Btrfs: make sure to flush queued bios if write_cache_pages waits

write_cache_pages tries to build up a large bio to stuff down the pipe.
But if it needs to wait for a page lock, it needs to make sure and send
down any pending writes so we don't deadlock with anyone who has the
page lock and is waiting for writeback of things inside the bio.

Dave Sterba triggered this as a deadlock between the autodefrag code and
the extent write_cache_pages

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

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 51372a5..6f58911 100644 (file)
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -2735,7 +2735,8 @@ int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
        return ret;
 }
 
-int btree_lock_page_hook(struct page *page)
+static int btree_lock_page_hook(struct page *page, void *data,
+                               void (*flush_fn)(void *))
 {
        struct inode *inode = page->mapping->host;
        struct btrfs_root *root = BTRFS_I(inode)->root;
@@ -2752,7 +2753,10 @@ int btree_lock_page_hook(struct page *page)
        if (!eb)
                goto out;
 
-       btrfs_tree_lock(eb);
+       if (!btrfs_try_tree_write_lock(eb)) {
+               flush_fn(data);
+               btrfs_tree_lock(eb);
+       }
        btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 
        if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
@@ -2767,7 +2771,10 @@ int btree_lock_page_hook(struct page *page)
        btrfs_tree_unlock(eb);
        free_extent_buffer(eb);
 out:
-       lock_page(page);
+       if (!trylock_page(page)) {
+               flush_fn(data);
+               lock_page(page);
+       }
        return 0;
 }
 

diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index bec3ea4..e678539 100644 (file)
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -83,8 +83,6 @@ int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
                             struct btrfs_fs_info *fs_info);
 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
                       struct btrfs_root *root);
-int btree_lock_page_hook(struct page *page);
-
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 void btrfs_init_lockdep(void);

diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index f284d4e..b40ba75 100644 (file)
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -2613,10 +2613,16 @@ retry:
                         * swizzled back from swapper_space to tmpfs file
                         * mapping
                         */
-                       if (tree->ops && tree->ops->write_cache_pages_lock_hook)
-                               tree->ops->write_cache_pages_lock_hook(page);
-                       else
-                               lock_page(page);
+                       if (tree->ops &&
+                           tree->ops->write_cache_pages_lock_hook) {
+                               tree->ops->write_cache_pages_lock_hook(page,
+                                                              data, flush_fn);
+                       } else {
+                               if (!trylock_page(page)) {
+                                       flush_fn(data);
+                                       lock_page(page);
+                               }
+                       }
 
                        if (unlikely(page->mapping != mapping)) {
                                unlock_page(page);

diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 325a346..cbd4824 100644 (file)
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -86,7 +86,8 @@ struct extent_io_ops {
                                  struct extent_state *other);
        void (*split_extent_hook)(struct inode *inode,
                                  struct extent_state *orig, u64 split);
-       int (*write_cache_pages_lock_hook)(struct page *page);
+       int (*write_cache_pages_lock_hook)(struct page *page, void *data,
+                                          void (*flush_fn)(void *));
 };
 
 struct extent_io_tree {