1 files changed, 46 insertions, 22 deletions

diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 315f21191643..42c745dbccfa 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -3995,6 +3995,10 @@ retry:
 			if (!ret) {
 				free_extent_buffer(eb);
 				continue;
+			} else if (ret < 0) {
+				done = 1;
+				free_extent_buffer(eb);
+				break;
 			}
 
 			ret = write_one_eb(eb, fs_info, wbc, &epd);
@@ -4153,6 +4157,14 @@ retry:
 		 */
 		scanned = 1;
 		index = 0;
+
+		/*
+		 * If we're looping we could run into a page that is locked by a
+		 * writer and that writer could be waiting on writeback for a
+		 * page in our current bio, and thus deadlock, so flush the
+		 * write bio here.
+		 */
+		flush_write_bio(data);
 		goto retry;
 	}
 	btrfs_add_delayed_iput(inode);
@@ -4418,6 +4430,8 @@ int try_release_extent_mapping(struct extent_map_tree *map,
 
 			/* once for us */
 			free_extent_map(em);
+
+			cond_resched(); /* Allow large-extent preemption. */
 		}
 	}
 	return try_release_extent_state(map, tree, page, mask);
@@ -4842,25 +4856,28 @@ err:
 static void check_buffer_tree_ref(struct extent_buffer *eb)
 {
 	int refs;
-	/* the ref bit is tricky.  We have to make sure it is set
-	 * if we have the buffer dirty.   Otherwise the
-	 * code to free a buffer can end up dropping a dirty
-	 * page
+	/*
+	 * The TREE_REF bit is first set when the extent_buffer is added
+	 * to the radix tree. It is also reset, if unset, when a new reference
+	 * is created by find_extent_buffer.
 	 *
-	 * Once the ref bit is set, it won't go away while the
-	 * buffer is dirty or in writeback, and it also won't
-	 * go away while we have the reference count on the
-	 * eb bumped.
+	 * It is only cleared in two cases: freeing the last non-tree
+	 * reference to the extent_buffer when its STALE bit is set or
+	 * calling releasepage when the tree reference is the only reference.
 	 *
-	 * We can't just set the ref bit without bumping the
-	 * ref on the eb because free_extent_buffer might
-	 * see the ref bit and try to clear it.  If this happens
-	 * free_extent_buffer might end up dropping our original
-	 * ref by mistake and freeing the page before we are able
-	 * to add one more ref.
+	 * In both cases, care is taken to ensure that the extent_buffer's
+	 * pages are not under io. However, releasepage can be concurrently
+	 * called with creating new references, which is prone to race
+	 * conditions between the calls to check_buffer_tree_ref in those
+	 * codepaths and clearing TREE_REF in try_release_extent_buffer.
 	 *
-	 * So bump the ref count first, then set the bit.  If someone
-	 * beat us to it, drop the ref we added.
+	 * The actual lifetime of the extent_buffer in the radix tree is
+	 * adequately protected by the refcount, but the TREE_REF bit and
+	 * its corresponding reference are not. To protect against this
+	 * class of races, we call check_buffer_tree_ref from the codepaths
+	 * which trigger io after they set eb->io_pages. Note that once io is
+	 * initiated, TREE_REF can no longer be cleared, so that is the
+	 * moment at which any such race is best fixed.
 	 */
 	refs = atomic_read(&eb->refs);
 	if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
@@ -4937,12 +4954,14 @@ struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
 		return eb;
 	eb = alloc_dummy_extent_buffer(fs_info, start);
 	if (!eb)
-		return NULL;
+		return ERR_PTR(-ENOMEM);
 	eb->fs_info = fs_info;
 again:
 	ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
-	if (ret)
+	if (ret) {
+		exists = ERR_PTR(ret);
 		goto free_eb;
+	}
 	spin_lock(&fs_info->buffer_lock);
 	ret = radix_tree_insert(&fs_info->buffer_radix,
 				start >> PAGE_CACHE_SHIFT, eb);
@@ -5325,6 +5344,11 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
 	clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 	eb->read_mirror = 0;
 	atomic_set(&eb->io_pages, num_reads);
+	/*
+	 * It is possible for releasepage to clear the TREE_REF bit before we
+	 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
+	 */
+	check_buffer_tree_ref(eb);
 	for (i = start_i; i < num_pages; i++) {
 		page = eb->pages[i];
 		if (!PageUptodate(page)) {
@@ -5400,9 +5424,9 @@ void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
 	}
 }
 
-int read_extent_buffer_to_user(const struct extent_buffer *eb,
-			       void __user *dstv,
-			       unsigned long start, unsigned long len)
+int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
+				       void __user *dstv,
+				       unsigned long start, unsigned long len)
 {
 	size_t cur;
 	size_t offset;
@@ -5423,7 +5447,7 @@ int read_extent_buffer_to_user(const struct extent_buffer *eb,
 
 		cur = min(len, (PAGE_CACHE_SIZE - offset));
 		kaddr = page_address(page);
-		if (copy_to_user(dst, kaddr + offset, cur)) {
+		if (probe_user_write(dst, kaddr + offset, cur)) {
 			ret = -EFAULT;
 			break;
 		}