On Wed, Nov 30, 2022 at 11:20:11AM +0800, Baokun Li wrote: > > If we can protect against the problem by adding a check that has other > > value as well (such as making usre that when ext4_iget fetches a > > special inode, we enforce that i_links_couint must be > 0), maybe > > that's worth it. > > Yes, but some special inodes allow i_links_couint to be zero, > such as the uninitialized boot load inode. That's a good point; but the only time when a special inode can validly have a zero i_links_count is when it has no blocks associated to it. Otherwise, when the file system releases the inode using iput() when the file system is unmounted, all of the blocks will get released when the inode is evicted. So we can have ext4_iget() allow fetching an inode if i_blocks[] is zeros. But if it has any blocks and i_links_count is non-zero, something must be terribly wrong with that inode. Cheers, - Ted
