On 5/9/23 16:01, Hao Xu wrote:
From: Hao Xu <howeyxu@xxxxxxxxxxx>
In FOPEN_DIRECT_IO, page cache may still be there for a file, direct
write should respect that and invalidate the corresponding pages so
that page cache readers don't get stale data. Another thing this patch
does is flush related pages to avoid its loss.
Signed-off-by: Hao Xu <howeyxu@xxxxxxxxxxx>
---
Reference:
https://lore.kernel.org/linux-fsdevel/ee8380b3-683f-c526-5f10-1ce2ee6f79ad@xxxxxxxxx/#:~:text=I%20think%20this%20problem%20exists%20before%20this%20patchset
fs/fuse/file.c | 14 +++++++++++++-
1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 89d97f6188e0..edc84c1dfc5c 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -1490,7 +1490,8 @@ ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
int write = flags & FUSE_DIO_WRITE;
int cuse = flags & FUSE_DIO_CUSE;
struct file *file = io->iocb->ki_filp;
- struct inode *inode = file->f_mapping->host;
+ struct address_space *mapping = file->f_mapping;
+ struct inode *inode = mapping->host;
struct fuse_file *ff = file->private_data;
struct fuse_conn *fc = ff->fm->fc;
size_t nmax = write ? fc->max_write : fc->max_read;
@@ -1516,6 +1517,17 @@ ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
inode_unlock(inode);
}
+ res = filemap_write_and_wait_range(mapping, pos, pos + count - 1);
Seems We don't need to flush dirty page if the page is only private
mmaped, because the pages are always clean. I'll fix this in v2.
+ if (res)
+ return res;
+
+ if (write) {
+ if (invalidate_inode_pages2_range(mapping,
+ idx_from, idx_to)) {
+ return -ENOTBLK;
+ }
+ }
+
io->should_dirty = !write && user_backed_iter(iter);
while (count) {
ssize_t nres;
