Name

ovn-trace -- Open Virtual Network logical network tracing utility

Synopsis

ovn-trace [options] datapath microflow

ovn-trace [options] --detach

Description

This utility simulates packet forwarding within an OVN logical network. It can be used to run through ``what-if'' scenarios: if a packet originates at a logical port, what will happen to it and where will it ultimately end up? Users already familiar with the Open vSwitch ofproto/trace command described in ovs-vswitch(8) will find ovn-trace to be a similar tool for logical networks.

ovn-trace works by reading the Logical_Flow and other tables from the OVN southbound database (see ovn-sb(5)). It simulates a packet's path through logical networks by repeatedly looking it up in the logical flow table, following the entire tree of possibilities.

ovn-trace simulates only the OVN logical network. It does not simulate the physical elements on which the logical network is layered. This means that, for example, it is unimportant how VMs are distributed among hypervisors, or whether their hypervisors are functioning and reachable, so ovn-trace will yield the same results regardless. There is one important exception: ovn-northd, the daemon that generates the logical flows that ovn-trace simulates, treats logical ports differently based on whether they are up or down. Thus, if you see surprising results, ensure that the ports involved in a simulation are up.

The simplest way to use ovn-trace is to provide datapath and microflow arguments on the command line. In this case, it simulates the behavior of a single packet and exits. For an alternate usage model, see Daemon Mode below.

The datapath argument specifies the name of a logical datapath. Acceptable names are the name from the northbound Logical_Switch or Logical_Router table, the UUID of a record from one of those tables, or the UUID of a record from the southbound Datapath_Binding table.

The microflow argument describes the packet whose forwarding is to be simulated, in the syntax of an OVN logical expression, as described in ovn-sb(5), to express constraints. The parser understands prerequisites; for example, if the expression refers to ip4.src, there is no need to explicitly state ip4 or eth.type == 0x800.

For reasonable L2 behavior, the microflow should include at least inport and eth.dst, plus eth.src if port security is enabled. For example:

    inport == "lp11" && eth.src == 00:01:02:03:04:05 && eth.dst == ff:ff:ff:ff:ff:ff
  

For reasonable L3 behavior, microflow should also include ip4.src and ip4.dst (or ip6.src and ip6.dst) and ip.ttl. For example:

    inport == "lp111" && eth.src == f0:00:00:00:01:11 && eth.dst == 00:00:00:00:ff:11
    && ip4.src == 192.168.11.1 && ip4.dst == 192.168.22.2 && ip.ttl == 64
  

Here's an ARP microflow example:

    inport == "lp123"
    && eth.dst == ff:ff:ff:ff:ff:ff && eth.src == f0:00:00:00:01:11
    && arp.op == 1 && arp.sha == f0:00:00:00:01:11 && arp.spa == 192.168.1.11
    && arp.tha == ff:ff:ff:ff:ff:ff && arp.tpa == 192.168.2.22
  

ovn-trace will reject erroneous microflow expressions, which beyond syntax errors fall into two categories. First, they can be ambiguous. For example, tcp.src == 80 is ambiguous because it does not state IPv4 or IPv6 as the Ethernet type. ip4 && tcp.src > 1024 is also ambiguous because it does not constrain bits of tcp.src to particular values. Second, they can be contradictory, e.g. ip4 && ip6.

Output

ovn-trace supports the three different forms of output, each described in a separate section below. Regardless of the selected output format, ovn-trace starts the output with a line that shows the microflow being traced in OpenFlow syntax.

Detailed Output

The detailed form of output is also the default form. This form groups output into sections headed up by the ingress or egress pipeline being traversed. Each pipeline lists each table that was visited (by number and name), the ovn-northd source file and line number of the code that added the flow, the match expression and priority of the logical flow that was matched, and the actions that were executed.

The execution of OVN logical actions naturally forms a ``control stack'' that resembles that of a program in conventional programming languages such as C or Java. Because the next action that calls into another logical flow table for a lookup is a recursive construct, OVN ``programs'' in practice tend to form deep control stacks that, displayed in the obvious way using additional indentation for each level, quickly use up the horizontal space on all but the widest displays. To make detailed output more readable, without loss of generality, ovn-trace omits indentation for ``tail recursion,'' that is, when next is the last action in a logical flow, it does not indent details of the next table lookup more deeply. Output still uses indentation when it is needed for clarity.

OVN ``programs'' traces also tend to encounter long strings of logical flows with match expression 1 (which matches every packet) and the single action next;. These are uninteresting and merely clutter output, so ovn-trace omits them entirely even from detailed output.

The following excerpt from detailed ovn-trace output shows a section for a packet traversing the ingress pipeline of logical datapath ls1 with ingress logical port lp111. The packet matches a logical flow in table 0 (aka ls_in_port_sec_l2) with priority 50 and executes next(1); to pass to table 1. Tables 1 through 11 are trivial and omitted. In table 12 (aka ls_in_l2_lkup), the packet matches a flow with priority 50 based on its Ethernet destination address and the flow's actions output the packet to the lrp11-attachement logical port.

    ingress(dp="ls1", inport="lp111")
    ---------------------------------
    0. ls_in_port_sec_l2: inport == "lp111", priority 50
    next(1);
    12. ls_in_l2_lkup: eth.dst == 00:00:00:00:ff:11, priority 50
    outport = "lrp11-attachment";
    output;
  

Summary Output

Summary output includes the logical pipelines visited by a packet and the logical actions executed on it. Compared to the detailed output, however, it removes details of tables and logical flows traversed by a packet. It uses a format closer to that of a programming language and does not attempt to avoid indentation. The summary output equivalent to the above detailed output fragment is:

    ingress(dp="ls1", inport="lp111") {
    outport = "lrp11-attachment";
    output;
    ...
    };
  

Minimal Output

Minimal output includes only actions that modify packet data (not including OVN registers or metadata such as outport) and output actions that actually deliver a packet to a logical port (excluding patch ports). The operands of actions that modify packet data are displayed reduced to constants, e.g. ip4.dst = reg0; might be show as ip4.dst = 192.168.0.1; if that was the value actually loaded. This yields output even simpler than the summary format. (Users familiar with Open vSwitch may recognize this as similar in spirit to the datapath actions listed at the bottom of ofproto/trace output.)

The minimal output format reflects the externally seen behavior of the logical networks more than it does the implementation. This makes this output format the most suitable for use in regression tests, because it is least likely to change when logical flow tables are rearranged without semantic change.

Stateful Actions

Some OVN logical actions use or update state that is not available in the southbound database. ovn-trace handles these actions as described below:

ct_next

By default ovn-trace treats flows as ``tracked'' and ``established.'' See the description of the --ct option for a way to override this behavior.

ct_dnat (without an argument)

Forks the pipeline. In one fork, advances to the next table as if next; were executed. The packet is not changed, on the assumption that no NAT state was available. In the other fork, the pipeline continues without change after the ct_dnat action.

ct_snat (without an argument)

This action distinguishes between gateway routers and distributed routers. A gateway router is defined as a logical datapath that contains an l3gateway port; any other logical datapath is a distributed router. On a gateway router, ct_snat; is treated as a no-op. On a distributed router, it is treated the same way as ct_dnat;.

ct_dnat(ip)

ct_snat(ip)

Forks the pipeline. In one fork, sets ip4.dst (or ip4.src) to ip and ct.dnat (or ct.snat) to 1 and advances to the next table as if next; were executed. In the other fork, the pipeline continues without change after the ct_dnat (or ct_snat) action.

ct_lb;

ct_lb(ip[:port]...);

Forks the pipeline. In one fork, sets ip4.dst (or ip6.dst) to one of the load-balancer addresses and the destination port to its associated port, if any, and sets ct.dnat to 1. With one or more arguments, gives preference to the address specified on --lb-dst, if any; without arguments, uses the address and port specified on --lb-dst. In the other fork, the pipeline continues without change after the ct_lb action.

ct_commit

put_arp

put_nd

These actions are treated as no-ops.

Daemon Mode

If ovn-trace is invoked with the --detach option (see Daemon Options, below), it runs in the background as a daemon and accepts commands from ovs-appctl (or another JSON-RPC client) indefinitely. The currently supported commands are described below.

trace [options] datapath microflow

Traces microflow through datapath and replies with the results of the trace. Accepts the options described under Trace Options below.

exit

Causes ovn-trace to gracefully terminate.

Options

Trace Options

--detailed

--summary

--minimal

These options control the form and level of detail in ovn-trace output. If more than one of these options is specified, all of the selected forms are output, in the order listed above, each headed by a banner line. If none of these options is given, --detailed is the default. See Output, above, for a description of each kind of output.

--all

Selects all three forms of output.

--ovs[=remote]

Makes ovn-trace attempt to obtain and display the OpenFlow flows that correspond to each OVN logical flow. To do so, ovn-trace connects to remote (by default, unix:@RUNDIR@/br-int.mgmt) over OpenFlow and retrieves the flows. If remote is specified, it must be an active OpenFlow connection method described in ovsdb(7).

To make the best use of the output, it is important to understand the relationship between logical flows and OpenFlow flows. ovn-architecture(7), under Architectural Physical Life Cycle of a Packet, describes this relationship. Keep in mind the following points:

• ovn-trace currently shows all the OpenFlow flows to which a logical flow corresponds, even though an actual packet ordinarily matches only one of these.
• Some logical flows can map to the Open vSwitch ``conjunctive match'' extension (see ovs-fields(7)). Currently ovn-trace cannot display the flows with conjunction actions that effectively produce the conj_id match.
• Some logical flows may not be represented in the OpenFlow tables on a given hypervisor, if they could not be used on that hypervisor.
• Some OpenFlow flows do not correspond to logical flows, such as OpenFlow flows that map between physical and logical ports. These flows will never show up in a trace.
• When ovn-trace omits uninteresting logical flows from output, it does not look up the corresponding OpenFlow flows.

--ct=flags

This option sets the ct_state flags that a ct_next logical action will report. The flags must be a comma- or space-separated list of the following connection tracking flags:

• trk: Include to indicate connection tracking has taken place. (This bit is set automatically even if not listed in flags.
• new: Include to indicate a new flow.
• est: Include to indicate an established flow.
• rel: Include to indicate a related flow.
• rpl: Include to indicate a reply flow.
• inv: Include to indicate a connection entry in a bad state.
• dnat: Include to indicate a packet whose destination IP address has been changed.
• snat: Include to indicate a packet whose source IP address has been changed.

The ct_next action is used to implement the OVN distributed firewall. For testing, useful flag combinations include:

• trk,new: A packet in a flow in either direction through a firewall that has not yet been committed (with ct_commit).
• trk,est: A packet in an established flow going out through a firewall.
• trk,rpl: A packet coming in through a firewall in reply to an established flow.
• trk,inv: An invalid packet in either direction.

A packet might pass through the connection tracker twice in one trip through OVN: once following egress from a VM as it passes outward through a firewall, and once preceding ingress to a second VM as it passes inward through a firewall. Use multiple --ct options to specify the flags for multiple ct_next actions.

When --ct is unspecified, or when there are fewer --ct options than ct_next actions, the flags default to trk,est.

--lb-dst=ip[:port]

Sets the IP from VIP pool to use as destination of the packet. --lb-dst is not available in daemon mode.

--friendly-names

--no-friendly-names

When cloud management systems such as OpenStack are layered on top of OVN, they often use long, human-unfriendly names for ports and datapaths, for example, ones that include entire UUIDs. They do usually include friendlier names, but the long, hard-to-read names are the ones that appear in matches and actions. By default, or with --friendly-names, ovn-trace substitutes these friendlier names for the long names in its output. Use --no-friendly-names to disable this behavior; this option might be useful, for example, if a program is going to parse ovn-trace output.

Daemon Options