This routine should return enum ofperr, but in a couple of places
doesn't. When adding one more meter when the meter table is full,
this results in an incorrect error message.
Signed-off-by: Tony van der Peet <tony.vanderpeet@alliedtelesis.co.nz>
Signed-off-by: Ben Pfaff <blp@ovn.org>
CirrusCI [1] is free for open-sorce projects and provides similar
to TravisCI interfaces. One significant difference is ability
to run tasks on FreeBSD instances.
This patch adds simple configuration file to test OVS build
on two FreeBSD releases with gcc and clang.
Unit tests are commented out because they are broken for now.
To enable the automated checks Cirrus CI application from GitHub
Marketplace should be installed. See details in Quick Start guide [2].
[1] https://cirrus-ci.org
[2] https://cirrus-ci.org/guide/quick-start/
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
This patch replaces most of direct accesses to the dp_packet_batch
internal components by appropriate APIs.
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Signed-off-by: Ian Stokes <ian.stokes@intel.com>
The daemon tests used files a lot when shell variables were easier to use
and easier to understand. This commit changes that.
The tests created empty databases that aren't really needed anymore. This
commit changes them to use the ovsdb-server --no-db option instead.
The tests had a lot of common code for checking the ancestry of processes.
This commit factors out a new shell function check_ancestors.
The tests tended to use random pidfile names. This switches to just using
the defaults, which are fine.
The tests didn't check the names of the child processes. This adds those
checks using the new check_process_name shell function. This should avoid
regression of the bug fixed by commit 266f79e32c60 ("daemon-unix: Use
same name for original or restarted children.")
Other minor improvements too.
I only made small updates to the Windows-specific test, because it is hard
for me to verify.
Acked-by: Alin Gabriel Serdean <aserdean@ovn.org>
Signed-off-by: Ben Pfaff <blp@ovn.org>
The commonly used function, opt_dpif_open(), recently became more complex
to check for a datapath argument. Unnecessary dummy parameters for most users
were hence added. Revert back and call the intended api, dp_arg_exists(), to
query for a datapath argument being supplied.
Fixes: 4eeec031d4c4 ("dpctl: Implement dpctl commands for conntrack per zone limit")
Acked-by: Yi-Hung Wei <yihung.wei@gmail.com>
Signed-off-by: Darrell Ball <dlu998@gmail.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
As far as RPMs are concerned, system-id.conf file is declared as being
owned by openvswitch.
At the first ovs startup, ovs-ctl creates this file if none exists without
ensuring this.
We end up with an inconsistency:
$ rpm -V openvswitch
.....UG.. c /etc/openvswitch/system-id.conf
Fix this when ovs-ctl is the one who creates the file.
Note: this issue ends up being hidden after a RPM upgrade, since the
openvswitch user is enforced on the whole /etc/openvswitch directory as a
%post operation.
Acked-by: Timothy Redaelli <tredaelli@redhat.com>
Acked-by: Flavio Leitner <fbl@sysclose.org>
Signed-off-by: David Marchand <david.marchand@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
Since commit bc4fd439586f ("rhel: Ship ovs shared libraries, fedora")
openvswitch-devel RPM package includes both static and shared library.
This is against the Fedora Packaging Guidelines [1].
This commit prevent the static libraries and libtool archives to be shipped.
[1] https://fedoraproject.org/wiki/Packaging:Guidelines#Packaging_Static_Libraries
Acked-by: Flavio Leitner <fbl@sysclose.org>
Signed-off-by: Timothy Redaelli <tredaelli@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
There are no users of these functions.
This change fixes clang build on FreeBSD:
lib/ovs-thread.c:158:1: error: \
mutex 'mutex' is still held at the end of function \
[-Werror,-Wthread-safety-analysis]
XPTHREAD_FUNC1(pthread_mutex_lock, pthread_mutex_t *);
^
lib/ovs-thread.c:138:5: note: expanded from macro 'XPTHREAD_FUNC1'
}
^
Fixes: 4dff0893c376 ("ovs-atomic-pthreads: Use global shared locks for atomic_flag also.")
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
'redzone' not supported in new versions of jemalloc
(since jemalloc 5.0.0).
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
FreeBSD 12 supports 'pthread_rwlock_tryrdlock' without 'pthread'
library. Let's add check for more rare function.
OTOH, Travis-CI environment supports 'pthread_rwlockattr_destroy',
but does not support 'pthread_rwlock_tryrdlock' without 'pthread'.
So, both checks needed.
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
The arguments to sed(1) need to be on the same line in the shell
script or it will just sit there awaiting input.
Signed-off-by: Scott Cheloha <scottcheloha@gmail.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
The command "hostname" is not available in OpenWrt by default. Strace
result of hostname-3.13 on centos7 shows that bare "hostname" command
calls uname() to fetch node name.
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
Oss-fuzz complains that 'vid << VLAN_VID_SHIFT' is causing an error of
"Undefined-shift in parse_odp_action". This is because an invalid
value of vid is passed in push_vlan. This patch adds validation to
the value of vid, in addition to the value of pcp.
Reported-at: https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=11520
Signed-off-by: Yifeng Sun <pkusunyifeng@gmail.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
Right now the man page of ovs-ofctl has "out_group=port". Correcting the
output to group instead of port.
Signed-off-by: Ashish Varma <ashishvarma.ovs@gmail.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
Busybox ip command will have exit code 1 for `ip -V` or `ip help` etc.,
use `ip link show` to cover both iproute2 and busybox ip command
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
The previous syntax installed the libtool wrapper script instead of the
actual binary. This fixes the problem.
CC: James Page <james.page@ubuntu.com>
Fixes: 3d8dededeaf8 ("debian: Rationalize packaging using new debhelper.")
Reported-by: hubo <hubo@jiedaibao.com>
Reported-at: https://mail.openvswitch.org/pipermail/ovs-discuss/2018-October/047625.html
Acked-by: Justin Pettit <jpettit@ovn.org>
Signed-off-by: Ben Pfaff <blp@ovn.org>
This makes the tests all pass cleanly when Address Sanitizer is enabled.
Acked-by: Mark Michelson <mmichels@redhat.com>
Acked-by: Justin Pettit <jpettit@ovn.org>
Signed-off-by: Ben Pfaff <blp@ovn.org>
If another request came in on a particular connection while the previous
request was still being processed, unixctl_server_wait() would wake up the
main loop but unixctl_server_run() wouldn't read the request, resulting in
100% CPU use.
I doubt whether this is a real problem because it's unusual for a client
to attempt to make requests in parallel. I found it while pursuing a 100%
CPU issue but it turned out not to be a bug (the 100% CPU was caused by
a client making requests as fast as possible).
Signed-off-by: Ben Pfaff <blp@ovn.org>
Acked-by: Justin Pettit <jpettit@ovn.org>
Commit dfaf00e started using the result of dpdk_buf_size() to calculate
the available size on each mbuf, as opposed to using the previous
MBUF_SIZE macro. However, this was calculating the mbuf size by adding
up the MTU with RTE_PKTMBUF_HEADROOM and only then aligning to
NETDEV_DPDK_MBUF_ALIGN. Instead, the accounting for the
RTE_PKTMBUF_HEADROOM should only happen after alignment, as per below.
Before alignment:
ROUNDUP(MTU(1500) + RTE_PKTMBUF_HEADROOM(128), 1024) = 2048
After aligment:
ROUNDUP(MTU(1500), 1024) + 128 = 2176
This might seem insignificant, however, it might have performance
implications in DPDK, where each mbuf is expected to have 2k +
RTE_PKTMBUF_HEADROOM of available space. This is because not only some
NICs have course grained alignments of 1k, they will also take
RTE_PKTMBUF_HEADROOM bytes from the overall available space in an mbuf
when setting up their Rx requirements. Thus, only the "After alignment"
case above would guarantee a 2k of available room, as the "Before
alignment" would report only 1920B.
Some extra information can be found at:
https://mails.dpdk.org/archives/dev/2018-November/119219.html
Note: This has been found by Ian Stokes while going through some
af_packet checks.
Reported-by: Ian Stokes <ian.stokes@intel.com>
Fixes: dfaf00e ("netdev-dpdk: fix mbuf sizing")
Signed-off-by: Tiago Lam <tiago.lam@intel.com>
Signed-off-by: Ian Stokes <ian.stokes@intel.com>
An OVN deployment can have multiple logical switches each with a
localnet port connected to a distributed logical router in which one
logical switch may provide external connectivity and the rest of
the localnet logical switches use VLAN tagging in the physical
network.
As reported in [1], external traffic from these localnet VLAN tagged
logical switches are tunnelled to the gateway chassis (chassis hosting
a distributed gateway port which applies NAT rules). As part of the
discussion in [1], there are few possible solutions proposed by
Russell [2]. This patch implements the first option in [2].
With this patch, a new option 'reside-on-redirect-chassis' in 'options'
column of Logical_Router_Port table is added. If the value of this
option is set to 'true' and if the logical router also have a
distributed gateway port, then routing for this logical router port
is centralized in the chassis hosting the distributed gateway port.
If a logical switch 'sw0' is connected to a router 'lr0' with the
router port - 'lr0-sw0' with the address - "00:00:00:00:af:12 192.168.1.1"
, and it has a distributed logical port - 'lr0-public', then the
below logical flow is added in the logical switch pipeline
of 'sw0' if the 'reside-on-redirect-chassis' option is set on 'lr-sw0' -
table=16(ls_in_l2_lkup), priority=50,
match=(eth.dst == 00:00:00:00:af:12 && is_chassis_resident("cr-lr0-public")),
action=(outport = "sw0-lr0"; output;)
"cr-lr0-public" is an internal port binding created by ovn-northd of type
'chassisredirect' for lr0-public in SB DB. Please see "man ovn-sb" for more details.
With the above flow, the packet doesn't enter the router pipeline in
the source chassis. Instead the packet is sent out via the localnet
port of 'sw0'. The gateway chassis upon receiving this packet, runs
the logical router pipeline applying NAT rules and sends the traffic
out via the localnet port of the logical switch providing external connectivity.
The gateway chassis will also reply to the ARP requests for the router port IPs.
With this approach, we avoid redirecting the external traffic to the
gateway chassis via the tunnel port. There are a couple of drawbacks
with this approach:
- East - West routing is no more distributed for the VLAN tagged
localnet logical switches if 'reside-on-redirect-chassis' option is defined
- 'dnat_and_snat' NAT rules with 'logical_mac' and 'logical_port'
columns defined will not work for these logical switches.
This approach is taken for now as it is simple. If there is a requirement
to support distributed routing for these VLAN tenant networks, we
can explore other possible solutions.
[1] - https://mail.openvswitch.org/pipermail/ovs-discuss/2018-April/046543.html
[2] - https://mail.openvswitch.org/pipermail/ovs-discuss/2018-April/046557.html
Reported-at: https://mail.openvswitch.org/pipermail/ovs-discuss/2018-April/046543.html
Reported-by: venkata anil <vkommadi@redhat.com>
Co-authored-by: venkata anil <vkommadi@redhat.com>
Signed-off-by: Numan Siddique <nusiddiq@redhat.com>
Signed-off-by: venkata anil <vkommadi@redhat.com>
Signed-off-by: Gurucharan Shetty <guru@ovn.org>
On my laptop, pthread_create() is always available without -lpthread, but
when I use -fsanitize=address, -lpthread is required to pull in other
threading functions such as pthread_rwlock_tryrdlock(). Thus, with
-fsanitize=address I have to manually add -lpthread to link commands one
way or another. This commit avoids that problem by checking for a
function that is sometimes only available in -lpthread.
Tested-by: Yifeng Sun <pkusunyifeng@gmail.com>
Reviewed-by: Yifeng Sun <pkusunyifeng@gmail.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
The functions to put ofpacts into ofpbufs have always padded them to
OFPACT_ALIGNTO boundaries, but the underlying structures weren't
necessarily padded out. That led to difficulties in a few places where
structures were allocated on the stack instead in an ofpbuf, because
functions like ofpact_init_*() would access beyond the end of the actual
structure. This is true, for example, in test_multipath_main() in
tests/test-multipath.c, which allocates a struct ofpact_multipath on the
stack, and in lswitch_handshake() in learning-switch.c, which allocates
a struct ofpact_output on the stack.
It's possible to fix these individual cases, but it's possible that there
are others that haven't been identified. This commit addresses the issue
another way, by padding all of the ofpact structures to a full multiple
of OFPACT_ALIGNTO and adding assertions to ensure that it can't be screwed
up in the future.
This commit removes the OFPACT_*_SIZE enums, because they are now
equivalent to sizeof(struct ofpact_*) in every case.
Acked-by: Mark Michelson <mmichels@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
With --detach but not --no-chdir, core files and Address Sanitizer logs
don't go into the testsuite directory but end up dropped because it tries
to write them in the root directory.
Acked-by: Mark Michelson <mmichels@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
dp_packet_put_uninit() can reallocate the data buffer, so find the L4
header pointer afterward instead of before.
Found by Address Sanitizer.
Acked-by: Mark Michelson <mmichels@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
When server A sends the leader a request to remove server B from the
cluster, where A != B, the leader sends both A and B a notification when
the removal is complete. Until now, however, the notification (which is a
raft_remove_server_reply message) did not say which server had been
removed, and the receiver did not check. Instead, the receiver assumed
that it had been removed. The result was that B was removed and A stopped
serving out the database even though it was still part of the cluster,
This commit fixes the problem.
Reported-by: ramteja tadishetti <ramtejatadishetti@gmail.com>
Acked-by: Mark Michelson <mmichels@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
When the server is leaving the cluster but remains leader, the
raft_find_server() call can return NULL. Previously this caused a null
dereference. This commit fixes the problem.
Acked-by: Mark Michelson <mmichels@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
raft_update_commit_index() iterates over a sequence of entries that may
have up to two components: a set of servers and a piece of data. When
a set of servers is present, it calls raft_run_reconfigure(), which can
call through the following chain of functions in some cases:
raft_log_reconfiguration()
raft_command_execute__()
raft_command_initiate()
raft_write_entry()
raft_add_entry()
and raft_add_entry() can reallocate raft->entries, which turns the pointer
'e' that raft_update_commit_index() has to the current entry into a wild
pointer.
This commit fixes the problem.
Acked-by: Mark Michelson <mmichels@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
For debugging, it is useful to know the source code line that sent a
given RPC message.
Acked-by: Mark Michelson <mmichels@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
Ordinarily the IDL finds out in advance whether a particular database is
on its server, or it finds out via notifications. But it's also a good
idea to adopt a belt-and-suspenders approach so that, if the IDL does
receive an "unknown database" error, we treat it as a "soft" error that
can be fixed by reconnecting to another server, rather than a "hard" error
that should cause an immediate abort.
Acked-by: Mark Michelson <mmichels@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
Until now the code here would happily try to send transactions to the
database server even if the database connection was not in the correct
state. In some cases this could lead to strange behavior, such as sending
a database transaction for a database that the IDL had just learned did not
exist on the server.
Acked-by: Mark Michelson <mmichels@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
The underlying type for an enum is somewhat unpredictable in that the
compiler and the ABI influence it. The format specifier I used here was
apparently correct for i386 on Linux but wrong for x86-64. It's better to
just use a cast.
Fixes: 597177a283da ("pcap-file: Add support for Linux SLL formatted PCAP files.")
Reported-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
Patch: 69c51582ff
broke Windows userpace - kernel communication.
On windows we create netlink sockets when the handlers are initiated and
reuse them.
This patch remaps the usage of the netlink socket pool.
Fixes:
https://github.com/openvswitch/ovs-issues/issues/164
Signed-off-by: Alin Gabriel Serdean <aserdean@ovn.org>
Acked-by: Shashank Ram <rams@vmware.com>
Tested-by: Shashank Ram <rams@vmware.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
Co-authored-by: Ben Pfaff <blp@ovn.org>
Some keywords and numbers highlighted. Added few spaces to
the examples.
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
This broke the unit tests.
Fixes: 597177a283da ("pcap-file: Add support for Linux SLL formatted PCAP files.")
Acked-by: Alin Gabriel Serdean <aserdean@ovn.org>
Tested-by: Alin Gabriel Serdean <aserdean@ovn.org>
Reported-by: Alin Gabriel Serdean <aserdean@ovn.org>
Tested-by: Timothy Redaelli <tredaelli@redhat.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
Until now, the code here would report an error to its caller as success.
This fixes the problem.
Found by inspection.
Acked-by: Alin Gabriel Serdean <aserdean@ovn.org>
Signed-off-by: Ben Pfaff <blp@ovn.org>
Previously, the buffer size of 'struct ofpbuf b' is less than the
size of 'char buf[512]', this could cause memory overflow of ofpbuf
when calling ofpbuf_put_hex. This patch fixes it.
Reported-at: https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=10865
Signed-off-by: Yifeng Sun <pkusunyifeng@gmail.com>
Signed-off-by: Ben Pfaff <blp@ovn.org>
