Re-vamp how I/O is handled in volumes and plexes.

Analogous to the drive level, give each volume and plex a worker thread
that picks up and processes incoming and completed BIOs.

This should fix the data corruption issues that have come up a few
weeks ago and improve performance, especially of RAID5 plexes.

The volume level needs a little work, though.

1 files changed, 21 insertions, 42 deletions

diff --git a/sys/geom/vinum/geom_vinum_raid5.h b/sys/geom/vinum/geom_vinum_raid5.h
index 454311f..8074f42 100644
--- a/sys/geom/vinum/geom_vinum_raid5.h
+++ b/sys/geom/vinum/geom_vinum_raid5.h
@@ -32,22 +32,23 @@
 /*
  * A single RAID5 request usually needs more than one I/O transaction,
  * depending on the state of the associated subdisks and the direction of the
- * transaction (read or write).  Every subrequest of a RAID5 request,
- * represented by a gv_raid_packet, is defined by a gv_raid5_bit.
+ * transaction (read or write).
  */
 
-/* A subrequest of a RAID5 read/write operation. */
-struct gv_raid5_bit {
-	struct bio	*bio;		/* BIO of this subrequest. */
-	caddr_t		buf;		/* Data buffer of this subrequest. */
-	int		malloc;		/* Flag if data buffer was malloced. */
-	struct g_consumer *consumer;	/* Consumer to send the BIO to. */
-	TAILQ_ENTRY(gv_raid5_bit) list;	/* Entry in the list of this request. */
-};
+#define	GV_ENQUEUE(bp, cbp, pbp)				\
+	do { 							\
+		if (bp->bio_driver1 == NULL) {			\
+			bp->bio_driver1 = cbp;			\
+		} else {					\
+			pbp = bp->bio_driver1;			\
+			while (pbp->bio_caller1 != NULL)	\
+				pbp = pbp->bio_caller1;		\
+			pbp->bio_caller1 = cbp;			\
+		}						\
+	} while (0);
 
-/* Container for one or more gv_raid5_bits; represents a RAID5 I/O request. */
 struct gv_raid5_packet {
-	caddr_t	buf;		/* Data buffer of this RAID5 request. */
+	caddr_t	data;		/* Data buffer of this sub-request- */
 	off_t	length;		/* Size of data buffer. */
 	off_t	lockbase;	/* Deny access to our plex offset. */
 	off_t	offset;		/* The drive offset of the subdisk. */
@@ -56,39 +57,17 @@ struct gv_raid5_packet {
 	int	rqcount;	/* Count of subrequests. */
 
 	struct bio	*bio;	/* Pointer to the original bio. */
-	caddr_t		 data;	/* Pointer to the original data. */
-
-	struct g_consumer *original;	/* Consumer to the data stripe. */
-	struct g_consumer *parity;	/* Consumer to the parity stripe. */
-
-	/* State of this RAID5 packet. */
-	enum {
-	    SETUP,		/* Newly created. */
-	    VALID,		/* Ready for processing. */
-	    IO,			/* Currently doing I/O. */
-	    FINISH		/* Packet has finished. */
-	} state;
-
-	/* Type of this RAID5 transaction. */
-	enum {
-	    JUNK,		/* Newly created, not valid. */
-	    NORMAL,		/* Normal read or write. */
-	    ISPARITY,		/* Containing only parity data. */
-	    NOPARITY,		/* Parity stripe not available. */
-	    DEGRADED,		/* Data stripe not available. */
-	    COMBINED		/* Data and parity stripes ok, others not. */
-	} type;
+	struct bio	*parity;  /* The bio containing the parity data. */
+	struct bio	*waiting; /* A bio that need to wait for other bios. */
 
-	TAILQ_HEAD(,gv_raid5_bit)    bits; /* List of subrequests. */
-	TAILQ_ENTRY(gv_raid5_packet) list; /* Entry in plex's packet list. */
+	TAILQ_HEAD(,gv_bioq)		bits; /* List of subrequests. */
+	TAILQ_ENTRY(gv_raid5_packet)	list; /* Entry in plex's packet list. */
 };
 
-int	gv_build_raid5_req(struct gv_raid5_packet *, struct bio *, caddr_t,
-	    long, off_t);
-void	gv_free_raid5_packet(struct gv_raid5_packet *);
-void	gv_raid5_done(struct bio *);
+int	gv_stripe_active(struct gv_plex *, struct bio *);
+int	gv_build_raid5_req(struct gv_plex *, struct gv_raid5_packet *,
+	    struct bio *, caddr_t, off_t, off_t);
 void	gv_raid5_worker(void *);
-struct gv_raid5_packet  *gv_new_raid5_packet(void);
-struct gv_raid5_bit	*gv_new_raid5_bit(void);
+void	gv_plex_done(struct bio *);
 
 #endif /* !_GEOM_VINUM_RAID5_H_ */