https://bugzilla.kernel.org/show_bug.cgi?id=196405 --- Comment #1 from Theodore Tso (tytso@xxxxxxx) --- On Mon, Jul 17, 2017 at 09:23:44PM +0000, bugzilla-daemon@xxxxxxxxxxxxxxxxxxx wrote: > See the following shell transcript, run in an ext4 directory, for an > illustration of the bug. The first four commands set things up. The 5th > command > 'mkdir d/64998' calls 'mkdir("d/64998, 0777)' and succeeds. The 6th command > shows that d's resulting st_nlink value is 1, which is incorrect; it should > be > 65000. I discovered this bug in Fedora 26, which runs Linux 4.11.10; from > inspecting the source it appears the bug is still there in 4.13-rc1. This is actually working as intended. In order to support a large number of directories, when we have a 16-bit link count field, we use a link count of 1 to mean, "lots". If you don't want this behavior because strict POSIX compliance is preferable to failing with ENOSPC when you hit that overflow condition, you can create the file system with the dir_nlink feature turned off. I will note that various userspace utilities that try to optimize directory walking by depending on the directory's link count to understand when they have found all of the subdirectories know that directory link count of 1 means "lots". This is why a link count of 1 was chosen. - Ted -- You are receiving this mail because: You are watching the assignee of the bug.
