MD RAID10: Improve redundancy for 'far' and 'offset' algorithms (part 2)

MD RAID10:  Improve redundancy for 'far' and 'offset' algorithms (part 2)

This patch addresses raid arrays that have a number of devices that cannot
be evenly divided by 'far_copies'.  (E.g. 5 devices, far_copies = 2)  This
case must be handled differently because it causes that last set to be of
a different size than the rest of the sets.  We must compute a new modulo
for this last set so that copied chunks are properly wrapped around.

Example use_far_sets=1, far_copies=2, near_copies=1, devices=5:
                "far" algorithm
        dev1 dev2 dev3 dev4 dev5
	==== ==== ==== ==== ====
	[ A   B ] [ C    D   E ]
        [ G   H ] [ I    J   K ]
                    ...
        [ B   A ] [ E    C   D ] --> nominal set of 2 and last set of 3
        [ H   G ] [ K    I   J ]     []'s show far/offset sets

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: NeilBrown <neilb@suse.de>

diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
index 70b58b4bcf896b229bff7e50c18e62b671c0283f..61ed150bd0cf6e065443e83d9bfd5efb73c2fbe7 100644 (file)
--- a/drivers/md/raid10.c
+++ b/drivers/md/raid10.c
@@ -550,6 +550,13 @@ static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio)
        sector_t stripe;
        int dev;
        int slot = 0;
+       int last_far_set_start, last_far_set_size;
+
+       last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1;
+       last_far_set_start *= geo->far_set_size;
+
+       last_far_set_size = geo->far_set_size;
+       last_far_set_size += (geo->raid_disks % geo->far_set_size);
 
        /* now calculate first sector/dev */
        chunk = r10bio->sector >> geo->chunk_shift;
@@ -575,9 +582,16 @@ static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio)
                for (f = 1; f < geo->far_copies; f++) {
                        set = d / geo->far_set_size;
                        d += geo->near_copies;
-                       d %= geo->far_set_size;
-                       d += geo->far_set_size * set;
 
+                       if ((geo->raid_disks % geo->far_set_size) &&
+                           (d > last_far_set_start)) {
+                               d -= last_far_set_start;
+                               d %= last_far_set_size;
+                               d += last_far_set_start;
+                       } else {
+                               d %= geo->far_set_size;
+                               d += geo->far_set_size * set;
+                       }
                        s += geo->stride;
                        r10bio->devs[slot].devnum = d;
                        r10bio->devs[slot].addr = s;
@@ -615,6 +629,18 @@ static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev)
        struct geom *geo = &conf->geo;
        int far_set_start = (dev / geo->far_set_size) * geo->far_set_size;
        int far_set_size = geo->far_set_size;
+       int last_far_set_start;
+
+       if (geo->raid_disks % geo->far_set_size) {
+               last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1;
+               last_far_set_start *= geo->far_set_size;
+
+               if (dev >= last_far_set_start) {
+                       far_set_size = geo->far_set_size;
+                       far_set_size += (geo->raid_disks % geo->far_set_size);
+                       far_set_start = last_far_set_start;
+               }
+       }
 
        offset = sector & geo->chunk_mask;
        if (geo->far_offset) {