On 01/16/2014 07:18 AM, Bart Van Assche wrote:
On 01/15/14 21:36, Robert Love wrote:
On 01/15/2014 05:23 AM, Natti Shwarts wrote:
Hello,
I'm trying to work with SCSI target over FCoE. (using fileio sample of
SCST)
I'm using Linux 3.8.13 , Intel 10G card and HP5900 switch with VSAN 10
for
FCoE.
I am not familiar with the SCST commands to be able to tell you if
you're using it right. I might be able to give you some pointers to help
debug though.
The problem is that the target doesn't send FLOGI to the switch.
(I have windows client that works with the same switch and configuration
and succeed to login)
The following is the operation sequence I perform:
- vconfig add eth2 10
- load scst, scst_user
- fileio_tgt disk0 file0 disk1 file1
- modprobe fcoe
- echo eth2.10 > /sys/module/libfcoe/parameters/create_vn2vn (fcoeadm -c
eth2.10 returned an error)
This is definitely a problem.
* Does eth2.10 exist? Was the VLAN configured on the switch? How did you
configure it on the host/target? Did you manually create the vlan or did
you use the 'fipvlan' application to discover/create it?
* What was the output of the 'echo' command? Why did it fail?
* What did the log files say (dmesg/messages)? You can also turn on
debug_logging for the fcoe kernel modules to get more details. I
wouldn't bother with fcoe.ko at this point as its debug_logging is very
verbose. You could try the following commands after the modprobe.
# echo 0xFFFF > /sys/module/libfcoe/parameters/debug_logging
# echo 0xFFFF > /sys/module/libfc/parameters/debug_logging
- modprobe fcst
- echo "add disk0 0" >
/sys/kernel/scst_tgt/targets/fcst/20\:01\:b8\:ca\:3a\:62\:82\:f9/luns/mgmt
- echo "add disk1 1" >
/sys/kernel/scst_tgt/targets/fcst/20\:01\:b8\:ca\:3a\:62\:82\:f9/luns/mgmt
- echo 1 >
/sys/kernel/scst_tgt/targets/fcst/20\:01\:b8\:ca\:3a\:62\:82\:f9/enabled
My questions are:
1. What is wrong with the above sequence ?
You should have stopped at the fcoeadm fail, no need to proceed with the
failure.
2. Do you have any idea why fcoeadm fails ?
No, I don't have any information about the failure to draw a conclusion.
3. Is "fcoeadm -c eth2.10" is exactly the same as "echo eth2.10 >
/sys/module/libfcoe/parameters/create_vn2vn " ?
No. 'fcoeadm -c eth2.10' with do 'Fabric Mode' and not VN2VN.
I'm not sure what kernel version you're using. For a while we did not
have VN2VN support in fcoeadm. Only in more recent kernels did we change
the sysfs interfaces to use files in /sys/bus/fcoe/. This change allowed
us to add VN2VN support to fcoeadm with the '-m|--mode' option. The
'echo' method is how you enable VN2VN on older kernels. On newer kernels
that method will still work, but it is deprecated because of the new
interfaces.
I have tried the sequence mentioned by Natti myself. What I see on the
initiator system with kernel 3.13-rc8 is the following:
* When not using VLAN (eth0), login works fine - the SCSI devices exported
by the target system appear at the initiator side.
* When using VLAN, no SCSI devices appear after reloading the fcoe module
on the initiator system. However, about 22 seconds later a kernel crash
is triggered. From the netconsole output:
I tried to reproduce this with TCM and fcoe with 3.13-rc8 and it sent me
in a slightly different direction from what you were testing, Bart.
I tried the fcoe_sysfs vn2vn create (i.e. ethX > ctlr_create, vn2vn >
mode, 1 > enabled). I tried with a VLAN and without and neither worked.
I tried the legacy create (i.e. echo ethX > create_vn2vn) with no-vlan
and it worked, luns discovered.
I think the problem for me is that with fcoe_sysfs the mode defaults to
FABRIC until the user sets the mode to VN2VN. When the fcoe_interface is
created via the 'echo ethX > ctlr_create' it does the dev_uc_add based
on the fact that it's in FABRIC mode and it doesn't tell the networking
layer to listen for the VN2VN well-known addresses. When the mode is set
by the user later, these MACs aren't updated.
if (fip->mode == FIP_MODE_VN2VN) {
printk(KERN_ERR "RWL RWL RWL - listening on ALL_VN2VN\n");
dev_mc_add(netdev, FIP_ALL_VN2VN_MACS);
dev_mc_add(netdev, FIP_ALL_P2P_MACS);
} else
dev_mc_add(netdev, FIP_ALL_ENODE_MACS);
I wrote a hacky patch to fix this, but it didn't seem to fix it for me,
so I've got something wrong.
Natti, I'm curious what happens for you if you don't do the 'fcoeadm -c
eth2.10' and instead only do the 'echo eth2.10 >
/sys/module/libfcoe/parameters/create_vn2vn'.
//Rob
_______________________________________________
fcoe-devel mailing list
[email protected]
http://lists.open-fcoe.org/mailman/listinfo/fcoe-devel
