Patchset below attempts to clean up the mess around iput() and icache eviction. I'm posting it here for review and it's also available in vfs git (#evict-inode). Changes: * ->clear_inode() and ->delete_inode() are replaced with a single new method (->evict_inode()) that is called whenever inode is kicked out, whether it's done by final iput(), or by memory pressure or by umount. Unlike ->clear_inode(), called has *not* done anything to pagecache or to associated metadata buffers; it's up to ->evict_inode() to deal with those. Note that ->delete_inode() already had been done that... * ->drop_inode() does *NOT* kick inode out itself now; it simply tells caller if fs wants that inode to be kicked out of icache. I.e. generic_delete_inode() is simply "return 1" now and generic_drop_inode() is "return !inode->i_nlink || hlist_unhashed(&inode->i_hash)". All actual work is done by iput_final() after it's called ->drop_inode(). * clear_inode() is gone; use end_writeback(inode) instead. Just as ->delete_inode() had to call clear_inode() exactly once, ->evict_inode() has to call end_writeback() exactly once. Unlike clear_inode(), end_writeback() doesn't do anything to inode-associated metadata; evict it yourself if you have it (i.e. if you use mark_buffer_dirty_inode()). * generic_detach_inode() is gone; it's been only needed because of hugetlbfs and that only because the normal codepath would've hit truncate_inode_pages() before calling into fs. Not an issue with new scheme, so it's not needed anymore. All in-tree filesystems converted. Also included: hostfs fixes (near the areas affected by the main patch sequence) and logfs metadata inode logics sanitizing (ditto). It seems to work here (i.e. survives some local beating) and I'd like to see comments before pushing it into -next. Have fun... -- 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
