Hi,
I tried to get multipathing failover running and failed. It seems to be that
there are two bugs in multipath.c :
- If a path/request fails mp_bh->path doesn't get updated to the new path.
Thus multipath_end_request will access a NULL pointer if a redirected
request returns.
- After fixing this I experienced several other errors. First of all in
bio->bi_flags the BIO_UPTODATE flag is not set when redirecting a request.
Causing the md driver to think that every redirected request failed when
it returns. Since that alone didn't fix the problems, I simply copied the
master_bio again and it looks like everything seems to work.
See patch below. There are probably better ways to fix these things...
Heiko
diff -urN linux-2.6.1/drivers/md/multipath.c linux-2.6.1-md/drivers/md/multipath.c
--- linux-2.6.1/drivers/md/multipath.c Fri Jan 9 07:59:27 2004
+++ linux-2.6.1-md/drivers/md/multipath.c Sat Jan 17 12:58:07 2004
@@ -54,7 +54,7 @@
kfree(mpb);
}
-static int multipath_map (mddev_t *mddev, mdk_rdev_t **rdevp)
+static int multipath_map (mddev_t *mddev, mdk_rdev_t **rdevp, int *path)
{
multipath_conf_t *conf = mddev_to_conf(mddev);
int i, disks = conf->raid_disks;
@@ -68,6 +68,7 @@
for (i = 0; i < disks; i++) {
mdk_rdev_t *rdev = conf->multipaths[i].rdev;
if (rdev && rdev->in_sync) {
+ *path = i;
*rdevp = rdev;
atomic_inc(&rdev->nr_pending);
spin_unlock_irq(&conf->device_lock);
@@ -355,7 +356,7 @@
bio->bi_sector = mp_bh->master_bio->bi_sector;
rdev = NULL;
- if (multipath_map (mddev, &rdev)<0) {
+ if (multipath_map (mddev, &rdev, &mp_bh->path)<0) {
printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
" error for block %llu\n",
bdevname(bio->bi_bdev,b),
@@ -366,7 +367,12 @@
" to another IO path\n",
bdevname(bio->bi_bdev,b),
(unsigned long long)bio->bi_sector);
+
+ *bio = *(mp_bh->master_bio);
bio->bi_bdev = rdev->bdev;
+ bio->bi_rw |= (1 << BIO_RW_FAILFAST);
+ bio->bi_end_io = multipath_end_request;
+ bio->bi_private = mp_bh;
generic_make_request(bio);
}
}
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
