Re: [PATCH 3/6] smb: client: short-circuit negative lookups when parent dir is fully cached

[Henrique Carvalho <henrique.carvalho@suse.com>]

Thanks, Enzo.

I'll address the comments in an updated patch.

On 9/19/25 3:54 PM, Enzo Matsumiya wrote:
> On 09/19, Henrique Carvalho wrote:
>> When the parent directory has a valid and complete cached enumeration we
>> can assume that negative dentries are not present in the directory, thus
>> we can return without issuing a request.
>>
>> This reduces traffic for common ENOENT when the directory entries are
>> cached.
>>
>> Signed-off-by: Henrique Carvalho <henrique.carvalho@suse.com>
> 
> (unrelated to the patch)
> Can a dentry ever get here being positive?  AFAICS it can't, but I might
> have missed some obscure code path.

AFAICS no, but I would also like someone else to confirm.

> 
>> +
>> +        /*
>> +         * We can only rely on negative dentries having the same
>> +         * spelling as the cached dirent if case insensitivity is
>> +         * forced on mount.
>> +         *
>> +         * XXX: if servers correctly announce Case Sensitivity Search
>> +         * on GetInfo of FileFSAttributeInformation, then we can take
>> +         * correct action even if case insensitive is not forced on
>> +         * mount.
>> +         *
>> +         */
> 
> If you're going into the case-sensitiveness hole, several other things
> can be fixed re: cached dirs (and other areas, as we know lol):
> 
>   # mount ... /mnt # not using nocase
>   # cd /mnt
>   # mkdir abc
>   # ls
>   # ls abc
>   # ls ABC
>   # ls ABc
>   # cat /proc/fs/cifs/open_dirs
>   # Version:1
>   # Format:
>   # <tree id> <sess id> <persistent fid> <path>
>   Num entries: 4
>   0x5 0x1d800a8000021 0x760051e015     \ABc       valid file info, valid
> dirents
>   0x5 0x1d800a8000021 0x760051e012     \ABC       valid file info, valid
> dirents
>   0x5 0x1d800a8000021 0x760051e00f     \abc       valid file info, valid
> dirents
>   0x5 0x1d800a8000021 0x760051e00c        valid file info, valid dirents

Agree. The code can likely be improved for users who don't care about
case-exact matching.

> So, as I understand, for the 'nocase' cases, it might be worth building
> the path string and comparing with strcasecmp().
> 
> Other than that, this seems to work fine though.
> 
> (also, extra line at the end of the comment)
> 

-- 
Henrique
SUSE Labs