[Virtio-fs] [PATCH-v2 0/2] fuse: remove dmap when truncating inode

[Virtio-fs] [PATCH-v2 0/2] fuse: remove dmap when truncating inode

[Peng Tao]

Hi Vivek,

The two patches try to remove dmap during truncation. virtiofs fuse
protocol is also modified to support removing multiple entries in
a single request. The host side virtiofsd patch will be sent separately.

Cheers,
Tao


Peng Tao (2):
  fuse: remove dmap when truncating inode
  virtiofs: FUSE_REMOVEMAPPING support multiple removing multiple
    entries

 fs/fuse/dir.c             |   5 ++
 fs/fuse/file.c            | 134 ++++++++++++++++++++++++++++++++------
 fs/fuse/fuse_i.h          |   2 +
 include/uapi/linux/fuse.h |   9 ++-
 4 files changed, 129 insertions(+), 21 deletions(-)

-- 
2.17.1

[Virtio-fs] [PATCH-v2 1/2] fuse: remove dmap when truncating inode

[Peng Tao]

The obseleted dmap entries can be put back to global free list
immediately and we don't have to rely on reclaim code to free them.

Signed-off-by: Peng Tao <tao.peng@linux.alibaba.com>
---
 fs/fuse/dir.c    |  5 +++++
 fs/fuse/file.c   | 55 ++++++++++++++++++++++++++++++++++++++----------
 fs/fuse/fuse_i.h |  2 ++
 3 files changed, 51 insertions(+), 11 deletions(-)

diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index 3f923fe7841a..233c3ed391f1 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -1751,6 +1751,11 @@ int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
 		down_write(&fi->i_mmap_sem);
 		truncate_pagecache(inode, outarg.attr.size);
 		invalidate_inode_pages2(inode->i_mapping);
+		// Free dmap beyond i_size
+		if (IS_DAX(inode) && oldsize > outarg.attr.size)
+			fuse_dax_free_mappings_range(fc, inode,
+						     outarg.attr.size, -1);
+
 		up_write(&fi->i_mmap_sem);
 	}
 
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 7362aab3ee74..b9f6688c4062 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -3848,7 +3848,6 @@ int fuse_dax_reclaim_dmap_locked(struct fuse_conn *fc, struct inode *inode,
 		return ret;
 	}
 
-	/* Remove dax mapping from inode interval tree now */
 	fuse_dax_interval_tree_remove(dmap, &fi->dmap_tree);
 	fi->nr_dmaps--;
 	return 0;
@@ -3927,6 +3926,49 @@ static struct fuse_dax_mapping *alloc_dax_mapping_reclaim(struct fuse_conn *fc,
 	}
 }
 
+/* Cleanup dmap entry and add back to free list */
+static void fuse_dax_do_free_mapping_locked(struct fuse_conn *fc, struct fuse_dax_mapping *dmap)
+{
+	pr_debug("fuse: freed memory range start=0x%llx end=0x%llx "
+		"window_offset=0x%llx length=0x%llx\n", dmap->start,
+		dmap->end, dmap->window_offset, dmap->length);
+	spin_lock(&fc->lock);
+	__dmap_remove_busy_list(fc, dmap);
+	dmap->inode = NULL;
+	dmap->start = dmap->end = 0;
+	__free_dax_mapping(fc, dmap);
+	spin_unlock(&fc->lock);
+}
+
+/*
+ * Free inode dmap entries whose range falls entirely inside [start, end].
+ * Called with inode->i_rwsem and fuse_inode->i_mmap_sem held.
+ */
+void fuse_dax_free_mappings_range(struct fuse_conn *fc, struct inode *inode, loff_t start, loff_t end)
+{
+	struct fuse_inode *fi = get_fuse_inode(inode);
+	struct fuse_dax_mapping *dmap;
+
+	WARN_ON(!inode_is_locked(inode));
+	WARN_ON(!rwsem_is_locked(&fi->i_mmap_sem));
+
+	/* interval tree search matches intersecting entries.
+	 * Adjust the range to avoid dropping partial valid entries. */
+	start = ALIGN(start, FUSE_DAX_MEM_RANGE_SZ);
+	end = ALIGN_DOWN(end, FUSE_DAX_MEM_RANGE_SZ);
+
+	pr_debug("fuse: fuse_dax_free_mappings_range start=0x%llx, end=0x%llx\n", start, end);
+	/* Lock ordering follows fuse_dax_free_one_mapping() */
+	down_write(&fi->i_dmap_sem);
+	while ((dmap = fuse_dax_interval_tree_iter_first(&fi->dmap_tree, start, end))) {
+		fuse_dax_interval_tree_remove(dmap, &fi->dmap_tree);
+		fi->nr_dmaps--;
+		fuse_dax_do_free_mapping_locked(fc, dmap);
+		fuse_removemapping_one(inode, dmap);
+	}
+	up_write(&fi->i_dmap_sem);
+}
+
 int fuse_dax_free_one_mapping_locked(struct fuse_conn *fc, struct inode *inode,
 				u64 dmap_start)
 {
@@ -3948,17 +3990,8 @@ int fuse_dax_free_one_mapping_locked(struct fuse_conn *fc, struct inode *inode,
 	if (ret < 0)
 		return ret;
 
-	/* Cleanup dmap entry and add back to free list */
-	spin_lock(&fc->lock);
-	__dmap_remove_busy_list(fc, dmap);
-	dmap->inode = NULL;
-	dmap->start = dmap->end = 0;
-	__free_dax_mapping(fc, dmap);
-	spin_unlock(&fc->lock);
+	fuse_dax_do_free_mapping_locked(fc, dmap);
 
-	pr_debug("fuse: freed memory range window_offset=0x%llx,"
-				" length=0x%llx\n", dmap->window_offset,
-				dmap->length);
 	return ret;
 }
 
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index 20a8bfd63b80..894fc5e78589 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -1206,5 +1206,7 @@ void fuse_dax_free_mem_worker(struct work_struct *work);
 void fuse_removemapping(struct inode *inode);
 void fuse_end_pending_request(struct fuse_conn *fc, struct fuse_pqueue *fpq,
 			      u64 unique, int err);
+void fuse_dax_free_mappings_range(struct fuse_conn *fc, struct inode *inode,
+			loff_t start, loff_t end);
 
 #endif /* _FS_FUSE_I_H */
-- 
2.17.1

[Virtio-fs] [PATCH-v2 2/2] virtiofs: FUSE_REMOVEMAPPING support multiple removing multiple entries

[Peng Tao]

Change FUSE_REMOVEMAPPING wire protocol so that we can remove
multiple mappings in one call.

Signed-off-by: Peng Tao <tao.peng@linux.alibaba.com>
---
 fs/fuse/file.c            | 85 +++++++++++++++++++++++++++++++++------
 include/uapi/linux/fuse.h |  9 ++++-
 2 files changed, 81 insertions(+), 13 deletions(-)

diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index b9f6688c4062..41ec7fdb780e 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -350,25 +350,80 @@ static int fuse_mkwrite_one_mapping(struct inode *inode, struct file *file,
 	return __fuse_setup_one_mapping(inode, file, offset, flags, dmap, true);
 }
 
-static int fuse_removemapping_one(struct inode *inode,
-					struct fuse_dax_mapping *dmap)
+static int fuse_send_removemapping_request(struct inode *inode,
+		struct fuse_removemapping_in_header *header,
+		struct fuse_removemapping_in *inargp)
 {
 	struct fuse_inode *fi = get_fuse_inode(inode);
 	struct fuse_conn *fc = get_fuse_conn(inode);
-	struct fuse_removemapping_in inarg;
 	FUSE_ARGS(args);
 
-	memset(&inarg, 0, sizeof(inarg));
-	inarg.moffset = dmap->window_offset;
-	inarg.len = dmap->length;
 	args.in.h.opcode = FUSE_REMOVEMAPPING;
 	args.in.h.nodeid = fi->nodeid;
-	args.in.numargs = 1;
-	args.in.args[0].size = sizeof(inarg);
-	args.in.args[0].value = &inarg;
+	args.in.numargs = 2;
+	args.in.args[0].size = sizeof(*header);
+	args.in.args[0].value = header;
+	args.in.args[1].size = header->num * sizeof(*inargp);
+	args.in.args[1].value = inargp;
 	return fuse_simple_request(fc, &args);
 }
 
+static int fuse_removemappings(struct inode *inode, unsigned num,
+				struct list_head *to_remove)
+{
+	struct fuse_removemapping_in *inargp, *ptr;
+	struct fuse_removemapping_in_header header;
+	struct fuse_dax_mapping *dmap;
+	int ret, i = 0;
+
+	if (num <= FUSE_REMOVEMAPPING_MAX_ENTRY) {
+		inargp = kmalloc_array(num, sizeof(*inargp), GFP_NOIO);
+	} else {
+		inargp = kmalloc_array(FUSE_REMOVEMAPPING_MAX_ENTRY,
+					sizeof(*inargp), GFP_NOIO);
+	}
+	if (inargp == NULL)
+		return -ENOMEM;
+
+	ptr = inargp;
+	list_for_each_entry(dmap, to_remove, list) {
+		ptr->moffset = dmap->window_offset;
+		ptr->len = dmap->length;
+		ptr++;
+		i++;
+		num--;
+		if (i >= FUSE_REMOVEMAPPING_MAX_ENTRY || num == 0) {
+			memset(&header, 0, sizeof(header));
+			header.num = i;
+			ret = fuse_send_removemapping_request(inode, &header, inargp);
+			if (ret)
+				goto out;
+			ptr = inargp;
+			i = 0;
+		}
+	}
+
+out:
+	kfree(inargp);
+	return ret;
+}
+
+static int fuse_removemapping_one(struct inode *inode,
+					struct fuse_dax_mapping *dmap)
+{
+	struct fuse_removemapping_in inarg;
+	struct fuse_removemapping_in_header header;
+
+	memset(&header, 0, sizeof(header));
+	/* TODO: fill in header.fh when available */
+	header.num = 1;
+	memset(&inarg, 0, sizeof(inarg));
+	inarg.moffset = dmap->window_offset;
+	inarg.len = dmap->length;
+
+	return fuse_send_removemapping_request(inode, &header, &inarg);
+}
+
 /*
  * It is called from evict_inode() and by that time inode is going away. So
  * this function does not take any locks like fi->i_dmap_sem for traversing
@@ -3947,7 +4002,9 @@ static void fuse_dax_do_free_mapping_locked(struct fuse_conn *fc, struct fuse_da
 void fuse_dax_free_mappings_range(struct fuse_conn *fc, struct inode *inode, loff_t start, loff_t end)
 {
 	struct fuse_inode *fi = get_fuse_inode(inode);
-	struct fuse_dax_mapping *dmap;
+	struct fuse_dax_mapping *dmap, *n;
+	int num = 0;
+	LIST_HEAD(to_remove);
 
 	WARN_ON(!inode_is_locked(inode));
 	WARN_ON(!rwsem_is_locked(&fi->i_mmap_sem));
@@ -3962,9 +4019,13 @@ void fuse_dax_free_mappings_range(struct fuse_conn *fc, struct inode *inode, lof
 	down_write(&fi->i_dmap_sem);
 	while ((dmap = fuse_dax_interval_tree_iter_first(&fi->dmap_tree, start, end))) {
 		fuse_dax_interval_tree_remove(dmap, &fi->dmap_tree);
-		fi->nr_dmaps--;
+		num++;
+		list_add(&dmap->list, &to_remove);
+	}
+	fi->nr_dmaps -= num;
+	fuse_removemappings(inode, num, &to_remove);
+	list_for_each_entry_safe(dmap, n, &to_remove, list) {
 		fuse_dax_do_free_mapping_locked(fc, dmap);
-		fuse_removemapping_one(inode, dmap);
 	}
 	up_write(&fi->i_dmap_sem);
 }
diff --git a/include/uapi/linux/fuse.h b/include/uapi/linux/fuse.h
index dbc5013ad747..04a4204e7f98 100644
--- a/include/uapi/linux/fuse.h
+++ b/include/uapi/linux/fuse.h
@@ -817,13 +817,20 @@ struct fuse_setupmapping_out {
         uint64_t	len[FUSE_SETUPMAPPING_ENTRIES];
 };
 
-struct fuse_removemapping_in {
+struct fuse_removemapping_in_header {
         /* An already open handle */
         uint64_t	fh;
+	/* number of fuse_removemapping_in follows */
+	unsigned	num;
+};
+
+struct fuse_removemapping_in {
 	/* Offset into the dax window start the unmapping */
 	uint64_t        moffset;
         /* Length of mapping required */
         uint64_t	len;
 };
+#define FUSE_REMOVEMAPPING_MAX_ENTRY	\
+		(PAGE_SIZE / sizeof(struct fuse_removemapping_in))
 
 #endif /* _LINUX_FUSE_H */
-- 
2.17.1