2011/7/6 Mike Christie <[email protected]>: > On 07/06/2011 11:45 AM, Joachim wrote: >> Hello, >> >> I am currently trying to implement an iscsi target driver for a remote >> storage space with tgt. >> >> At first, the driver was slow (quite normal), but worked quite well, >> with a testing script i wrote. >> Then, I added some features for it to get faster, and thus the problem >> appeared. >> >> Actually, after the login to the target with iscsiadm, I partition the >> disk, create some filesystem, and then make a copy of a random file >> tree, before umounting the partition, and logout. >> When i try to mount the partition again (after all required >> manipulations with iscsiadm), i stumble upon a series of IO Errors, >> meaning that my filesystem is totally corrupted, and that some of the >> data weren't saved by my driver (be it that i didnt receive it or that >> i could not save it) >> >> When I use the sync command before umounting, I actually succeed to >> read the whole file system again afterwards. >> >> While searching a solution, I found this in another thread of this >> mailing list (http://groups.google.com/group/open-iscsi/browse_thread/ >> thread/e445471fab2af459/b8fae3dfb6e6c68f?pli=1) : >> >>>> is the Sync cache that occurs at logout actually sending data? or is >>>> it just sending a scsi msg to flush/checkpoint the target side? >>> >>> No data. It is just telling the target side to basically flush what it >>> has in its cache. For example tgt does a sync() to write out data to the >>> backing store. I believe IET does the equivalent of a sync() but it is >>> done from the kernel since IET is implemented there. Look at the source >>> for IET or just fire off a mail to that list if you want to know all the >>> details of what it does. >> >> >> If i'm not misunderstanding what's written here, That means that when >> i type the command >> $ iscsiadm --mode node --targetname mytarget --logout >> All that is done is actually sending a sync command to the target (tgt >> in my case) without sending any data relating to the system/iscsid >> cache ? > > Right. The iscsi layer removes the devices that were getting accessed by > the session and this causes the scsi layer to send a sync cache command > if the device had caching on. > > But the iscsi layer does not do any caching of data. It passes/sends > whatever it has when it gets it. Upper layers like filesystems or memory > layer cache data. If you were writing out data and ran the logout > command, the logout and sync cache can occur before all the writing is > done. The iscsi layer does not know what the layers above have to write > out, so the iscsi layer just kills the devices and has the scsi layer > send the sync cache when ever you run the logout command. > > The user needs to make sure they have unmounted FSes or had they apps > sync their data before running the logout command. > >> >> My hypothesis is thus that my problem originates from this behavior of >> the iscsiadm logout action. If I'm not wrong about that, how can i >> make it so that a synchronization is done before the iscsi logout ? Or >> I am completely wrong, and then do you have any insight as to what my >> problem really is ? > > I do not think it is really a iscsiadm issue or at least there is > nothing the iscsi layer can do. It just does not know what the upper > layers are doing. The iscsi layer just handles iscsi stuff. What is in > the FS/VM/block/userspace buffers/cache at the time of the logout > command is not known to us. You need to sync your data, unmount > filesystems, etc, before you logout a session. For example when the > system shutsdown, the FSes will be unmounted and data written out, then > the iscsi service is shutdown which does the logout command. > > > Did you check if the target is even getting the sync cache command btw? > Some older kernels had a bug in the scsi or driver model layer where it > never got sent. Also check your cache settings and make sure the cache > is on or the sync cache does not get sent. >
Yeah, I checked that my tgt driver received the sync when i do the logout command, and that is the case . No problem over this side. By the way, I work on an ubuntu 10.4 LTS, so I don't think that could come from an old kernel version (or a too old one). What seemed strange to me is that without the sync command, my driver did not update many blocs/sectors on my storage, even though I did properly umount and logout. To sum it up, I could not find where were the lost data, and that is the reason why i came here with my question. Anyways, thanks for the answer, it seems i'll have to dig a little bit more in how the FS side works to understand where is my shortcoming then. Thanks again :) -- David Pineau, Etudiant Epitech -Promotion 2012- Assistant C/Unix, Assistant Langages Avancés Membre LSE (Laboratoire Système Epita/Epitech). -- You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/open-iscsi?hl=en.
