Giuseppe Scrivano
2009-Aug-02 22:24 UTC
[PATCH] Btrfs: don''t allocate more than max_available bytes when looped
Hello, I had a problem running this simple test case that I was writing to test the GNU coreutils cp tool: truncate --size=256M btrfs.img mkfs.btrfs btrfs.img mkdir btrfs mount -t btrfs -o loop btrfs.img btrfs dd bs=1M count=200 if=/dev/zero of=btrfs/alloc.test with the current git version I get an ENOSPC error on dd, even if the available space is 256M. The simple patch that I have attached seems to fix it. Am I missing someting? If there is something wrong please let me know and I''ll try to fix it. Regards, Giuseppe From d6345f720f9fa5e60edd904d90d56bf9e96fb118 Mon Sep 17 00:00:00 2001 From: Giuseppe Scrivano <gscrivano@gnu.org> Date: Sun, 2 Aug 2009 23:29:21 +0200 Subject: [PATCH] Btrfs: don''t allocate more than max_available bytes when looped avoid that __btrfs_alloc_chunk can try to allocate more than max_available bytes if it wasn''t possible on the first attempt. --- fs/btrfs/volumes.c | 13 ++++++++----- 1 files changed, 8 insertions(+), 5 deletions(-) diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c index 5dbefd1..38d93f6 100644 --- a/fs/btrfs/volumes.c +++ b/fs/btrfs/volumes.c @@ -2180,8 +2180,8 @@ static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), max_chunk_size); -again: max_avail = 0; +again: if (!map || map->num_stripes != num_stripes) { kfree(map); map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); @@ -2196,11 +2196,12 @@ again: do_div(calc_size, stripe_len); calc_size *= stripe_len; } - /* we don''t want tiny stripes */ - calc_size = max_t(u64, min_stripe_size, calc_size); - do_div(calc_size, stripe_len); - calc_size *= stripe_len; + /* we don''t want tiny stripes */ + calc_size = max_t(u64, min_stripe_size, calc_size); + + do_div(calc_size, stripe_len); + calc_size *= stripe_len; cur = fs_devices->alloc_list.next; index = 0; @@ -2217,6 +2218,8 @@ again: */ if (!looped) min_free += 1024 * 1024; + else + min_free = min_t(u64, max_avail, min_free); INIT_LIST_HEAD(&private_devs); while (index < num_stripes) { -- 1.6.3.3 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Giuseppe Scrivano
2009-Aug-03 20:53 UTC
Re: [PATCH] Btrfs: don''t allocate more than max_available bytes when looped
Hi Chris, Thanks for your comments, I changed the patch according to them. From my tests, I can see that it helps to get 10% more usable space on a 256MB partition and 5% on a 750MB one. Cheers, Giuseppe From 4eff62a7132018f8685cdeff3aa439089c708258 Mon Sep 17 00:00:00 2001 From: Giuseppe Scrivano <gscrivano@gnu.org> Date: Mon, 3 Aug 2009 22:26:43 +0200 Subject: [PATCH] Btrfs: reduce the default min_stripe_size The default min_stripe_size used by __btrfs_alloc_chunk was reduced to 32MB for data and 16MB for metadata chunks. Also, after a failed attempt to alloc data, try again reducing the minimum size by a factor of 4. It may help allocating space that wouldn''t otherwise. --- fs/btrfs/volumes.c | 7 +++++-- 1 files changed, 5 insertions(+), 2 deletions(-) diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c index 5dbefd1..a39b651 100644 --- a/fs/btrfs/volumes.c +++ b/fs/btrfs/volumes.c @@ -2166,10 +2166,10 @@ static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, if (type & BTRFS_BLOCK_GROUP_DATA) { max_chunk_size = 10 * calc_size; - min_stripe_size = 64 * 1024 * 1024; + min_stripe_size = 32 * 1024 * 1024; } else if (type & BTRFS_BLOCK_GROUP_METADATA) { max_chunk_size = 4 * calc_size; - min_stripe_size = 32 * 1024 * 1024; + min_stripe_size = 16 * 1024 * 1024; } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { calc_size = 8 * 1024 * 1024; max_chunk_size = calc_size * 2; @@ -2190,6 +2190,9 @@ again: map->num_stripes = num_stripes; } + if (looped) + min_stripe_size = min_stripe_size / 4; + if (calc_size * num_stripes > max_chunk_size) { calc_size = max_chunk_size; do_div(calc_size, num_stripes); -- 1.6.3.3 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html