Re: [PATCH 2/2] RFC: The be2iscsi driver support for bsg
On Mon, 22 Mar 2010 11:16:31 -0400 James Smart james.sm...@emulex.com wrote: About the implementation, I think that it's better to have the common library code rather than just copying the fs bsg code into iscsi. Note: I tried to library-ize the transport implementation on the first pass of the RFC. But it was making things more complex. I tried to explain this, perhaps not very well (http://marc.info/?l=linux-scsim=125725904205688w=2). Ah, I overlooked it. I'll think about it later. Maybe Mike already has some ideas about it. Anyway, library-izing is not urgent at all. We can work on it later. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@googlegroups.com. To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/open-iscsi?hl=en.
Re: [PATCH 2/2] RFC: The be2iscsi driver support for bsg
On Fri, 19 Mar 2010 08:56:30 -0400 James Smart james.sm...@emulex.com wrote: I still want to know why vendors can't do this via the existing netlink interface. open-iscsi uses the netlink interface for some pdu so I guess that having a different channel for management might be a good idea. Separate this request into two needs: The first need is for the iscsi driver to have some kind of entry point to kick off a vendor specific thing - primarily diagnostics and internal f/w and flash mgmt items. Here, using the same mechanism that we had on the FC side, which also supports dma payloads, made a lot of sense. I like and prefer the symmetry. The second need is for common iscsi link/stack mgmt. All vendors would be expected to implement the same items the same way - thus the formalization of the api in the transport. It also makes sense that all use of these common interfaces comes via the open-iscsi mgmt utilities. Given the data set, it could be done by netlink or bsg. I gave some pros/cons on the interfaces in (http://marc.info/?l=linux-scsim=124811693510903w=2). In my mind, the main reason these settings ended up in bsg vs netlink is - the functionality is typically migrating from a vendor-specific ioctl set, which maps rather easily to the bsg model. Not that netlink is that more difficult (although to NLA_ or not may confuse some of the contributors). And, if you already had the bsg infrastructure for the first need, you had to add very little to support it. Thus, the main reason they are together is one of expediency. The first had to be done, so it was very easy to use the same methodology for the second. If vendors use the common data structures via bsg, it's totally fine by me. I see why bsg is preferable. The only thing that I care about is managing any iSCSI HBA with iscsiadm instead of various vendor specific utilities. About the implementation, I think that it's better to have the common library code rather than just copying the fs bsg code into iscsi. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@googlegroups.com. To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/open-iscsi?hl=en.
Re: [PATCH 2/2] RFC: The be2iscsi driver support for bsg
On Thu, 18 Mar 2010 16:02:52 -0500 Mike Christie micha...@cs.wisc.edu wrote: On 03/18/2010 08:58 AM, FUJITA Tomonori wrote: - You invent your hardware specific data structure for the simplest operation such as setting IP address. I think this is what Jay is not trying to do. I think the patch has some extra code like the ISCSI_BSG_HST_VENDOR parts that makes it confusing - it got me too. The ISCSI_BSG_HST_VENDOR code in be2iscsi looks like it is basically disabled (should remove for a formal patch when he sends for merging). It looks like there is a common struct iscsi_bsg_common_format that is getting passed around, and then in be2iscsi the driver is using that info to make a be2iscsi specific command. So scsi_transport_iscsi / ISCSI_SET_IP_ADDR / iscsi_bsg_common_format gets translated by b2iscsi to b2iscsi / OPCODE_COMMON_ISCSI_NTWK_MODIFY_IP_ADDR / be_modify_ip_addr. Yeah, seems you are right. But looks like this patchset also adds the vendor specific message support (ISCSI_BSG_HST_VENDOR)? I still want to know why vendors can't do this via the existing netlink interface. open-iscsi uses the netlink interface for some pdu so I guess that having a different channel for management might be a good idea. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@googlegroups.com. To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/open-iscsi?hl=en.
Re: [PATCH 2/2] RFC: The be2iscsi driver support for bsg
On Wed, 17 Mar 2010 23:37:07 +0530 Jayamohan Kallickal jayamoh...@serverengines.com wrote: This patch contains the necessary changes to support the bsg interface Signed-off-by: Jayamohan Kallickal jayamoh...@serverengines.com --- drivers/scsi/be2iscsi/be_cmds.h | 137 --- drivers/scsi/be2iscsi/be_iscsi.c |3 +- drivers/scsi/be2iscsi/be_main.c | 99 ++-- drivers/scsi/be2iscsi/be_main.h |6 +- drivers/scsi/be2iscsi/be_mgmt.c | 230 +- drivers/scsi/be2iscsi/be_mgmt.h | 107 ++ include/scsi/scsi_bsg_iscsi.h|1 + 7 files changed, 547 insertions(+), 36 deletions(-) diff --git a/drivers/scsi/be2iscsi/be_cmds.h b/drivers/scsi/be2iscsi/be_cmds.h index 49fcc78..d5c14cf 100644 --- a/drivers/scsi/be2iscsi/be_cmds.h +++ b/drivers/scsi/be2iscsi/be_cmds.h @@ -18,6 +18,7 @@ #ifndef BEISCSI_CMDS_H #define BEISCSI_CMDS_H +#include scsi/scsi_bsg_iscsi.h /** * The driver sends configuration and managements command requests to the * firmware in the BE. These requests are communicated to the processor @@ -162,6 +163,13 @@ struct be_mcc_mailbox { #define OPCODE_COMMON_ISCSI_CFG_POST_SGL_PAGES 2 #define OPCODE_COMMON_ISCSI_CFG_REMOVE_SGL_PAGES3 #define OPCODE_COMMON_ISCSI_NTWK_GET_NIC_CONFIG 7 +#define OPCODE_COMMON_ISCSI_NTWK_SET_VLAN14 +#define OPCODE_COMMON_ISCSI_NTWK_CONFIGURE_STATELESS_IP_ADDR 17 +#define OPCODE_COMMON_ISCSI_NTWK_MODIFY_IP_ADDR 21 +#define OPCODE_COMMON_ISCSI_NTWK_GET_DEFAULT_GATEWAY 22 +#define OPCODE_COMMON_ISCSI_NTWK_MODIFY_DEFAULT_GATEWAY 23 +#define OPCODE_COMMON_ISCSI_NTWK_GET_ALL_IF_ID 24 +#define OPCODE_COMMON_ISCSI_NTWK_GET_IF_INFO 25 So this patchset adds the user-kernel space interface for management via bsg, right? Then I have two questions. - open-iscsi already has the user-kernel space interface for management via netlink. Why do you need another via bsg? IOW, why can't you do this via the existing netlink interface? - You invent your hardware specific data structure for the simplest operation such as setting IP address. It means that every vendor invents their hardware specific data structure for management. Users have to use the vendor specific management software or open-iscsi management tool (iscsiadm) needs all the vendor specific code to build the vendor specific data structure (and parse the response). I think that it is exactly what open-iscsi tries to avoid. iscsiadm builds the generic management data structure and scsi_transport_iscsi.c passes it to the drivers. Then the drivers have to handle it. That's the way open-iscsi works now. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@googlegroups.com. To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/open-iscsi?hl=en.
Re: iSCSI initiator implementation question
On Thu, 18 Jun 2009 12:12:49 +0200 Hannes Reinecke h...@suse.de wrote: Hi all, Joachim Worringen wrote: On Jun 18, 11:19 am, Boaz Harrosh bharr...@panasas.com wrote: On 06/18/2009 10:56 AM, Joachim Worringen wrote: Greetings, I tried to use Open-iSCSI with a non-tcp socket type and failed (timeout after connection has been established). Looking at the source, the reason is obvious: for sending data (iscsi_send()), the function pointers from sock-sk are used via the kernel socket API. This works well with non-tcp sockets. Howver, for reading data (see callback handler iscsi_tcp_data_ready()), tcp_read_sock() is used instead of the related kernel socket API call. Sounds good. Could you test out your solution and send a patch? If it tests out, I don't see why not. We don't have a solution yet, just a problem... Is there a specific reason for this? iSCSI would surely benefit from using high-performance, non-tcp sockets if available (I'm talking about SuperSockets in this case, seehttp://www.dolphinics.com/products/dolphin-supersockets.html). Sure sounds nice. If it is a simple and compatible change it sounds very good. I haven't looked into it enough to claim it'll be a simple change. I figure the iscsi recv code will change completely, but would be simplified as we don't need to deal with the tcp buffer details etc. when using a simple socket recv call, but it remains to be seen how this affects compatibility, i.e. conc. partial or interrupted transfer. It is not decided whether we allocate resources for this. Well, the obvious solution here is to implement another transport module (eg iscsi_supersockets.c) much like it's done for iSER. The rest of the code should be sufficiently abstracted to handle it properly. There is a reason why it's called iscsi_tcp.c ... Yeah, I guess that all niche high performance interconnect technology (such as Dolphin, Myrinet, etc) for HPC support RDMA. They could do something like iSER. They support upport kinda socket interface but socket interface is not optimal for them. So modifying iscsi_tcp for them in a strange way doesn't make sense much. Well, anyway seems that 10GbE is slowly killing all niche high performace interconnect technology. --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: iSCSI initiator implementation question
On Thu, 18 Jun 2009 03:53:47 -0700 (PDT) Joachim Worringen worrin...@googlemail.com wrote: On Jun 18, 12:37 pm, FUJITA Tomonori fujita.tomon...@lab.ntt.co.jp wrote: Yeah, I guess that all niche high performance interconnect technology (such as Dolphin, Myrinet, etc) for HPC support RDMA. They could do something like iSER. They support upport kinda socket interface but socket interface is not optimal for them. So modifying iscsi_tcp for them in a strange way doesn't make sense much. As Hannes proposed, deriving a iscsi_ssocks.c from iscsi_tcp.c is the most obvious and painless solution. I just wonder why iscsi_tcp.c calls tcp-functions directly for receiving data although there's a simpler way (which is used for sending data) for sockets. Because it's much faster. Well, anyway seems that 10GbE is slowly killing all niche high performace interconnect technology. Of course Ethernet always wins conc. market share. But 10GigE doesn't help you if you need really low latency; it does hardly give you an advantage over 1GigE in this respect. Seems that the latency diffence with 10GbE is smaller than one with 1GbE. There were lots of niche high performance interconnect companies in 1GbE era but now there are not many. Yes, the latency diffence still matters for some people but the market is getting smaller. --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: iSCSI initiator implementation question
On Thu, 18 Jun 2009 13:33:42 +0200 Bart Van Assche bart.vanass...@gmail.com wrote: On Thu, Jun 18, 2009 at 1:14 PM, FUJITA Tomonorifujita.tomon...@lab.ntt.co.jp wrote: On Thu, 18 Jun 2009 03:53:47 -0700 (PDT) Joachim Worringen worrin...@googlemail.com wrote: On Jun 18, 12:37 pm, FUJITA Tomonori fujita.tomon...@lab.ntt.co.jp wrote: Well, anyway seems that 10GbE is slowly killing all niche high performance interconnect technology. Of course Ethernet always wins conc. market share. But 10GigE doesn't help you if you need really low latency; it does hardly give you an advantage over 1GigE in this respect. Seems that the latency diffence with 10GbE is smaller than one with 1GbE. There were lots of niche high performance interconnect companies in 1GbE era but now there are not many. Yes, the latency diffence still matters for some people but the market is getting smaller. Numbers please. From the last numbers I have seen the latency difference between 10 GbE and fast non-Ethernet networks is still significant. http://74.125.153.132/search?q=cache:lWqSaJ-iJ50J:www.chelsio.com/poster.html%3FsearchText%3Dcybermedia+latency+chelsio+osaka+10.49+microsecondscd=2hl=jact=clnkgl=jp I think that 10GbE vendors are happy to give you lots of results. Maybe it's still significant for you but it's not for some, which I had worked with. --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Crash tgtd and tgtadm with more than 112 volumes.
On Tue, 10 Mar 2009 23:23:30 +0100 Tomasz Chmielewski man...@wpkg.org wrote: Konrad Rzeszutek schrieb: On Tue, Mar 10, 2009 at 02:52:41PM -0700, Ben Greear wrote: I wrote a script to create lots of iscsi volumes on loop devices. Seems to run fine up to 111, and then tgtd crashes and tgtadm gets a buffer overflow. The script I used to create the problem and a capture of the crash is attached. I'm using standard RPMs on Fedora 10, kernel Linux iSCSi-test 2.6.27.15-170.2.24.fc10.x86_64 #1 SMP Wed Feb 11 23:14:31 EST 2009 x86_64 x86_64 x86_64 GNU/Linux Let me know if I can provide any additional info. You are on the wrong mailing list. This is for the Open-iSCSI _initiator_. The mailing list you want is for the Open-iSCSI _target_, which is: iscsitarget-de...@lists.sourceforge.net. The address is s...@vger.kernel.org for a year or so, the iscsitarget-de...@lists.sourceforge.net is an old address which is not used any more. This part is not correct. iscsitarget-de...@lists.sourceforge.net is the mailing list for iSCSI Enterprise Target project (IET), which is not tgt project. I had maintained IET but we have never used IET mailing list for tgt. FYI, our old mailing list is stgt-de...@lists.berlios.de. CC: s...@vger.kernel.org --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Limiting the number of ISCSI-Sessions per target
On Fri, 17 Oct 2008 11:32:41 +0200 Dr. Volker Jaenisch [EMAIL PROTECTED] wrote: Hi Mike! Thank you for the fast reply. Mike Christie schrieb: I am not sure if this is what you are looking for but you can control which sessions get made. If I get you right your solution is not excactly what I want. Do you have multiple portals per target and so we can create a session through each portal to the same target? Something like this: # iscsiadm -m node 20.15.0.12:3260,1 iqn.2001-04.com.home:meanminna 20.15.0.100:3260,1 iqn.2001-04.com.home:meanminna And right now we create two sessions (one through each portal on the target)? If so you can run iscsiadm -m node -T target -p ip:port,tpgt -o update -n node.startup -v manual This means I can set the initiator on one specific node not to start the session automatically? It would be possible to use this approach, but it seems to me no robust solution. The handling of the manual sessions have to be driven by the HA-System (heartbeat). I case of split brain this will give a session to both machines - and booom! I think the a robust solution have to be a session counting instance and thus at best located at the target side. Is there really no Target-side appoach? Some target implementations support such feature, limits the maximum number of sessions. Ask on the mailing list of a target implementation that you use. --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---