Amir Goldstein <amir73il@xxxxxxxxx> writes: > On Wed, Apr 9, 2025 at 5:01 PM André Almeida <andrealmeid@xxxxxxxxxx> wrote: >> >> To add overlayfs support casefold filesystems, make >> ovl_cache_entry_find() support casefold dentries. >> >> For the casefold support, just comparing the strings does not work >> because we need the dentry enconding, so make this function find the >> equivalent dentry for a giving directory, if any. >> >> Also, if two strings are not equal, strncmp() return value sign can be >> either positive or negative and this information can be used to optimize >> the walk in the rb tree. utf8_strncmp(), in the other hand, just return >> true or false, so replace the rb walk with a normal rb_next() function. > > You cannot just replace a more performance implementation with a > less performant one for everyone else just for your niche use case. > Also it is the wrong approach. > > This code needs to use utf8_normalize() to store the normalized > name in the rbtree instead of doing lookup and d_same_name(). > and you need to do ovl_cache_entry_add_rb() with the normalized > name anotherwise you break the logic of ovl_dir_read_merged(). > > Gabriel, > > Do you think it makes sense to use utf8_normalize() from this code > directly to generate a key for "is this name found in another layer" > search tree? utf8_normalize is on its way out of the kernel and I don't think it would help here, since it doesn't handle case-insensitive equivalent names either, bug is just as expensive. utf8_casefold might do what you want, but it is expensive as well. With it, you can store the folded version and be sure it is a byte-per-byte match. Alternatively, you can keep the existing name and open code something similar to what generic_ci_match does: check with strncmp first and only if the mountpoint must consider case-insensitive, do a utf8_strncasecmp_folded if the first check wasn't a match. -- Gabriel Krisman Bertazi
