btrfs: flush_space: treat return value of do_chunk_alloc properly
authorAlex Lyakas <alex@zadarastorage.com>
Sun, 6 Dec 2015 10:32:31 +0000 (12:32 +0200)
committerChris Mason <clm@fb.com>
Thu, 25 Aug 2016 10:58:18 +0000 (03:58 -0700)
commiteecba891d38051ebf7f4af6394d188a5fd151a6a
tree941a1a6626018519dab16e4373a884bb36ec05a0
parentf3bca8028bd934e96257b8bd1143e6474fe98465
btrfs: flush_space: treat return value of do_chunk_alloc properly

do_chunk_alloc returns 1 when it succeeds to allocate a new chunk.
But flush_space will not convert this to 0, and will also return 1.
As a result, reserve_metadata_bytes will think that flush_space failed,
and may potentially return this value "1" to the caller (depends how
reserve_metadata_bytes was called). The caller will also treat this as an error.
For example, btrfs_block_rsv_refill does:

int ret = -ENOSPC;
...
ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
if (!ret) {
        block_rsv_add_bytes(block_rsv, num_bytes, 0);
        return 0;
}

return ret;

So it will return -ENOSPC.

Signed-off-by: Alex Lyakas <alex@zadarastorage.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
fs/btrfs/extent-tree.c