Reimplement gvinum orphanization.

gvinum was the only GEOM class, using consumer nstart/nend fields. Making
it do its own accounting for orphanization purposes allows in perspective
to remove burden of that expensive for SMP accounting from GEOM.

Also the previous implementation spinned in a tight event loop, waiting
for all active BIOs to complete, while the new one knows exactly when it
is possible to close the consumer.

MFC after:	1 month
Sponsored by:	iXsystems, Inc.
This commit is contained in:
Alexander Motin 2019-12-27 01:36:53 +00:00
parent 6b5d8e30f1
commit a29df733fa
6 changed files with 33 additions and 8 deletions

View File

@ -127,6 +127,7 @@ void gv_post_event(struct gv_softc *, int, void *, void *, intmax_t,
void gv_worker_exit(struct gv_softc *);
struct gv_event *gv_get_event(struct gv_softc *);
void gv_remove_event(struct gv_softc *, struct gv_event *);
void gv_drive_done(struct gv_drive *);
void gv_drive_tasted(struct gv_softc *, struct g_provider *);
void gv_drive_lost(struct gv_softc *, struct gv_drive *);
void gv_setup_objects(struct gv_softc *);

View File

@ -194,6 +194,20 @@ gv_drive_tasted(struct gv_softc *sc, struct g_provider *pp)
g_topology_unlock();
}
/*
* Count completed BIOs and handle orphanization when all are done.
*/
void
gv_drive_done(struct gv_drive *d)
{
KASSERT(d->active >= 0, ("Negative number of BIOs (%d)", d->active));
if (--d->active == 0 && (d->flags & GV_DRIVE_ORPHANED)) {
d->flags &= ~GV_DRIVE_ORPHANED;
gv_post_event(d->vinumconf, GV_EVENT_DRIVE_LOST, d, NULL, 0, 0);
}
}
/*
* When losing a drive (e.g. hardware failure), we cut down the consumer
* attached to the underlying device and bring the drive itself to a
@ -214,10 +228,10 @@ gv_drive_lost(struct gv_softc *sc, struct gv_drive *d)
cp = d->consumer;
if (cp != NULL) {
if (cp->nstart != cp->nend) {
G_VINUM_DEBUG(0, "dead drive '%s' has still active "
if (d->active > 0) {
G_VINUM_DEBUG(2, "dead drive '%s' has still active "
"requests, unable to detach consumer", d->name);
gv_post_event(sc, GV_EVENT_DRIVE_LOST, d, NULL, 0, 0);
d->flags |= GV_DRIVE_ORPHANED;
return;
}
g_topology_lock();

View File

@ -278,6 +278,7 @@ gv_plex_normal_request(struct gv_plex *p, struct bio *bp, off_t boff,
cbp->bio_data = addr;
cbp->bio_done = gv_done;
cbp->bio_caller1 = s;
s->drive_sc->active++;
/* Store the sub-requests now and let others issue them. */
bioq_insert_tail(p->bqueue, cbp);
@ -580,10 +581,10 @@ gv_sync_request(struct gv_plex *from, struct gv_plex *to, off_t offset,
return (ENOMEM);
}
bp->bio_length = length;
bp->bio_done = gv_done;
bp->bio_done = NULL;
bp->bio_pflags |= GV_BIO_SYNCREQ;
bp->bio_offset = offset;
bp->bio_caller1 = from;
bp->bio_caller1 = from;
bp->bio_caller2 = to;
bp->bio_cmd = type;
if (data == NULL)
@ -694,7 +695,7 @@ gv_grow_request(struct gv_plex *p, off_t offset, off_t length, int type,
}
bp->bio_cmd = type;
bp->bio_done = gv_done;
bp->bio_done = NULL;
bp->bio_error = 0;
bp->bio_caller1 = p;
bp->bio_offset = offset;
@ -802,7 +803,7 @@ gv_init_request(struct gv_sd *s, off_t start, caddr_t data, off_t length)
}
bp->bio_cmd = BIO_WRITE;
bp->bio_data = data;
bp->bio_done = gv_done;
bp->bio_done = NULL;
bp->bio_error = 0;
bp->bio_length = length;
bp->bio_pflags |= GV_BIO_INIT;
@ -819,6 +820,7 @@ gv_init_request(struct gv_sd *s, off_t start, caddr_t data, off_t length)
}
cbp->bio_done = gv_done;
cbp->bio_caller1 = s;
d->active++;
/* Send it off to the consumer. */
g_io_request(cbp, cp);
}
@ -905,7 +907,7 @@ gv_parity_request(struct gv_plex *p, int flags, off_t offset)
}
bp->bio_cmd = BIO_WRITE;
bp->bio_done = gv_done;
bp->bio_done = NULL;
bp->bio_error = 0;
bp->bio_length = p->stripesize;
bp->bio_caller1 = p;

View File

@ -93,11 +93,13 @@ gv_raid5_start(struct gv_plex *p, struct bio *bp, caddr_t addr, off_t boff,
if (wp->waiting != NULL) {
if (wp->waiting->bio_cflags & GV_BIO_MALLOC)
g_free(wp->waiting->bio_data);
gv_drive_done(wp->waiting->bio_caller1);
g_destroy_bio(wp->waiting);
}
if (wp->parity != NULL) {
if (wp->parity->bio_cflags & GV_BIO_MALLOC)
g_free(wp->parity->bio_data);
gv_drive_done(wp->parity->bio_caller1);
g_destroy_bio(wp->parity);
}
g_free(wp);
@ -118,6 +120,7 @@ gv_raid5_start(struct gv_plex *p, struct bio *bp, caddr_t addr, off_t boff,
while (cbp != NULL) {
if (cbp->bio_cflags & GV_BIO_MALLOC)
g_free(cbp->bio_data);
gv_drive_done(cbp->bio_caller1);
g_destroy_bio(cbp);
cbp = bioq_takefirst(p->bqueue);
}
@ -657,6 +660,7 @@ gv_raid5_clone_bio(struct bio *bp, struct gv_sd *s, struct gv_raid5_packet *wp,
cbp->bio_length = wp->length;
cbp->bio_done = gv_done;
cbp->bio_caller1 = s;
s->drive_sc->active++;
if (use_wp)
cbp->bio_caller2 = wp;

View File

@ -260,10 +260,12 @@ struct gv_drive {
#define GV_DRIVE_REFERENCED 0x01 /* The drive isn't really existing,
but was referenced by a subdisk
during taste. */
#define GV_DRIVE_ORPHANED 0x02 /* The drive was orphaned. */
struct gv_hdr *hdr; /* The drive header. */
struct g_consumer *consumer; /* Consumer attached to this drive. */
int active; /* Number of active requests. */
int freelist_entries; /* Count of freelist entries. */
LIST_HEAD(,gv_freelist) freelist; /* List of freelist entries. */

View File

@ -163,4 +163,6 @@ gv_bio_done(struct gv_softc *sc, struct bio *bp)
gv_plex_raid5_done(p, bp);
break;
}
gv_drive_done(s->drive_sc);
}