On Mon, Jun 15, 2026 at 11:00:48PM +0200, Xose Vazquez Perez wrote:
> On 6/15/26 9:07 PM, Benjamin Marzinski wrote:
>
> > This patchset is to deprecate the rr_weight, rr_min_io_rq, and rr_min_io
> > multipath config options. It's a companion of the patchset to handle
> > their deprecation in the man pages.
> > https://lore.kernel.org/all/[email protected]/
> >
> > It does not completely remove the minio setting code, since there are
> > ideas for reusing this in the future. The final patch is simply a change
> > to avoid a kernel warning message that I hit while testing these
> > patches.
> >
> > Benjamin Marzinski (4):
> > libmultipath: deprecate rr_weight, and make it do nothing
> > libmultipath: deprecate rr_min_io_rq, and make it do nothing
> > libmultipath: deprecate rr_min_io
> > libmultipath: don't set hwhander for bio based devices
> >
> > libmultipath/config.c | 6 ---
> > libmultipath/config.h | 6 ---
> > libmultipath/configure.c | 1 -
> > libmultipath/defaults.h | 2 +-
> > libmultipath/dict.c | 90 +++++++---------------------------------
> > libmultipath/dmparser.c | 28 +------------
> > libmultipath/hwtable.c | 3 --
> > libmultipath/propsel.c | 54 ++++++------------------
> > libmultipath/propsel.h | 1 -
> > libmultipath/structs.h | 1 -
> > tests/hwtable.c | 13 ++----
> > tests/test-lib.c | 1 -
> > 12 files changed, 33 insertions(+), 173 deletions(-)
> >
>
> Ben, thanks for this cleanup.
>
> I ran grep to see what remains of these options in the codebase.
> Could you please check if any of the following occurrences should have been
> removed, or if they are intended to stay for now?
>
> 8<---
> $ git grep -Ei "rr_weight|minio|min_io" | grep -vE
> "multipath.conf.5.in|deprecated"
>
Good catch. It appears I misses some things.
> libmultipath/config.c: merge_num(minio);
> libmultipath/config.c: merge_num(minio);
> libmultipath/config.c: hwe->minio = dhwe->minio;
> libmultipath/config.h: int minio;
> libmultipath/config.h: int minio;
> libmultipath/config.h: int minio;
> libmultipath/configure.c: select_minio(conf, mpp);
> libmultipath/configure.c: if (cmpp->minio != mpp->minio) {
> libmultipath/configure.c: select_reload_action(mpp, "minio
> change");
> libmultipath/defaults.h:#define DEFAULT_MINIO 1
> libmultipath/defaults.h:#define DEFAULT_MINIO_RQ 1
> libmultipath/defaults.h:#define DEFAULT_RR_WEIGHT RR_WEIGHT_NONE
> libmultipath/dict.h:int print_rr_weight(struct strbuf *buff, long v);
These were supposed to go.
> libmultipath/dmparser.c: int minio;
> libmultipath/dmparser.c: minio = mp->minio;
> libmultipath/dmparser.c: if (print_strbuf(&buff, " %s
> %d", pp->dev_t, minio) < 0)
> libmultipath/dmparser.c: int def_minio = 0;
> libmultipath/dmparser.c: def_minio =
> atoi(word);
> libmultipath/dmparser.c: if (def_minio
> != mpp->minio)
> libmultipath/dmparser.c:
> mpp->minio = def_minio;
> libmultipath/prioritizers/path_latency.c:#define MIN_IO_NUM 20
> libmultipath/prioritizers/path_latency.c: if ((io_num < MIN_IO_NUM) ||
> (io_num > MAX_IO_NUM)) {
> libmultipath/propsel.c:select_minio_bio (struct config *conf, struct
> multipath * mp)
> libmultipath/propsel.c: mp_set_mpe(minio);
> libmultipath/propsel.c: mp_set_ovr(minio);
> libmultipath/propsel.c: mp_set_hwe(minio);
> libmultipath/propsel.c: mp_set_conf(minio);
> libmultipath/propsel.c: mp_set_default(minio, DEFAULT_MINIO);
> libmultipath/propsel.c: condlog(3, "%s: minio = %i %s", mp->alias, mp->minio,
> origin);
> libmultipath/propsel.c:int select_minio(struct config *conf, struct multipath
> *mp)
> libmultipath/propsel.c: return select_minio_bio(conf, mp);
> libmultipath/propsel.c: do_default(mp->minio, DEFAULT_MINIO_RQ);
This should be mp_set_default(). The generated code is the same, but
there's no reason not to use the standard call here.
> libmultipath/propsel.c: condlog(3, "%s: minio = %i %s", mp->alias, mp->minio,
> origin);
> libmultipath/propsel.h:int select_minio(struct config *conf, struct multipath
> *mp);
> libmultipath/structs.h:enum rr_weight_mode {
> libmultipath/structs.h: RR_WEIGHT_UNDEF,
> libmultipath/structs.h: RR_WEIGHT_NONE,
> libmultipath/structs.h: RR_WEIGHT_PRIO
More missed rr_weight stuff.
> libmultipath/structs.h: int minio;
> tests/test-lib.c: /* pretend new dm, use minio_rq, */
This comment no longer makes any sense, since select_minio no longer
checks the kernel version.
I'll clean these up and resend. If anyone has opinions on what I did
choose to keep around (you could easily make an argument that
DEFAULT_MINIO_RQ should go), let me know.
-Ben
> tests/test-lib.c: select_minio(conf, mp);
> 8<---
>
> Regards.
