Re: RFC / RFT - patch for nfsd_operations interface
Jeff Garzik wrote: > > Neil Brown wrote: > >3/ With this patch, only ext2fs can be exported. Naturally when I > > submit to Linus, all other filesystems which it makes sense to > > export will provide an nfsd_operations structure. > > One word: ug. > > Why does NFS need to stick its fingers into every other filesystem? > And who decides whether or not it makes sense to export a certain > filesystem? (if the answer is "the programmer", I'm sure you can find > several admins who disagree...) > > Jeff > > -- > Jeff Garzik | > Building 1024| Yossarian lives. > MandrakeSoft, Inc. | Neil is addressing a real problem, and you are complaining. NFS may suck, but it is real world important. Nothing in your email suggests an alternative to the problems he addresses. Hans
Re: RFC / RFT - patch for nfsd_operations interface
Neil Brown wrote: >3/ With this patch, only ext2fs can be exported. Naturally when I > submit to Linus, all other filesystems which it makes sense to > export will provide an nfsd_operations structure. One word: ug. Why does NFS [server] need to stick its fingers into every other filesystem? In general for the same reasons VFS (i.e. open/read/close/rmdir/etc) sticks it its fingers into every file system. NFS servers behave rather differently than VFS programs, especially in respect to accessing files without the moral equivalent of open and I/O sequencing. The former makes access across server reboots fairly seemless, the latter tends to confuse the simplistic sequential I/O checks that work with VFS. It's a struggle, even Tru64 space, to keep physical FS designers from doing things that work well with local access but don't with NFS server. As for the patch at hand, I'm not passing judgement cause I don't know the issues here. -Ric Werme -- Eric (Ric) Werme | [EMAIL PROTECTED] Compaq Computer Corp.| http://people.ne.mediaone.net/werme
Re: RFC / RFT - patch for nfsd_operations interface
On Thu, 3 Aug 2000, Jeff Garzik wrote: > Neil Brown wrote: > >3/ With this patch, only ext2fs can be exported. Naturally when I > > submit to Linus, all other filesystems which it makes sense to > > export will provide an nfsd_operations structure. > > > One word: ug. ug, indeed, but for a different reason. "multiple dentries" stuff is BOGUS. Period. Don't do that. > Why does NFS need to stick its fingers into every other filesystem? For the same reasons why it plays games with iget(). Jeff, look at that code. Really (nfsd/nfsfh.c). > And who decides whether or not it makes sense to export a certain > filesystem? (if the answer is "the programmer", I'm sure you can find > several admins who disagree...) Right now "no ->read_inode()" == "no NFS exports".
Re: RFC / RFT - patch for nfsd_operations interface
On Thursday August 3, [EMAIL PROTECTED] wrote: > Neil Brown wrote: > >3/ With this patch, only ext2fs can be exported. Naturally when I > > submit to Linus, all other filesystems which it makes sense to > > export will provide an nfsd_operations structure. > > > One word: ug. > > Why does NFS need to stick its fingers into every other filesystem? NFSD is a client of filesystems (it uses their services). It has certain special needs beyond the needs of VFS (the other client of filesystems). > And who decides whether or not it makes sense to export a certain > filesystem? (if the answer is "the programmer", I'm sure you can find > several admins who disagree...) It it not about which instances of filesystems should be exported. It is about which types of filesystems should be exportable. Certainly the sysadmin decides which actual filesystems should be exported. The programmer decides if hir filesystem is exportable. ext2fs should be. procfs shouldn't. It might be nice if FAT32 could be, but it is hard to implement reliably. Maybe I should have said "should be exportable" rather than the more vague "it makes sense to export". Does that address your issue? NeilBrown
Re: RFC / RFT - patch for nfsd_operations interface
Neil Brown wrote: >3/ With this patch, only ext2fs can be exported. Naturally when I > submit to Linus, all other filesystems which it makes sense to > export will provide an nfsd_operations structure. One word: ug. Why does NFS need to stick its fingers into every other filesystem? And who decides whether or not it makes sense to export a certain filesystem? (if the answer is "the programmer", I'm sure you can find several admins who disagree...) Jeff -- Jeff Garzik | Building 1024| Yossarian lives. MandrakeSoft, Inc. |