On 12/4/06, Wendy Cheng <wcheng@xxxxxxxxxx> wrote:
Russell Cattelan wrote: > Wendy Cheng wrote: > >> Linux kernel, particularly the VFS layer, is starting to show signs >> of inadequacy as the software components built upon it keep growing. >> I have doubts that it can keep up and handle this complexity with a >> development policy like you just described (filesystem is a dumb >> layer ?). Aren't these DIO_xxx_LOCKING flags inside >> __blockdev_direct_IO() a perfect example why trying to do too many >> things inside vfs layer for so many filesystems is a bad idea ? By >> the way, since we're on this subject, could we discuss a little bit >> about vfs rename call (or I can start another new discussion thread) ? >> >> Note that linux do_rename() starts with the usual lookup logic, >> followed by "lock_rename", then a final round of dentry lookup, and >> finally comes to filesystem's i_op->rename call. Since lock_rename() >> only calls for vfs layer locks that are local to this particular >> machine, for a cluster filesystem, there exists a huge window between >> the final lookup and filesystem's i_op->rename calls such that the >> file could get deleted from another node before fs can do anything >> about it. Is it possible that we could get a new function pointer >> (lock_rename) in inode_operations structure so a cluster filesystem >> can do proper locking ? > > It looks like the ocfs2 guys have the similar problem? > > http://ftp.kernel.org/pub/linux/kernel/people/mfasheh/ocfs2/ocfs2_git_patches/ocfs2-upstream-linus-20060924/0009-PATCH-Allow-file-systems-to-manually-d_move-inside-of-rename.txt > > > Thanks for the pointer. Same as ocfs2, under current VFS code, both GFS1/2 also need FS_ODD_RENAME flag for the rename problem - got an ugly ~200 line draft patch ready for GFS1 (and am looking into GFS2 at this moment). The issue here is, for GFS, if vfs lock_rename() can call us, this complication can be greatly reduced. Will start another thread to see whether the wish can be granted.
Hi Wendy, Have you (or others) made any progress on a possible solution to simplify handling cluster fs do_rename() races (e.g. your proposed lock_rename in inode_operations)? I couldn't find a newer thread that continued this discussion... please advise, thanks. Mike - To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
