Skip to content

Latest commit

 

History

History
147 lines (112 loc) · 5.55 KB

README-troubleshooting.md

File metadata and controls

147 lines (112 loc) · 5.55 KB
Raw

Saw "Add port operation failed" when starting simple_switch

Symptom: You tried running simple_switch and you see a message like this:

[10:59:18.649] [bmv2] [D] [thread 20021] Adding interface veth0 as port 0
[10:59:18.649] [bmv2] [E] [thread 20021] Add port operation failed

Note: simple_switch does not quit when this happens. It keeps running, but whatever port it had trouble adding will not be present in your software switch. You won't be able to send packet to your P4 program on that port, and packets sent to that port by your P4 program will disappear.

If the interface that failed was a virtual Ethernet one like veth0, the most likely fix is to run this command to create those interfaces:

$ sudo $HOME/behavioral-model/tools/veth_setup.sh

or if you do not have the directory behavioral-model in your directory $HOME, replace that with the path to your copy of that repository.

Trying to run simple_switch but get error "Nanomsg returned a exception ..."

Symptom: You tried running simple_switch (or some p4c or behavioral-model make check command you ran tried doing so for you), and you see an error message like this:

Nanomsg returned a exception when trying to bind to address 'ipc:///tmp/bmv2-0-notifications.ipc'.
The exception is: Address already in use
This may happen if
1) the address provided is invalid,
2) another instance of bmv2 is running and using the same address, or
3) you have insufficent permissions (e.g. you are using an IPC socket on Unix, the file already exists and you don't have permission to access it)

The most likely fix is to run this command:

$ sudo rm /tmp/bmv2-0-notifications.ipc

It is also possible you have one or more simple_switch processes running on your machine already. See the next section for that.

Explanation:

When you run simple_switch, it normally creates a "Unix domain socket" for possible communication of logging data to other processes. If you use this command, you will see what looks like a file in the /tmp directory named bmv2-0-notifications.ipc:

$ ls -l /tmp/bmv2-0-notifications.ipc 
srwxr-xr-x 1 root root 0 Sep 20 13:11 /tmp/bmv2-0-notifications.ipc

This file will be owned by whatever Unix user you last ran simple_switch as, which is most commonly your normal login user id, or root if you use sudo to run simple_switch as specified elsewhere in this repository (to give simple_switch the needed permissions to send/receive packets to virtual Ethernet ports).

If you earlier ran simple_switch as root and this file was left behind owned as root, and later try to run simple_switch as you, e.g. because you forgot sudo, or because you try to run the tests that come with the p4c or behavioral-model repositories, which do not need root permissions, then when running as you, simple_switch will try to create that file in /tmp and fail, causing the message above. Deleting the file owned by root enables simple_switch to create the file as your user id.

Killing old simple_switch processes

The symptoms could be any of several that I do not have handy at the time of writing this. Feel free to submit some specific symptoms if you have the details that would help others recognize this situation.

Running multiple simple_switch processes that are all trying to read packets from, or send packets to, the same virtual Ethernet ports on the same machine simultaneously, might be exactly what you want in order to simulate a network of multiple switches communicating via Ethernet links to each other.

However, if it is an accident that you have multiple simple_switch processes running, and you want to ensure none are currently running before starting a new one, read below.

$ ps axguw | grep simple_switch
jafinger 20161  0.2  0.3 223116 14452 pts/1    Sl+  11:11   0:00 simple_switch --log-console -i 0@veth0 action-profile-and-selector/action-profile.json
jafinger 20174  0.0  0.0  21536  1036 pts/3    S+   11:11   0:00 grep --color=auto simple_switch

To kill all processes with the name simple_switch (but not simple_switch_CLI):

$ sudo killall -s 9 simple_switch

Compiler gives error message about mark_to_drop

If you see an error message like this when compiling one of these programs:

demo1.p4_16.p4(92): [--Werror=legacy] error: mark_to_drop: Passing 1 arguments when 0 expected
        mark_to_drop(stdmeta);
        ^^^^^^^^^^^^^^^^^^^^^
demo1.p4_16.p4(137): [--Werror=legacy] error: mark_to_drop: Passing 1 arguments when 0 expected
        mark_to_drop(stdmeta);
        ^^^^^^^^^^^^^^^^^^^^^

this is most likely because you are using a version of p4c compiled from source code from 2019-Apr-18 or sooner. In particular, it was compiled before this change was committed to the compiler:

commit 73f100e460e60d8a872522991603a2b2c3cf77ea
Author: Mihai Budiu <mbudiu@vmware.com>
Date:   Thu Apr 18 16:40:56 2019 -0700

    Deprecate mark_to_drop(); replace with mark_to_drop(standard_metadata) function (#1835)
    
    * Deprecate @mark_to_drop; replace with mark_to_drop(standard_metadata)

It was discovered shortly before that change was made that the v1model architecture's definition of mark_to_drop should take the standard_metadata struct as a parameter, because it locally modifies some fields of that structure.

You have two choices for resolving this:

  • Change the P4 program so it calls mark_to_drop() with no parameters, instead of mark_to_drop(stdmeta).
  • Leave the P4 program unchanged, but update your version of the p4c compiler to one built from source from the commit mentioned above, or later.