Btrfs: Write bio checksumming outside the FS mutex

This significantly improves streaming write performance by allowing
concurrency in the data checksumming.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index ff15b8513f908d0ef1055baaac36178b0e0b7acd..b56ae19506589406e8bc3879fac7f05c2c0e7a41 100644 (file)
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -1515,7 +1515,9 @@ int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
                             u64 bytenr, int mod);
 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
                           struct btrfs_root *root, struct inode *inode,
-                          struct bio *bio);
+                          struct bio *bio, char *sums);
+int btrfs_csum_one_bio(struct btrfs_root *root,
+                      struct bio *bio, char **sums_ret);
 struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
                                          struct btrfs_root *root,
                                          struct btrfs_path *path,

diff --git a/fs/btrfs/file-item.c b/fs/btrfs/file-item.c
index 374d0245b29d469e3390eef244e7c077a36e4d6d..9259aece6ed11137a6f5c72f77cf6bbbae48b2ed 100644 (file)
--- a/fs/btrfs/file-item.c
+++ b/fs/btrfs/file-item.c
@@ -134,9 +134,36 @@ int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
        return ret;
 }
 
+int btrfs_csum_one_bio(struct btrfs_root *root,
+                      struct bio *bio, char **sums_ret)
+{
+       u32 *sums;
+       char *data;
+       struct bio_vec *bvec = bio->bi_io_vec;
+       int bio_index = 0;
+
+       sums = kmalloc(bio->bi_vcnt * BTRFS_CRC32_SIZE, GFP_NOFS);
+       if (!sums)
+               return -ENOMEM;
+       *sums_ret = (char *)sums;
+
+       while(bio_index < bio->bi_vcnt) {
+               data = kmap_atomic(bvec->bv_page, KM_USER0);
+               *sums = ~(u32)0;
+               *sums = btrfs_csum_data(root, data + bvec->bv_offset,
+                                       *sums, bvec->bv_len);
+               kunmap_atomic(data, KM_USER0);
+               btrfs_csum_final(*sums, (char *)sums);
+               sums++;
+               bio_index++;
+               bvec++;
+       }
+       return 0;
+}
+
 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
                           struct btrfs_root *root, struct inode *inode,
-                          struct bio *bio)
+                          struct bio *bio, char *sums)
 {
        u64 objectid = inode->i_ino;
        u64 offset;
@@ -150,12 +177,11 @@ int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
        struct btrfs_csum_item *item_end;
        struct extent_buffer *leaf = NULL;
        u64 csum_offset;
-       u32 csum_result;
+       u32 *sums32 = (u32 *)sums;
        u32 nritems;
        u32 ins_size;
        int bio_index = 0;
        struct bio_vec *bvec = bio->bi_io_vec;
-       char *data;
        char *eb_map;
        char *eb_token;
        unsigned long map_len;
@@ -278,15 +304,6 @@ found:
                                      btrfs_item_size_nr(leaf, path->slots[0]));
        eb_token = NULL;
 next_bvec:
-       data = kmap_atomic(bvec->bv_page, KM_USER0);
-       csum_result = ~(u32)0;
-       csum_result = btrfs_csum_data(root, data + bvec->bv_offset,
-                                     csum_result, bvec->bv_len);
-       kunmap_atomic(data, KM_USER0);
-       btrfs_csum_final(csum_result, (char *)&csum_result);
-       if (csum_result == 0) {
-               printk("csum result is 0 for inode %lu offset %Lu\n", inode->i_ino, offset);
-       }
 
        if (!eb_token ||
           (unsigned long)item  + BTRFS_CRC32_SIZE >= map_start + map_len) {
@@ -304,13 +321,14 @@ next_bvec:
        }
        if (eb_token) {
                memcpy(eb_token + ((unsigned long)item & (PAGE_CACHE_SIZE - 1)),
-                      &csum_result, BTRFS_CRC32_SIZE);
+                      sums32, BTRFS_CRC32_SIZE);
        } else {
-               write_extent_buffer(leaf, &csum_result, (unsigned long)item,
+               write_extent_buffer(leaf, sums32, (unsigned long)item,
                                    BTRFS_CRC32_SIZE);
        }
        bio_index++;
        bvec++;
+       sums32++;
        if (bio_index < bio->bi_vcnt) {
                item = (struct btrfs_csum_item *)((char *)item +
                                                  BTRFS_CRC32_SIZE);

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 457daf725d29013bb969863f4e8ff50268429640..3e1f21c632c18c43961466b2b5cd00de0cb23b39 100644 (file)
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -330,14 +330,23 @@ int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
        struct btrfs_root *root = BTRFS_I(inode)->root;
        struct btrfs_trans_handle *trans;
        int ret = 0;
+       char *sums = NULL;
+
+       ret = btrfs_csum_one_bio(root, bio, &sums);
+       BUG_ON(ret);
 
        mutex_lock(&root->fs_info->fs_mutex);
        trans = btrfs_start_transaction(root, 1);
+
        btrfs_set_trans_block_group(trans, inode);
-       btrfs_csum_file_blocks(trans, root, inode, bio);
+       btrfs_csum_file_blocks(trans, root, inode, bio, sums);
+
        ret = btrfs_end_transaction(trans, root);
        BUG_ON(ret);
        mutex_unlock(&root->fs_info->fs_mutex);
+
+       kfree(sums);
+
        return btrfs_map_bio(root, rw, bio, mirror_num);
 }