[Topic] Support for checkpoint/restart in fs/mm. [Abstract] Linux-CR implements transparent application checkpoint-restart aiming for inclusion in the mainline kernel. We'd like to briefly present our approach, and discuss pending issues. (for more info on linux-cr: http://www.linux-cr.org/) * Quick review of the approach: Checkpoint: we add .checkpoint method to 'struct file_operations", (which so far simply calls a generic checkpoint function). Restart: to restore the state of an open fd we reopen it based on the pathname saved during checkpoint, and then some adjustments. * Handling of unlinked files: Unlinked files are accessible by their open fd only, but at restart the fd cannot be restored (re-opened) since the original pathname is gone. We'd like to suggset a .relink filesystem method to attach a (new) pathname to an existing inode. * Handling of fsnotify/inotify: Both fsnotify/inotify track inode(s) directly without respective dentry(s). However, to restore an inotify fd, we need to have had recorded the pathname to that inode to re-add the watch at restart. We'd like to discuss possibilities for doing so. * Tracking of mm/pages modification per-process: We envision two optimizations - pre-copy (for fast live-migration) and incremental checkpoints. Both require tracking memory changes such as mm layout and modified pages. Tracking needs to be done on a per-process basis. We'd like to discuss possible approaches. Thanks, Oren. -- 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
