nfs: don't call __mark_inode_dirty while holding i_lock

nfs_scan_commit() is called with the inode->i_lock held, but it then
calls __mark_inode_dirty() while still holding the lock. This causes
a deadlock.

Push the inode->i_lock into nfs_scan_commit() so it can protect only
the parts of the code it needs to and can be dropped before the call
to __mark_inode_dirty() to avoid the deadlock.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Tested-by: NWill Simoneau <simoneau@ele.uri.edu>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

fs/nfs/write.c

@@ -542,11 +542,15 @@ nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, u
 	if (!nfs_need_commit(nfsi))
 		return 0;
 
+	spin_lock(&inode->i_lock);
 	ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT);
 	if (ret > 0)
 		nfsi->ncommit -= ret;
+	spin_unlock(&inode->i_lock);
+
 	if (nfs_need_commit(NFS_I(inode)))
 		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
+
 	return ret;
 }
 #else
@@ -1483,9 +1487,7 @@ int nfs_commit_inode(struct inode *inode, int how)
 	res = nfs_commit_set_lock(NFS_I(inode), may_wait);
 	if (res <= 0)
 		goto out_mark_dirty;
-	spin_lock(&inode->i_lock);
 	res = nfs_scan_commit(inode, &head, 0, 0);
-	spin_unlock(&inode->i_lock);
 	if (res) {
 		int error;