[v2,01/11] btrfs: fix race with memory mapped writes when activating swap file

Commit Message

Filipe Manana Dec. 11, 2024, 3:04 p.m. UTC

From: Filipe Manana <[e-mail]>

When activating the swap file we flush all delalloc and wait for ordered
extent completion, so that we don't miss any delalloc and extents before
we check that the file's extent layout is usable for a swap file and
activate the swap file. We are called with the inode's VFS lock acquired,
so we won't race with buffered and direct IO writes, however we can still
race with memory mapped writes since they don't acquire the inode's VFS
lock. The race window is between flushing all delalloc and locking the
whole file's extent range, since memory mapped writes lock an extent range
with the length of a page.

Fix this by acquiring the inode's mmap lock before we flush delalloc.

Signed-off-by: Filipe Manana <[e-mail]>
---
 fs/btrfs/inode.c | 31 ++++++++++++++++++++++++-------
 1 file changed, 24 insertions(+), 7 deletions(-)

Patch

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index c4997200dbb2..926d82fbdbae 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -9809,6 +9809,15 @@  static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
 	u64 isize;
 	u64 start;
 
+	/*
+	 * Acquire the inode's mmap lock to prevent races with memory mapped
+	 * writes, as they could happen after we flush delalloc below and before
+	 * we lock the extent range further below. The inode was already locked
+	 * up in the call chain.
+	 */
+	btrfs_assert_inode_locked(BTRFS_I(inode));
+	down_write(&BTRFS_I(inode)->i_mmap_lock);
+
 	/*
 	 * If the swap file was just created, make sure delalloc is done. If the
 	 * file changes again after this, the user is doing something stupid and
@@ -9816,22 +9825,25 @@  static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
 	 */
 	ret = btrfs_wait_ordered_range(BTRFS_I(inode), 0, (u64)-1);
 	if (ret)
-		return ret;
+		goto out_unlock_mmap;
 
 	/*
 	 * The inode is locked, so these flags won't change after we check them.
 	 */
 	if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
 		btrfs_warn(fs_info, "swapfile must not be compressed");
-		return -EINVAL;
+		ret = -EINVAL;
+		goto out_unlock_mmap;
 	}
 	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
 		btrfs_warn(fs_info, "swapfile must not be copy-on-write");
-		return -EINVAL;
+		ret = -EINVAL;
+		goto out_unlock_mmap;
 	}
 	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
 		btrfs_warn(fs_info, "swapfile must not be checksummed");
-		return -EINVAL;
+		ret = -EINVAL;
+		goto out_unlock_mmap;
 	}
 
 	/*
@@ -9846,7 +9858,8 @@  static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
 	if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
 		btrfs_warn(fs_info,
 	   "cannot activate swapfile while exclusive operation is running");
-		return -EBUSY;
+		ret = -EBUSY;
+		goto out_unlock_mmap;
 	}
 
 	/*
@@ -9860,7 +9873,8 @@  static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
 		btrfs_exclop_finish(fs_info);
 		btrfs_warn(fs_info,
 	   "cannot activate swapfile because snapshot creation is in progress");
-		return -EINVAL;
+		ret = -EINVAL;
+		goto out_unlock_mmap;
 	}
 	/*
 	 * Snapshots can create extents which require COW even if NODATACOW is
@@ -9881,7 +9895,8 @@  static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
 		btrfs_warn(fs_info,
 		"cannot activate swapfile because subvolume %llu is being deleted",
 			btrfs_root_id(root));
-		return -EPERM;
+		ret = -EPERM;
+		goto out_unlock_mmap;
 	}
 	atomic_inc(&root->nr_swapfiles);
 	spin_unlock(&root->root_item_lock);
@@ -10036,6 +10051,8 @@  static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
 
 	btrfs_exclop_finish(fs_info);
 
+out_unlock_mmap:
+	up_write(&BTRFS_I(inode)->i_mmap_lock);
 	if (ret)
 		return ret;