On Thu, Oct 20, 2022 at 02:58:42PM +0000, Martin Wilck wrote:
> On Fri, 2022-10-07 at 12:35 -0500, Benjamin Marzinski wrote:
> > device-mapper is not able to change the queue_mode on a table reload.
> > Make sure that when multipath sets up the map, both on regular
> > reloads
> > and reconfigures, it keeps the queue_mode the same.
> >
> > Signed-off-by: Benjamin Marzinski <bmarzins@xxxxxxxxxx>
>
> Reviewed-by: Martin Wilck <mwilck@xxxxxxxx>
>
> Some remarks below.
>
> > ---
> > libmultipath/configure.c | 4 +++
> > libmultipath/dmparser.c | 2 ++
> > libmultipath/propsel.c | 55
> > ++++++++++++++++++++++++++++++++++++++
> > libmultipath/structs.h | 7 +++++
> > multipath/multipath.conf.5 | 7 +++--
> > 5 files changed, 73 insertions(+), 2 deletions(-)
> >
> > diff --git a/libmultipath/configure.c b/libmultipath/configure.c
> > index 8af7cd79..41641e30 100644
> > --- a/libmultipath/configure.c
> > +++ b/libmultipath/configure.c
> > @@ -1075,6 +1075,7 @@ int coalesce_paths (struct vectors *vecs,
> > vector mpvec, char *refwwid,
> > struct config *conf = NULL;
> > int allow_queueing;
> > struct bitfield *size_mismatch_seen;
> > + struct multipath * cmpp;
> >
> > /* ignore refwwid if it's empty */
> > if (refwwid && !strlen(refwwid))
> > @@ -1184,6 +1185,9 @@ int coalesce_paths (struct vectors *vecs,
> > vector mpvec, char *refwwid,
> > }
> > verify_paths(mpp);
> >
> > + cmpp = find_mp_by_wwid(curmp, mpp->wwid);
> > + if (cmpp)
> > + mpp->queue_mode = cmpp->queue_mode;
> > if (setup_map(mpp, ¶ms, vecs)) {
> > remove_map(mpp, vecs->pathvec, NULL);
> > continue;
> > diff --git a/libmultipath/dmparser.c b/libmultipath/dmparser.c
> > index 50d13c08..3b37a926 100644
> > --- a/libmultipath/dmparser.c
> > +++ b/libmultipath/dmparser.c
> > @@ -151,6 +151,8 @@ int disassemble_map(const struct _vector
> > *pathvec,
> >
> > free(word);
> > }
> > + mpp->queue_mode = strstr(mpp->features, "queue_mode bio") ?
> > + QUEUE_MODE_BIO : QUEUE_MODE_RQ;
>
> Nitpick: You have spent effort to make multipath-tools support any
> whitepace characters in the previous patches, but here you don't. I can
> see that disassemble_map() generally assumes space characters as word
> delimiters, but at least I see some inconsistency here.
>
> Do you intend to generalize the whitespace handling in
> disassemble_map(), too? Or am I overlooking something?
The kernel will never output any whitespace characters other than space,
so that seems like unnecessary complexity. We also know that the kernel
will never output a feature named something like "enqueue_mode" so we
don't need to check for the space before "queue_mode". But I suppose the
the kernel could add feature named "<something>queue_mode" in the
future. If you think it's important, I send a patch to add a space
before "queue_mode", in the check to deal with that possibility, but I'm
not particularly worried about this.
> >
> > /*
> > * hwhandler
> > diff --git a/libmultipath/propsel.c b/libmultipath/propsel.c
> > index 98e3aad1..d4f19897 100644
> > --- a/libmultipath/propsel.c
> > +++ b/libmultipath/propsel.c
> > @@ -26,6 +26,7 @@
> > #include "strbuf.h"
> > #include <inttypes.h>
> > #include <libudev.h>
> > +#include <ctype.h>
> >
> > pgpolicyfn *pgpolicies[] = {
> > NULL,
> > @@ -413,6 +414,59 @@ void reconcile_features_with_options(const char
> > *id, char **features, int* no_pa
> > }
> > }
> >
> > +static void reconcile_features_with_queue_mode(struct multipath *mp)
> > +{
> > + char *space = NULL, *val = NULL, *mode_str = NULL, *feat;
> > + int features_mode = QUEUE_MODE_UNDEF;
> > +
> > + if (!mp->features)
> > + return;
> > +
> > + pthread_cleanup_push(cleanup_free_ptr, &space);
> > + pthread_cleanup_push(cleanup_free_ptr, &val);
> > + pthread_cleanup_push(cleanup_free_ptr, &mode_str);
>
> I was wondering why we need pthread_cleanup() complexity here, seeing
> no cancellation points in this function. I eventually realized that
> condlog()->dlog()->log_safe()->pthread_mutex_lock() is a cancellation
> point. I suppose we need to clean that up some time.
>
So is fprintf(), so even as a systemd style daemon, condlog() is a
cancellation point.
> > +
> > + if (!(feat = strstr(mp->features, "queue_mode")) ||
> > + feat == mp->features || !isspace(*(feat - 1)) ||
> > + sscanf(feat, "queue_mode%m[ \f\n\r\t\v]%ms", &space,
> > &val) != 2)
>
> Nit: Given that mp->features comes from the multipath.conf, I'm pretty
> sure that it can't contain \n or \r as whitespace characters
> (read_line() would remove them()). Not sure about \f and \v; guess they
> are allowed but I wouldn't swear that they can be used in
> multipath.conf without causing trouble elsewhere.
I was matching the characters that isspace() checks for, for
consistency, since we used isspace() to check that there was a space
before "queue_mode".
-Ben
> > + goto sync_mode;
> > + if (asprintf(&mode_str, "queue_mode%s%s", space, val) < 0) {
> > + condlog(1, "failed to allocate space for queue_mode
> > feature string");
> > + mode_str = NULL; /* value undefined on failure */
> > + goto exit;
> > + }
> > +
> > + if (!strcmp(val, "rq") || !strcmp(val, "mq"))
> > + features_mode = QUEUE_MODE_RQ;
> > + else if (!strcmp(val, "bio"))
> > + features_mode = QUEUE_MODE_BIO;
> > + if (features_mode == QUEUE_MODE_UNDEF) {
> > + condlog(2, "%s: ignoring invalid feature '%s'",
> > + mp->alias, mode_str);
> > + goto sync_mode;
> > + }
> > +
> > + if (mp->queue_mode == QUEUE_MODE_UNDEF)
> > + mp->queue_mode = features_mode;
> > + if (mp->queue_mode == features_mode)
> > + goto exit;
> > +
> > + condlog(2,
> > + "%s: ignoring feature '%s' because queue_mode is set
> > to '%s'",
> > + mp->alias, mode_str,
> > + (mp->queue_mode == QUEUE_MODE_RQ)? "rq" : "bio");
> > +
> > +sync_mode:
> > + if (mode_str)
> > + remove_feature(&mp->features, mode_str);
> > + if (mp->queue_mode == QUEUE_MODE_BIO)
> > + add_feature(&mp->features, "queue_mode bio");
> > +exit:
> > + pthread_cleanup_pop(1);
> > + pthread_cleanup_pop(1);
> > + pthread_cleanup_pop(1);
> > +}
> > +
> > int select_features(struct config *conf, struct multipath *mp)
> > {
> > const char *origin;
> > @@ -428,6 +482,7 @@ out:
> > reconcile_features_with_options(mp->alias, &mp->features,
> > &mp->no_path_retry,
> > &mp->retain_hwhandler);
> > + reconcile_features_with_queue_mode(mp);
> > condlog(3, "%s: features = \"%s\" %s", mp->alias, mp-
> > >features, origin);
> > return 0;
> > }
> > diff --git a/libmultipath/structs.h b/libmultipath/structs.h
> > index 5a713d46..129bdf0e 100644
> > --- a/libmultipath/structs.h
> > +++ b/libmultipath/structs.h
> > @@ -170,6 +170,12 @@ enum max_sectors_kb_states {
> > MAX_SECTORS_KB_MIN = 4, /* can't be smaller than page size
> > */
> > };
> >
> > +enum queue_mode_states {
> > + QUEUE_MODE_UNDEF = 0,
> > + QUEUE_MODE_BIO,
> > + QUEUE_MODE_RQ,
> > +};
> > +
> > enum scsi_protocol {
> > SCSI_PROTOCOL_FCP = 0, /* Fibre Channel */
> > SCSI_PROTOCOL_SPI = 1, /* parallel SCSI */
> > @@ -396,6 +402,7 @@ struct multipath {
> > int needs_paths_uevent;
> > int ghost_delay;
> > int ghost_delay_tick;
> > + int queue_mode;
> > uid_t uid;
> > gid_t gid;
> > mode_t mode;
> > diff --git a/multipath/multipath.conf.5 b/multipath/multipath.conf.5
> > index e098d555..46a4126c 100644
> > --- a/multipath/multipath.conf.5
> > +++ b/multipath/multipath.conf.5
> > @@ -459,8 +459,11 @@ precedence. See KNOWN ISSUES.
> > <mode> can be \fIbio\fR, \fIrq\fR or \fImq\fR, which corresponds to
> > bio-based, request-based, and block-multiqueue (blk-mq) request-
> > based,
> > respectively.
> > -The default depends on the kernel parameter \fBdm_mod.use_blk_mq\fR.
> > It is
> > -\fImq\fR if the latter is set, and \fIrq\fR otherwise.
> > +Before kernel 4.20 The default depends on the kernel parameter
> > +\fBdm_mod.use_blk_mq\fR. It is \fImq\fR if the latter is set, and
> > \fIrq\fR
> > +otherwise. Since kernel 4.20, \fIrq\fR and \fImq\fR both correspond
> > to
> > +block-multiqueue. Once a multipath device has been created, its
> > queue_mode
> > +cannot be changed.
> > .TP
> > The default is: \fB<unset>\fR
> > .RE
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://listman.redhat.com/mailman/listinfo/dm-devel
