Hide Forgot
Description of problem: Bug 682790 has enable biosdevname in RHEL 6.1, it will give nic with name 'em1', 'p1p1' instead of 'ethX'. configuration like "/etc/fcoe/cfg-p1p1" will not be loaded. fcoeadm -i show nothing about p1p1 NIC. Work around: ======== echo p1p1 >/sys/module/fcoe/parameters/create ======== Version-Release number of selected component (if applicable): RHEL 6.1 GA kernel-2.6.32-131.0.15.el6.x86_64 fcoe-utils-1.0.18-2.el6.x86_64 How reproducible: 100% Steps to Reproduce: 1. Find a server with SMBIOS 2.6 or higher 2. Install fcoe-utils. 3. cp /etc/fcoe/cfg-ethx /etc/fcoe-${nic_name} 4. service lldpad start (skip this is that NIC is not 10Gb NIC with switch enabled DCBX) 5. dcbtool sc ${nic_name} dcb on 6. dcbtool sc ${nic_name} app:fcoe e:1 7. service fcoe start Actual results: fcoe initiator cannot started. Expected results: fcoe initiator started as expected. Additional info:
I don't see any issues in the code. It should work when you rename the interfaces to their biosdevname equivalents. The server you mentioned (note: a private comment) uses the ethX scheme.
Petr, Sorry for the late response. That server do using biosdevname scheme. Check p1p1 and p1p2, by default it's in down mode. I check again on fcoe-utils-1.0.20-2.el6.x86_64, still same problem. Does these steps correct? 1. cp /etc/fcoe/cfg-ethx /etc/fcoe/cfg-${nic_name} 2. service lldpad start 3. dcbtool sc ${nic_name} dcb on 4. dcbtool sc ${nic_name} app:fcoe e:1 5. service fcoe start [root@storageqe-51 ~]# fcoeadm -i No FCoE interfaces created.
Petr Sabata, It seems I found the root cause of this. After I disable "trunk" on the FCoE switch. Linux has to use ethX for FCoE rather than ethX.802-fcoe sub-NIC. With "trunk" enabled in FCoE, fcoe services will get fcoe host created automatically, but without it, we need "echo p1p1 >/sys/module/fcoe/parameters/create" I checked on server using ethX name scheme, same results. Can I reopen this bug and change subject?
Gris, I'm not sure what this "trunk" really is but it seems to respond to FIP VLAN discovery; when it's disabled, users have to create their interfaces manually. The interface name (eth1 vs p1p1) should make no difference. What exactly do you think is wrong?
Yes. The interface name make no difference. Current issue is if FCoE switch has disabled VLAN trunk, then fcoe daemon (or lldapd) will not create FC host. You mentioned user have to create their interfaces manually, you mean: 1. ifconfig eth0.fcoe up 2. echo eth0.fcoe > /sys/module/fcoe/parameters/create
(In reply to comment #6) > Yes. The interface name make no difference. > > Current issue is if FCoE switch has disabled VLAN trunk, then fcoe daemon (or > lldapd) will not create FC host. > > You mentioned user have to create their interfaces manually, you mean: > 1. ifconfig eth0.fcoe up > 2. echo eth0.fcoe > /sys/module/fcoe/parameters/create Yes, either that or via fcoeadm -c.
Thanks for the explanation. I keep it as NOTABUG.
the name scheme for ethX.* is for vlan trunk only. command "ifconfig eth0.fcoe up" is not executable. For sub nic like "ethX:X", it cannot echo to /sys/module/fcoe/parameters/create So, for use who disabled the vlan trunk, it can only use ethX directly. In that way, fcoe daemon don't create it's fcoe session automatically. Does Red Hat really support FCoE without vlan trunk? BTW: I change the bug title for better understanding.
(In reply to comment #9) > the name scheme for ethX.* is for vlan trunk only. > > command "ifconfig eth0.fcoe up" is not executable. > > For sub nic like "ethX:X", it cannot echo to /sys/module/fcoe/parameters/create > > So, for use who disabled the vlan trunk, it can only use ethX directly. In that > way, fcoe daemon don't create it's fcoe session automatically. > > Does Red Hat really support FCoE without vlan trunk? It's never really worked for me (but that doesn't mean anything) ... > > BTW: I change the bug title for better understanding.