Btrfs: pass fs_info to btrfs_map_block() instead of mapping_tree
[deliverable/linux.git] / fs / btrfs / scrub.c
index a67b1a17a0092b9a049fbb9de72dffbc0a1ef04e..894bb2732fcc55ab192d732045ee79033a76afda 100644 (file)
@@ -152,7 +152,7 @@ static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
 static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
-                                    struct btrfs_mapping_tree *map_tree,
+                                    struct btrfs_fs_info *fs_info,
                                     u64 length, u64 logical,
                                     struct scrub_block *sblock);
 static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
@@ -523,7 +523,7 @@ static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *ctx)
        }
 
        if (PageUptodate(page)) {
-               struct btrfs_mapping_tree *map_tree;
+               struct btrfs_fs_info *fs_info;
                if (PageDirty(page)) {
                        /*
                         * we need to write the data to the defect sector. the
@@ -544,8 +544,8 @@ static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *ctx)
                        ret = -EIO;
                        goto out;
                }
-               map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
-               ret = repair_io_failure(map_tree, offset, PAGE_SIZE,
+               fs_info = BTRFS_I(inode)->root->fs_info;
+               ret = repair_io_failure(fs_info, offset, PAGE_SIZE,
                                        fixup->logical, page,
                                        fixup->mirror_num);
                unlock_page(page);
@@ -754,7 +754,7 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
        }
 
        /* setup the context, map the logical blocks and alloc the pages */
-       ret = scrub_setup_recheck_block(sctx, &fs_info->mapping_tree, length,
+       ret = scrub_setup_recheck_block(sctx, fs_info, length,
                                        logical, sblocks_for_recheck);
        if (ret) {
                spin_lock(&sctx->stat_lock);
@@ -1012,7 +1012,7 @@ out:
 }
 
 static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
-                                    struct btrfs_mapping_tree *map_tree,
+                                    struct btrfs_fs_info *fs_info,
                                     u64 length, u64 logical,
                                     struct scrub_block *sblocks_for_recheck)
 {
@@ -1036,7 +1036,7 @@ static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
                 * with a length of PAGE_SIZE, each returned stripe
                 * represents one mirror
                 */
-               ret = btrfs_map_block(map_tree, WRITE, logical, &mapped_length,
+               ret = btrfs_map_block(fs_info, WRITE, logical, &mapped_length,
                                      &bbio, 0);
                if (ret || !bbio || mapped_length < sublen) {
                        kfree(bbio);
This page took 0.026177 seconds and 5 git commands to generate.