Hello Kent,
I've been wanting to set dc->disk.stripe_size and
dc->partial_stripes_expensive for my hardware RAID5 controller since it
defines neither q->limits.io_opt nor q->limits.raid_partial_stripes_expensive
(I think only md sets these at the moment).
I started a patch for the userspace tools to use one of the pad[8]'s as
the stripe_size before starting a kernel patch and thought I would get
some feedback before proceeding.
I figured the superblock is the best place for this since it allocates
memory in bcache_device_init() for this purpose and needs to know as early
as possible. If the SB defines stripe_size, then
partial_stripes_expensive would be implicit.
However, I think sysfs would be best, but stripe_size is read-only
(presumably) because stripe_sectors_dirty and full_dirty_stripes would
need re-allocated.
What do you think the best way for this might be?
--
Eric Wheeler
diff --git a/bcache-super-show.c b/bcache-super-show.c
index 26cc40e..a35a1c1 100644
--- a/bcache-super-show.c
+++ b/bcache-super-show.c
@@ -161,9 +161,11 @@ int main(int argc, char **argv)
printf("dev.uuid\t\t%s\n", uuid);
printf("dev.sectors_per_block\t%u\n"
- "dev.sectors_per_bucket\t%u\n",
+ "dev.sectors_per_bucket\t%u\n"
+ "dev.stripe_size\t\t%" PRIu64 "\n",
sb.block_size,
- sb.bucket_size);
+ sb.bucket_size,
+ sb.stripe_size);
if (!SB_IS_BDEV(&sb)) {
// total_sectors includes the superblock;
diff --git a/bcache.h b/bcache.h
index 61e4252..5213761 100644
--- a/bcache.h
+++ b/bcache.h
@@ -56,7 +56,8 @@ struct cache_sb {
uint64_t flags;
uint64_t seq;
- uint64_t pad[8];
+ uint64_t stripe_size;
+ uint64_t pad[7];
union {
struct {
diff --git a/make-bcache.c b/make-bcache.c
index c626eae..442a8a3 100644
--- a/make-bcache.c
+++ b/make-bcache.c
@@ -151,6 +151,7 @@ void usage()
" -b, --bucket bucket size\n"
" -w, --block block size (hard sector size of SSD, often 2k)\n"
" -o, --data-offset data offset in sectors\n"
+ " -z, --stripe-size bdev stripe size in bytes\n"
" --cset-uuid UUID for the cache set\n"
// " -U UUID\n"
" --writeback enable writeback\n"
@@ -171,6 +172,7 @@ static void write_sb(char *dev, unsigned block_size, unsigned bucket_size,
bool writeback, bool discard, bool wipe_bcache,
unsigned cache_replacement_policy,
uint64_t data_offset,
+ uint64_t stripe_size,
uuid_t set_uuid, bool bdev)
{
int fd;
@@ -232,15 +234,19 @@ static void write_sb(char *dev, unsigned block_size, unsigned bucket_size,
sb.data_offset = data_offset;
}
+ sb.stripe_size = stripe_size;
+
printf("UUID: %s\n"
"Set UUID: %s\n"
"version: %u\n"
"block_size: %u\n"
"data_offset: %ju\n",
+ "stripe_size: %ju\n",
uuid_str, set_uuid_str,
(unsigned) sb.version,
sb.block_size,
- data_offset);
+ data_offset,
+ stripe_size);
} else {
sb.nbuckets = getblocks(fd) / sb.bucket_size;
sb.nr_in_set = 1;
@@ -346,6 +352,7 @@ int main(int argc, char **argv)
int writeback = 0, discard = 0, wipe_bcache = 0;
unsigned cache_replacement_policy = 0;
uint64_t data_offset = BDEV_DATA_START_DEFAULT;
+ uint64_t stripe_size = 0;
uuid_t set_uuid;
uuid_generate(set_uuid);
@@ -362,6 +369,8 @@ int main(int argc, char **argv)
{ "cache-replacement-policy", 1, NULL, 'p' },
{ "data_offset", 1, NULL, 'o' },
{ "data-offset", 1, NULL, 'o' },
+ { "stripe_size", 1, NULL, 'z' },
+ { "stripe-size", 1, NULL, 'z' },
{ "cset-uuid", 1, NULL, 'u' },
{ "help", 0, NULL, 'h' },
{ NULL, 0, NULL, 0 },
@@ -402,6 +411,8 @@ int main(int argc, char **argv)
BDEV_DATA_START_DEFAULT);
exit(EXIT_FAILURE);
}
+ case 'z':
+ stripe_size = atoll(optarg);
break;
case 'u':
if (uuid_parse(optarg, set_uuid)) {
@@ -449,13 +460,13 @@ int main(int argc, char **argv)
write_sb(cache_devices[i], block_size, bucket_size,
writeback, discard, wipe_bcache,
cache_replacement_policy,
- data_offset, set_uuid, false);
+ data_offset, stripe_size, set_uuid, false);
for (i = 0; i < nbacking_devices; i++)
write_sb(backing_devices[i], block_size, bucket_size,
writeback, discard, wipe_bcache,
cache_replacement_policy,
- data_offset, set_uuid, true);
+ data_offset, stripe_size, set_uuid, true);
return 0;
}
--
To unsubscribe from this list: send the line "unsubscribe linux-bcache" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
