On Tue, Mar 8, 2011 at 5:29 AM, Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote: > Sorry Steve, > > first you need to explain why a totall different protocol is better > suited stuffed into the fucking mess than the cifs module than a clean > start. Second you absolutely need to post patches for review before > commiting them. While others are cleaning up the mess you > singlehandedly are making it worse constantly. It can't go on that way > much longer. Jeff's note may have explain more background, on some of your questions. I agree the code needs cleanup - that is why it is currently marked as config EXPERIMENTAL and BROKEN. The original prototype was a distinct module (the smb2.ko from the smb2.git tree on kernel.org) and had been tested at the previous 3 test events. It would rely on a cifs_common directory to allow cifs and smb2 to share some (perhaps 1/3) library functions. In that implementation I had done more aggressive cleanup (removing all camel case e.g., some other structural changes are now in cifs.ko) It passed some basic file system tests in its old form but was not complete. I had favored keeping the smb2 support in a distinct module to allow more aggressive rewrite of some common areas without risking cifs.ko stability and also because the SMB2 protocol differs from the cifs protocol in many ways (it is not really a "dialect" of the same protocol) but had feedback from Jeff Layton and informally from Andrew Morton to keep it in cifs.ko to make code reuse easier. At Connectathon Jeff, Jeremy Allison, and I met again, also talking with the nfs developers, and reached what seems like a reasonable compromise which will keep the cifs code stable while getting the experimental smb2 code - move from two binaries (cifs.ko and smb2.ko) to one with a common fs type (still cifs), but require that the version be passed in on mount (to use smb2) - make modest changes to cifs structures to handle smb2 specific needs (various new fields in the key global structures, socket/session/tree-connect) - move smb2 code from the prototype to smb2 specific files in cifs.ko (configurable by ifdef) and clean it up as we go Jeff and others have done some reasonable cleanup on cifs.ko over the last three years so getting the smb2 code in tree is important so it can take advantage of some of the good cleanup done in session establishment, socket handling and name resolution in particular. After working on the port, I am seeing more code reuse than I expected, but there is significant cleanup needed. SMB2 code is badly needed. It is a much better protocol than cifs (not just because of security and scalability improvements) and better than nfs in some areas as well. With code upstream, at least those we can make progress, rather than continually having to backport global fixes back. The prototype showed some interesting performance characteristics. Pavel's summer of code work on the smb2.git tree got performance of sequential write faster than nfs, and Jeremy Bongio's patches earlier in the year did the same for the readpages path. This will require rework, but I think is worth getting in experimental code so we can make progress. I welcome cleanup ideas (and patches of course) - but it is going to take a few weeks to get the original port converted to the new structure. -- Thanks, Steve -- To unsubscribe from this list: send the line "unsubscribe linux-cifs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
