[Virtio-fs] [PATCH 1/2] Virtio-fs: fix hang due to ENOSPC in shared backend fs

[Liu Bo]

Currently fuse/virtio-fs de-allocation doesn't clean up dax mapping range,
which may result in hang problems when the shared backend fs experiences
"NO Space Error".

The root cause is that the first writing to a dax mapping range triggers a
WRITE page fault on host side, which calls ->page_mkwrite() where block
allocation is required, if the fs is already full, ->page_mkwrite() returns
error so that page fault fails, however, for kvm is not able to propogate
errors while handling EPT_VIOLATION, thus guest keeps trying to resolve the
fault.

Fortunately, we can fix/work around the problem by dropping dax mapping
range for de-allocation operations.

Signed-off-by: Liu Bo <bo.liu at linux.alibaba.com>
---
 fs/fuse/dir.c    | 3 +++
 fs/fuse/file.c   | 9 +++++++--
 fs/fuse/fuse_i.h | 2 +-
 fs/fuse/inode.c  | 2 +-
 4 files changed, 12 insertions(+), 4 deletions(-)

diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index ed740a5..99a218c 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -1805,6 +1805,9 @@ int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
 
 		truncate_pagecache(inode, outarg.attr.size);
 		invalidate_inode_pages2(inode->i_mapping);
+		if (IS_DAX(inode) && oldsize > outarg.attr.size)
+			fuse_cleanup_inode_mappings(inode, outarg.attr.size,
+						    (loff_t)-1);
 		up_write(&fi->i_mmap_sem);
 	}
 
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index c52260c..4f2d908 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -417,6 +417,7 @@ static void inode_reclaim_dmap_range(struct fuse_conn *fc, struct inode *inode,
 	start = ALIGN(start, FUSE_DAX_MEM_RANGE_SZ);
 	end = ALIGN_DOWN(end, FUSE_DAX_MEM_RANGE_SZ);
 
+	down_write(&fi->i_dmap_sem);
 	while (1) {
 		dmap = fuse_dax_interval_tree_iter_first(&fi->dmap_tree, start,
 							 end);
@@ -426,6 +427,7 @@ static void inode_reclaim_dmap_range(struct fuse_conn *fc, struct inode *inode,
 		num++;
 		list_add(&dmap->list, &to_remove);
 	}
+	up_write(&fi->i_dmap_sem);
 
 	/* Nothing to remove */
 	if (list_empty(&to_remove))
@@ -478,7 +480,7 @@ static int dmap_removemapping_one(struct inode *inode,
  * that fuse inode interval tree. If that lock is taken then lock validator
  * complains of deadlock situation w.r.t fs_reclaim lock.
  */
-void fuse_cleanup_inode_mappings(struct inode *inode)
+void fuse_cleanup_inode_mappings(struct inode *inode, loff_t start, loff_t end)
 {
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	/*
@@ -486,7 +488,7 @@ void fuse_cleanup_inode_mappings(struct inode *inode)
 	 * before we arrive here. So we should not have to worry about
 	 * any pages/exception entries still associated with inode.
 	 */
-	inode_reclaim_dmap_range(fc, inode, 0, -1);
+	inode_reclaim_dmap_range(fc, inode, start, end);
 }
 
 void fuse_finish_open(struct inode *inode, struct file *file)
@@ -3867,6 +3869,9 @@ static long __fuse_file_fallocate(struct file *file, int mode,
 		}
 
 		truncate_pagecache_range(inode, offset, offset + length - 1);
+		if (IS_DAX(inode))
+			fuse_cleanup_inode_mappings(inode, offset,
+						    offset + length - 1);
 		up_write(&fi->i_mmap_sem);
 	}
 	fuse_invalidate_attr(inode);
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index 6956b62..6104d61 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -1213,6 +1213,6 @@ ssize_t fuse_getxattr(struct inode *inode, const char *name, void *value,
  */
 u64 fuse_get_unique(struct fuse_iqueue *fiq);
 void fuse_dax_free_mem_worker(struct work_struct *work);
-void fuse_cleanup_inode_mappings(struct inode *inode);
+void fuse_cleanup_inode_mappings(struct inode *inode, loff_t start, loff_t end);
 
 #endif /* _FS_FUSE_I_H */
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index e0d792b..629c1a7 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -125,7 +125,7 @@ static void fuse_evict_inode(struct inode *inode)
 		struct fuse_conn *fc = get_fuse_conn(inode);
 		struct fuse_inode *fi = get_fuse_inode(inode);
 		if (IS_DAX(inode)) {
-			fuse_cleanup_inode_mappings(inode);
+			fuse_cleanup_inode_mappings(inode, 0, (loff_t)-1);
 			WARN_ON(fi->nr_dmaps);
 		}
 		fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
-- 
1.8.3.1