Hmm, I'm not sure what the context is here? Features / fs | xfs | ext3 | jfs | reiserfs | ------------------------------------------------- removable fs | ? | ? | N | ? | -------------------------------------------------
I have removed / replaced both jfs and ext2 filesystems housed on hotswap scsi with no visible issues/problems As long as the drive has been cleanly unmounted (and the interface is a hot-swap type) I think all of these should be safe, but perhaps I'm misreading the context. Additional considerations, applications which are likely to care about fs implementation details are listed below. I regret I don't have as many answers to add as I would like (and should be able) to provide, however I presently only have bsd and linux w/ jfs running so I can't make additions. -- forrest 1. AFS AFS - Andrew file system ACL-based access control with kerberos(IV) authentication. - On the server afs volumes modify inode structure of the host FS, This is not fsck-safe! I don't know if it would be safe for journal operations, but it's safe to say that it's only likely to work on ext3. xfs has been reported as both working and not on the server side. ext3 is reported as working 'in theory' as a place for the cache. Some versions of reiser are said to have worked. AFS runs on nearly every unix implementation (and therefor in some supported native FS on all of these unices). While the Transarc-supported versions may provide fsck-saftey through system patches, in all OpenAFS cases I know of, the code is making raw changes to the disk metadata which *require* that the server-volume disks never be fsck-ed. Arla is the kth implementation of AFS I expect it's internals are similar to transarc / openafs but do not know this with any certainty. I run arla as a client on openbsd and it does not seem to play the games with the ffs that are done by afs on it's host filesystems. 2. SE-Linux, NSA's security design runnning on Linux through LSM. SE-Linux expects to track security ID and security context of both processes and persistent (disk) data. It assumes that there is a file:inode mapping which cannot be changed except by the known system call interface (which is hooked by LSM). A fs which does not adhere to this assumption can be expected to break SELinux. I beleive at this time ext3 and reiserfs are tested *by the developers* to work on SE-Linux. I have personally found that jfs works and xfs used to be noted in the se-linux notes as working, and I beleive that it still does. The one potential issue I found was that in checking against the Inode consistency requirement, Reiser indicated that a future release of reiserfs may change that assumption. I suggested to Hans that at least having the option of preserving backward compatibility would be a good idea(tm) and he agreed and said they would plan on that. 3. VmWare VMware inc promises that virtual machines can run off of virtual disks based on ext2-3 and specifically state that there are technical limitations on some journaling fs's (I forget which, either xfs or reiserfs) - they work but with some limitations. I found that jfs worked well as a host for vmware guest os's but did encounter the limitation mentioned by vmware inc.) Naturally, all of these journaled fs's work in guest OS's 4. UML user mode linux Because UML, unlike vmware is provided as a 'translation' layer I assume it's going to have different semantics in how it talks to the disks, and that some low-level calls may not be available, however I have no experience to say what will work or not. Features \ fs | xfs | ext3 | jfs | reiserfs | ------------------------------------------------- chattr | ? | Y | ? | ? | ------------------------------------------------- quotas | Y | ? | Y | ? | ------------------------------------------------- removable fs | ? | ? | N | ? | ------------------------------------------------- NFS | Y | Y | Y | Y | ------------------------------------------------- samba | Y | Y | Y | Y | ------------------------------------------------- coda | ? | ? | ? | ? | ------------------------------------------------- O_DIRECT | Y | N | ? | Y | ------------------------------------------------- SE_Linux | Y | Y | Y* | maybe | ------------------------------------------------- afs (server|cache)| N | poss?| N | N | -- must not fsck server afs volumes ------------------------------------------------- -- ext3 journal replay safe?? arla " | ? | ? | ? | ? | ------------------------------------------------- Vmware (host) | ? | Y | part| part | ------------------------------------------------- _______________________________________________ Jfs-discussion mailing list [EMAIL PROTECTED] http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jfs-discussion
