"Theodore Y. Ts'o" <tytso@xxxxxxx> writes: > On Tue, Jul 03, 2018 at 01:06:58PM -0400, Gabriel Krisman Bertazi wrote: >> + * Even if we are doing encodings, an exact-match lookup >> + * could still benefit from DX, so we don't skip it >> + * entirely. Only if it fails to find a match, we >> + * fallback to linear search. > > Why supply the normalized form of the name to ext4_dirhash()? (Or > have ext4_dirhash call the normalization function, but that would > require passing the sbi->encoding to ext4_dirhash) > > That way we can do efficient lookups even if the lookup uses a > different normalized form than what was originally used when the file > was created. (We can also use this to do efficient hash lookups once > we add case folding support.) Makes total sense. Thanks. I forgot to address dx when I changed the design from the mountpoint approach I presented at LSFMM. I will address this in v2. -- Gabriel Krisman Bertazi
