On Mon, 10 Sep 2018, Todd C. Miller wrote:
...
> --- /dev/null 1 Jan 1970 00:00:00 -0000
> +++ lib/libc/gen/pwcache.h 10 Sep 2018 00:46:44 -0000
...
> +/*
> + * Constants and data structures used to implement group and password file
> + * caches. Traditional passwd/group cache routines perform quite poorly with
> + * archives. The chances of hitting a valid lookup with an archive is quite a
> + * bit worse than with files already resident on the file system. These
> misses
> + * create a MAJOR performance cost. To adress this problem, these routines
> + * cache both hits and misses.
> + *
> + * NOTE: name lengths must be as large as those stored in ANY PROTOCOL and
> + * as stored in the passwd and group files. CACHE SIZES MUST BE PRIME
> + */
Oh yeah, this comment should be massaged to match the new context,
perhaps noting that the limits on these functions are set to provide
effective caching, both positive and negative, for the standard
archive formats, blah blah blah.
(I mentioned "maul"ing before. I'm been tempted to make malloc earn its
pay and use
struct uidc {
uid_t uid;
char valid;
char name[];
};
instead, mallocing the correct size for the name...and eliminating the
UNMLEN/GNMLEN limits!)
Philip
