On 11/13/20 6:51 AM, Gao Xiang wrote: > If rootino is wrong and misspecified to a subdir inode #, > the following assertion could be triggered: > assert(ino != persp->p_rootino || hardh == persp->p_rooth); > > This patch adds a '-x' option (another awkward thing is that > the codebase doesn't support long options) to address > problamatic images by searching for the real dir, the reason > that I don't enable it by default is that I'm not very confident > with the xfsrestore codebase and xfsdump bulkstat issue will > also be fixed immediately as well, so this function might be > optional and only useful for pre-exist corrupted dumps. > > In details, my understanding of the original logic is > 1) xfsrestore will create a rootdir node_t (p_rooth); > 2) it will build the tree hierarchy from inomap and adopt > the parent if needed (so inodes whose parent ino hasn't > been detected will be in the orphan dir, p_orphh); > 3) during this period, if ino == rootino and > hardh != persp->p_rooth (IOWs, another node_t with > the same ino # is created), that'd be definitely wrong. > > So the proposal fix is that > - considering the xfsdump root ino # is a subdir inode, it'll > trigger ino == rootino && hardh != persp->p_rooth condition; > - so we log this node_t as persp->p_rooth rather than the > initial rootdir node_t created in 1); > - we also know that this node is actually a subdir, and after > the whole inomap is scanned (IOWs, the tree is built), > the real root dir will have the orphan dir parent p_orphh; > - therefore, we walk up by the parent until some node_t has > the p_orphh, so that's it. > > Signed-off-by: Gao Xiang <hsiangkao@xxxxxxxxxx> Thank you for looking into this - I think you now understand xfsdump & xfsrestore better than anyone else on the planet. ;) One question - what happens if the wrong "root inode" is not a directory? I think that it is possible from the old "get the first active inode" heuristic to find any type of file and save it as the root inode. I think that your approach still works in this case, but wanted to double check and see what you think. Thanks, -Eric
