On Thu 06-02-20 16:04:16, Andreas Dilger wrote: > On Feb 6, 2020, at 3:16 AM, Jan Kara <jack@xxxxxxx> wrote: > > > > On Wed 05-02-20 11:11:13, Andreas Dilger wrote: > >> On Feb 5, 2020, at 3:01 AM, Jan Kara <jack@xxxxxxx> wrote: > >>> > >>> Add two tests adding 50000 files into a htree directory to test various > >>> cases of htree modification. > >> > >> Note that there is already tests/f_large_dir that is creating a large > >> directory via debugfs. Maybe that could be extended rather than adding > >> another long-running test to do almost the same thing? > > > > I didn't know tests/f_large_dir exists. Thanks for the pointer. There are > > just two problems with this: > > > > 1) I wanted to test both with & without metadata_csum because the code > > paths are somewhat different. > > > > 2) Currently we don't have implemented conversion of normal dir into > > indexed one so I need to start with a fs image that already has indexed > > directory. > > > > I suppose I could modify tests/f_large_dir to start with an image to > > address 2) if people are OK with that. And I could just create > > tests/f_large_dir_csum to address 1). > > This would be quite a large image? I thought "e2fsck -fD" would re-pack > htree directories (via e2fsck/rehash.c), so it seems like you could create > a non-htree test filesystem like f_large_dir, set the feature and inode > flags, and then run e2fsck -fD on it? That would also test the rehash code. The image is not big - ~70k packed - (it is enough to have two filled directory blocks for the kernel to enable DIR_INDEX feature). But the idea with using e2fsck -fD is certainly interesting, I'll try that. Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
