18 -0800)]
ofproto-dpif: Remove avg_subfacet_life.

This is going to be hard to compute efficiently in the new
architecture introduced in future patches.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto-dpif-upcall: New ovs-appctl upcall/show.
Ethan Jackson [Thu, 21 Nov 2013 02:06:12 +0000 (18:06 -0800)]
ofproto-dpif-upcall: New ovs-appctl upcall/show.

Shows debugging information related to upcall handling.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto-dpif-upcall: Add memory usage stats.
Ethan Jackson [Thu, 21 Nov 2013 01:41:02 +0000 (17:41 -0800)]
ofproto-dpif-upcall: Add memory usage stats.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto: Remove run_fast() functions.
Ethan Jackson [Thu, 21 Nov 2013 01:12:40 +0000 (17:12 -0800)]
ofproto: Remove run_fast() functions.

They don't really make sense in a multithreaded architecture.  Once
flow miss batches are dispatched with, they will be extra useless.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto: Replace reval_seq with a struct seq.
Ethan Jackson [Tue, 17 Sep 2013 21:35:53 +0000 (14:35 -0700)]
ofproto: Replace reval_seq with a struct seq.

Future patches will need to poll_block() on it.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto-dpif-xlate: Add a mechanism to skip wildcard calculation.
Ethan Jackson [Thu, 5 Dec 2013 21:09:27 +0000 (13:09 -0800)]
ofproto-dpif-xlate: Add a mechanism to skip wildcard calculation.

As time goes on and the classifier becomes more complicated, calculate
the wildcard mask will get more and more expensive.  This patch adds a
mechanism to xlate_actions() allowing callers to disable wildcard
calculation when it isn't really necessary.  Used in future patches.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto: Move all statistics accounting into xlate_actions().
Ethan Jackson [Wed, 13 Nov 2013 02:18:01 +0000 (18:18 -0800)]
ofproto: Move all statistics accounting into xlate_actions().

This patch moves statistics accounting for netflow, bonding, netdev,
and mirroring inside xlate_actions().  Moving all statistics into one
place makes it very difficult to mess up.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto: Retrieve ipfix, sflow and netflow in xlate_receive().
Ethan Jackson [Thu, 31 Oct 2013 23:23:13 +0000 (16:23 -0700)]
ofproto: Retrieve ipfix, sflow and netflow in xlate_receive().

This seems cleaner than having separate accessors for them.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoovs-ofctl: Update push_mpls documentation
Simon Horman [Wed, 11 Dec 2013 01:23:36 +0000 (10:23 +0900)]
ovs-ofctl: Update push_mpls documentation

* The ethertype of the packet is always set to
  the ethertype of the push_mpls action
* Drop references to unicast and multicast MPLS ethertypes,
  in RFC5332 0x8847 and 0x8848 do not always have
  those meanings.

Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto: Handle netflow similarly to ipfix and sflow in xlate.
Ethan Jackson [Wed, 30 Oct 2013 23:29:58 +0000 (16:29 -0700)]
ofproto: Handle netflow similarly to ipfix and sflow in xlate.

This patch changes how ofproto-dpif-xlate handles netflow so that it's
more similar to how ipfix and sflow are handled.  This not only
simplifies the code, but it will allow xlate to call netflow in future
if necessary.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto: New function netflow_exists().
Ethan Jackson [Sat, 7 Dec 2013 22:38:14 +0000 (14:38 -0800)]
ofproto: New function netflow_exists().

Useful in future patches.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agonetflow: Make thread safe.
Ethan Jackson [Wed, 30 Oct 2013 19:59:15 +0000 (12:59 -0700)]
netflow: Make thread safe.

In future patches upcall handler threads will need to update netflow.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto: Modularize netflow.
Ethan Jackson [Wed, 23 Oct 2013 06:07:55 +0000 (23:07 -0700)]
ofproto: Modularize netflow.

The netflow code has its tentacles all over the ofproto-dpif module.
This is fine today, but in future facets, which correspond roughly to
netflow_flows, will be retired.  In preparation, this patch hides as
much implementation detail as possible inside the netflow module.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto-dpif: Only run bundles when lacp or bonds are enabled
Joe Stringer [Fri, 13 Dec 2013 03:33:48 +0000 (03:33 +0000)]
ofproto-dpif: Only run bundles when lacp or bonds are enabled

When dealing with a large number of ports, bundle_run() and
bundle_wait() add significant unnecessary processing to the main run
loop, even when there are no bonds and lacp is not configured. This
patch skips such execution if it is unneeded, reducing average CPU usage
of the main thread from about 25% to about 20% in a test environment of
5000 internal ports and 50 tunnel ports running bfd.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
10 years agoovs-dev.py: Build with both GCC and Clang.
Ethan Jackson [Fri, 15 Nov 2013 23:55:19 +0000 (15:55 -0800)]
ovs-dev.py: Build with both GCC and Clang.

Many deployments prefer to use gcc in production and it supports
sparse, while clang supports thread safety analysis and has
more/better warnings and error messages.  This patch changes
ovs-dev.py to build with both so that developers are unlikely to miss
something that one or the other could catch.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Joe Stringer <joestringer@nicira.com>
10 years agobridge: Only store instant_stats on device changes
Joe Stringer [Fri, 13 Dec 2013 03:33:49 +0000 (03:33 +0000)]
bridge: Only store instant_stats on device changes

Previously, we iterated through all interfaces in instant_stats_run(),
grabbing up-to-date information about device and port status. After
assembling all of this information for all interfaces, we would
determine whether anything changed and only send an update to
ovsdb-server if something changed.

This patch uses the new global connectivity_seq to determine whether
there have been any changes before polling all interfaces, which reduces
unnecessary processing in the average case. In a test environment of
5000 internal ports and 50 tunnel ports with bfd, this reduces average
CPU usage of the main thread from about 15% to about 5%. When ports
change status more often than every 100ms, CPU usage is expected to
increase to previous rates.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
10 years agoofproto-dpif: Don't poll ports when nothing changes
Joe Stringer [Fri, 13 Dec 2013 03:33:47 +0000 (03:33 +0000)]
ofproto-dpif: Don't poll ports when nothing changes

Previously, as part of ofproto-dpif run() processing, we would loop
through all ports and poll for changes to carrier, bfd, cfm and lacp
status. This information is used to determine whether bundles may be
enabled, and to perform revalidation when needed.

This patch makes the bfd, cfm, lacp and stp modules aware of the new
global connectivity_seq, notifying on changes in port status. We can
then use connectivity_seq to check if anything has changed before
looping through all ports in ofproto-dpif. In a test environment of 5000
internal ports and 50 tunnel ports with bfd, this reduces average CPU
usage of the main thread from about 35% to about 25%.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
10 years agonetdev: Globally track port status changes
Joe Stringer [Fri, 13 Dec 2013 03:33:46 +0000 (03:33 +0000)]
netdev: Globally track port status changes

Previously, we tracked status changes for ofports on a per-device basis.
Each time in the main thread's loop, we would inspect every ofport
to determine whether the status had changed for corresponding devices.

This patch replaces the per-netdev change_seq with a global 'struct seq'
which tracks status change for all ports. In the average case where
ports are not constantly going up or down, this allows us to check the
sequence once per main loop and not poll any ports. In the worst case,
execution is expected to be similar to how it is currently.

In a test environment of 5000 internal ports and 50 tunnel ports with
bfd, this reduces average CPU usage of the main thread from about 40% to
about 35%.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
10 years agoofproto: Remove arbitrary handler thread limit.
Ethan Jackson [Tue, 12 Nov 2013 02:06:13 +0000 (18:06 -0800)]
ofproto: Remove arbitrary handler thread limit.

There's no particular reason to limit the number of handler threads to
64.  If someone wants to try more, let them.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agoofproto-dpif: Remove dead code.
Ethan Jackson [Thu, 5 Dec 2013 03:32:02 +0000 (19:32 -0800)]
ofproto-dpif: Remove dead code.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agolib/util: More portable use of builtin popcnt.
Jarno Rajahalme [Thu, 12 Dec 2013 16:27:41 +0000 (08:27 -0800)]
lib/util: More portable use of builtin popcnt.

- Use the GCC predefined macro __POPCNT__ to detect the availability
  of fast __builtin_popcnt function.
- Use portable preprocessor macros to detect 64-bit build.
- Only define the 32-bit parts when needed and declare the
  count_1bits_8 at file scope to silence a warning.

This time I have tested all code paths to make sure no warnigns are
generated.

Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Reviewed-by: Simon Horman <horms@verge.net.au>
10 years agobridge: Support changing port numbers.
Ben Pfaff [Wed, 11 Dec 2013 23:40:52 +0000 (15:40 -0800)]
bridge: Support changing port numbers.

Feature #21293.
Requested-by: Anuprem Chalvadi <achalvadi@vmware.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agobridge: Reconfigure in single pass.
Ben Pfaff [Thu, 12 Dec 2013 01:55:26 +0000 (17:55 -0800)]
bridge: Reconfigure in single pass.

This greatly simplifies the reconfiguration code, making it much easier
to understand and modify.  The old multi-pass configuration had the
property that it didn't delay block packet processing as much, but that's
not much of a worry anymore now that latency critical activities have
been moved outside the main thread.

Signed-off-by: Ben Pfaff <blp@nicira.com>
10 years agoovs-thread: Reduce logging level for cpuinfo parsing
Joe Stringer [Thu, 12 Dec 2013 05:46:22 +0000 (05:46 +0000)]
ovs-thread: Reduce logging level for cpuinfo parsing

This was causing test failures on non-Linux platforms.

Reported-by: Ed Maste <emaste@freebsd.org>
Signed-off-by: Joe Stringer <joestringer@nicira.com>
Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
10 years agoofproto-dpif: Ignore non-packet field masks during flow revalidation
Andy Zhou [Wed, 11 Dec 2013 07:32:51 +0000 (23:32 -0800)]
ofproto-dpif: Ignore non-packet field masks during flow revalidation

Commit bcd2633a5be(ofproto-dpif: Store relevant fields for wildcarding
in facet) implements flow revalidation by comparing the newly looked
up flow mask with that of the existing facet.

The non-packet fields, such as register masks, are always cleared
by xlate_actions in the masks stored within facets, but they are not
cleared in the newly looked up flow masks, causing otherwise valid
flows to be declared as invalid flows and be removed from the datapath.

This patch provides a fix. I was able to verify the fix on a system set
up by Ying Chen where the bug can be reproduced.

Bug #21680
Reported by: Ying Chen <yingchen@vmware.com>

Acked-by: Justin Pettit <jpettit@nicira.com>
Signed-off-by: Andy Zhou <azhou@nicira.com>
10 years agolib/jhash.c: Fix comment.
Jarno Rajahalme [Wed, 11 Dec 2013 19:20:00 +0000 (11:20 -0800)]
lib/jhash.c: Fix comment.

Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agodatapath: Fix sparse warning.
Jarno Rajahalme [Wed, 11 Dec 2013 19:20:00 +0000 (11:20 -0800)]
datapath: Fix sparse warning.

Make the new compat function skb_flow_get_ports() static to silence a
sparse warning.

Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
10 years agolacp: Make lacp negotiation test hard-code aggregation keys.
Ben Pfaff [Wed, 11 Dec 2013 06:42:07 +0000 (22:42 -0800)]
lacp: Make lacp negotiation test hard-code aggregation keys.

The lacp implementation takes the aggregation key from the key or portid
of the first slave to be added to the lacp object.  When multiple slaves
are added initially to a lacp object (the most common case), which is the
"first" is arbitrary.  Until now, it seems that the "first" was actually
predictable enough for the tests to (unknowingly) rely on it, but an
upcoming commit will break that.  This commit fixes the test by forcing
a particular aggregation key for each lacp object.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
10 years agonetdev: Log a warning when netdev_set_config() fails.
Ben Pfaff [Wed, 11 Dec 2013 18:50:10 +0000 (10:50 -0800)]
netdev: Log a warning when netdev_set_config() fails.

This allows its callers to avoid duplicating the code.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
10 years agoofproto: Fix alphabetical order in list of structs.
Ben Pfaff [Mon, 9 Dec 2013 23:39:07 +0000 (15:39 -0800)]
ofproto: Fix alphabetical order in list of structs.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
10 years agoofproto: Make ofproto_enumerate_types() match its comment.
Ben Pfaff [Mon, 9 Dec 2013 23:38:25 +0000 (15:38 -0800)]
ofproto: Make ofproto_enumerate_types() match its comment.

The comment says that it clears the passed-in sset, but it didn't.  The bug
didn't actually affect any of the existing callers, which all passed in an
empty sset.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
10 years agoofproto-provider: Add comment on struct ofport.
Ben Pfaff [Wed, 11 Dec 2013 00:34:07 +0000 (16:34 -0800)]
ofproto-provider: Add comment on struct ofport.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
10 years agoofproto-dpif: Improve comment.
Ben Pfaff [Mon, 9 Dec 2013 23:36:43 +0000 (15:36 -0800)]
ofproto-dpif: Improve comment.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
10 years agoClassifier: Track address prefixes.
Jarno Rajahalme [Wed, 11 Dec 2013 19:07:01 +0000 (11:07 -0800)]
Classifier: Track address prefixes.

Add a prefix tree (trie) structure for tracking the used address
space, enabling skipping classifier tables containing longer masks
than necessary for an address field value in a packet header being
classified.  This enables less unwildcarding for datapath flows in
parts of the address space without host routes.

Trie lookup is interwoven to the staged lookup, so that a trie is
searched only when the configured trie field becomes relevant
for the lookup.  The trie lookup results are retained so that each
trie is checked at most once for each classifier lookup.

This implementation tracks the number of rules at each address prefix
for the whole classifier.  More aggressive table skipping would be
possible by maintaining lists of tables that have prefixes at the
lengths encountered on tree traversal, or by maintaining separate
tries for subsets of rules separated by metadata fields.

Prefix tracking is configured via OVSDB.  A new column "prefixes" is
added to the database table "Flow_Table".  "prefixes" is a set of
string values listing the field names for which prefix lookup should
be used.

As of now, the fields for which prefix lookup can be enabled are:
- tun_id, tun_src, tun_dst
- nw_src, nw_dst (or aliases ip_src and ip_dst)
- ipv6_src, ipv6_dst

There is a maximum number of fields that can be enabled for any one
flow table.  Currently this limit is 3.

Examples:

ovs-vsctl set Bridge br0 flow_tables:0=@N1 -- \
 --id=@N1 create Flow_Table name=table0
ovs-vsctl set Bridge br0 flow_tables:1=@N1 -- \
 --id=@N1 create Flow_Table name=table1

ovs-vsctl set Flow_Table table0 prefixes=ip_dst,ip_src
ovs-vsctl set Flow_Table table1 prefixes=[]

Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agodpif-netdev: Properly create exact match masks.
Jarno Rajahalme [Wed, 11 Dec 2013 19:07:01 +0000 (11:07 -0800)]
dpif-netdev: Properly create exact match masks.

Normally OVS userspace supplies a mask along with a flow key for each
new data path flow that should be created.  OVS also provides an
option to disable the kernel wildcarding, in which case the flows are
created without a mask.  When kernel wildcarding is disabled, the
datapath should use exact match, i.e. not wildcard any bits in the
flow key.  Currently, what happens with the userspace datapath instead
is that a datapath flow with mostly empty mask is created (i.e., most
fields are wildcarded), as the current code does not examine the given
mask key length to find out that the mask key is actually empty.  This
results in the same datapath flow matching on packets of multiple
different flows, wrong actions being processed, and stats being
incorrect.

This patch refactors userspace datapath code to explicitly initialize
a suitable exact match mask when a flow put without a mask is
executed.

Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
10 years agocoverage: Fix build when linker sections not supported.
Ben Pfaff [Sat, 7 Dec 2013 16:17:55 +0000 (08:17 -0800)]
coverage: Fix build when linker sections not supported.

With this change, OVS builds and runs fine without linker section support.

CC: Linda Sun <lsun@vmware.com>
CC: Saurabh Shah <ssaurabh@vmware.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>