On 2/5/16, 12:00 PM, "Bart Van Assche" wrote: >On 02/05/2016 09:26 AM, Himanshu Madhani wrote: >> On 2/4/16, 10:16 AM, "Bart Van Assche" >>wrote: >>> On 02/04/2016 08:45 AM, Himanshu Madhani wrote: >>>> From: Quinn Tran >>>> >>>> #cat /sys/kernel/debug/qla2xxx/qla2xxx_31/tgt_sess >>>> qla2xxx_31 >>>> Port ID Port Name Handle >>>> ff:fc:01 21:fd:00:05:33:c7:ec:16 0 >>>> 01:0e:00 21:00:00:24:ff:7b:8a:e4 1 >>>> 01:0f:00 21:00:00:24:ff:7b:8a:e5 2 >>>> .... >>> >>> Hello Quinn and Himanshu, >>> >>> The above information is not only useful to people who are debugging >>>the >>> QLogic target driver but also to end users who want to check which >>> initiator ports have already logged in to a target port. Hence my >>> proposal to move this information from debugfs to another location >>>(e.g. >>> configfs or sysfs). Users of other target drivers are probably also >>> interested in seeing which sessions are active. How about adding the >>> functionality for reporting session information per target port in the >>> LIO core in such a way that some attributes are available for all >>>target >>> drivers (e.g. initiator port name, SCSI command statistics) and such >>> that target drivers can define additional attributes to exported for >>> each session (e.g. port ID, handle, ...). >> >> We had initially implemented this as a sysfs hook, but knowing that >>sysfs >> is not encouraged, we decided to put this information via debugfs. >>Would it >> make more sense if we send a sysfs patch? > >Hello Himanshu, > >Let's try to reach agreement about the approach first before starting to >rework this patch. > >Five years ago I explained in a message that I posted on the linux-scsi >mailing list why LIO should use sysfs to export information that changes >dynamically (see also http://thread.gmane.org/gmane.linux.scsi/65615/). >I think this patch shows that there is a real need to have detailed >session information from LIO target drivers in user space. We need one >directory per session instead of exporting all session information >through a single file. sysfs is the right filesystem to export such >information because configfs directories should be created by the user >and not from inside the kernel. If no agreement can be reached about >this over e-mail my proposal is to discuss this further at the 2016 >LSF/MM summit. Bart, I see lot of target customers requesting for such information and having something that can be created inside kernel space will be helpful. Let us discuss further at 2016 LSF/MM summit. ‹ Giri > >Thanks, > >Bart.