ceph: disable fscache when inode is opened for write

All other filesystems do not add dirty pages to fscache. They all
disable fscache when inode is opened for write. Only ceph adds
dirty pages to fscache, but the code is buggy.

Signed-off-by: Yan, Zheng <zyan@redhat.com>

4 files changed, 52 insertions, 53 deletions

diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
index 4ff62fc..26a9d10 100644
--- a/fs/ceph/addr.c
+++ b/fs/ceph/addr.c
@@ -537,8 +537,6 @@ static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
 	    CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
 		set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
 
-	ceph_readpage_to_fscache(inode, page);
-
 	set_page_writeback(page);
 	err = ceph_osdc_writepages(osdc, ceph_vino(inode),
 				   &ci->i_layout, snapc,
diff --git a/fs/ceph/cache.c b/fs/ceph/cache.c
index c052b5b..c19db6a 100644
--- a/fs/ceph/cache.c
+++ b/fs/ceph/cache.c
@@ -181,32 +181,26 @@ static const struct fscache_cookie_def ceph_fscache_inode_object_def = {
 	.now_uncached	= ceph_fscache_inode_now_uncached,
 };
 
-void ceph_fscache_register_inode_cookie(struct ceph_fs_client* fsc,
-					struct ceph_inode_info* ci)
+void ceph_fscache_register_inode_cookie(struct inode *inode)
 {
-	struct inode* inode = &ci->vfs_inode;
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 
 	/* No caching for filesystem */
 	if (fsc->fscache == NULL)
 		return;
 
 	/* Only cache for regular files that are read only */
-	if ((ci->vfs_inode.i_mode & S_IFREG) == 0)
+	if (!S_ISREG(inode->i_mode))
 		return;
 
-	/* Avoid multiple racing open requests */
-	inode_lock(inode);
-
-	if (ci->fscache)
-		goto done;
-
-	ci->fscache = fscache_acquire_cookie(fsc->fscache,
-					     &ceph_fscache_inode_object_def,
-					     ci, true);
-	fscache_check_consistency(ci->fscache);
-done:
+	inode_lock_nested(inode, I_MUTEX_CHILD);
+	if (!ci->fscache) {
+		ci->fscache = fscache_acquire_cookie(fsc->fscache,
+					&ceph_fscache_inode_object_def,
+					ci, false);
+	}
 	inode_unlock(inode);
-
 }
 
 void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci)
@@ -222,6 +216,34 @@ void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci)
 	fscache_relinquish_cookie(cookie, 0);
 }
 
+static bool ceph_fscache_can_enable(void *data)
+{
+	struct inode *inode = data;
+	return !inode_is_open_for_write(inode);
+}
+
+void ceph_fscache_file_set_cookie(struct inode *inode, struct file *filp)
+{
+	struct ceph_inode_info *ci = ceph_inode(inode);
+
+	if (!fscache_cookie_valid(ci->fscache))
+		return;
+
+	if (inode_is_open_for_write(inode)) {
+		dout("fscache_file_set_cookie %p %p disabling cache\n",
+		     inode, filp);
+		fscache_disable_cookie(ci->fscache, false);
+		fscache_uncache_all_inode_pages(ci->fscache, inode);
+	} else {
+		fscache_enable_cookie(ci->fscache, ceph_fscache_can_enable,
+				inode);
+		if (fscache_cookie_enabled(ci->fscache)) {
+			dout("fscache_file_set_cookie %p %p enabing cache\n",
+			     inode, filp);
+		}
+	}
+}
+
 static void ceph_vfs_readpage_complete(struct page *page, void *data, int error)
 {
 	if (!error)
diff --git a/fs/ceph/cache.h b/fs/ceph/cache.h
index 5ac591b..dfe9b58 100644
--- a/fs/ceph/cache.h
+++ b/fs/ceph/cache.h
@@ -35,9 +35,9 @@ int ceph_fscache_register_fs(struct ceph_fs_client* fsc);
 void ceph_fscache_unregister_fs(struct ceph_fs_client* fsc);
 
 void ceph_fscache_inode_init(struct ceph_inode_info *ci);
-void ceph_fscache_register_inode_cookie(struct ceph_fs_client* fsc,
-					struct ceph_inode_info* ci);
+void ceph_fscache_register_inode_cookie(struct inode *inode);
 void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci);
+void ceph_fscache_file_set_cookie(struct inode *inode, struct file *filp);
 
 int ceph_readpage_from_fscache(struct inode *inode, struct page *page);
 int ceph_readpages_from_fscache(struct inode *inode,
@@ -48,12 +48,6 @@ void ceph_readpage_to_fscache(struct inode *inode, struct page *page);
 void ceph_invalidate_fscache_page(struct inode* inode, struct page *page);
 void ceph_queue_revalidate(struct inode *inode);
 
-static inline void ceph_fscache_update_objectsize(struct inode *inode)
-{
-	struct ceph_inode_info *ci = ceph_inode(inode);
-	fscache_attr_changed(ci->fscache);
-}
-
 static inline void ceph_fscache_invalidate(struct inode *inode)
 {
 	fscache_invalidate(ceph_inode(inode)->fscache);
@@ -112,8 +106,16 @@ static inline void ceph_fscache_inode_init(struct ceph_inode_info *ci)
 {
 }
 
-static inline void ceph_fscache_register_inode_cookie(struct ceph_fs_client* parent_fsc,
-						      struct ceph_inode_info* ci)
+static inline void ceph_fscache_register_inode_cookie(struct inode *inode)
+{
+}
+
+static inline void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci)
+{
+}
+
+static inline void ceph_fscache_file_set_cookie(struct inode *inode,
+						struct file *filp)
 {
 }
 
@@ -141,10 +143,6 @@ static inline void ceph_readpage_to_fscache(struct inode *inode,
 {
 }
 
-static inline void ceph_fscache_update_objectsize(struct inode *inode)
-{
-}
-
 static inline void ceph_fscache_invalidate(struct inode *inode)
 {
 }
@@ -154,10 +152,6 @@ static inline void ceph_invalidate_fscache_page(struct inode *inode,
 {
 }
 
-static inline void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci)
-{
-}
-
 static inline int ceph_release_fscache_page(struct page *page, gfp_t gfp)
 {
 	return 1;
diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index 8eeb9f5..ce2f579 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -137,23 +137,11 @@ static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
 {
 	struct ceph_file_info *cf;
 	int ret = 0;
-	struct ceph_inode_info *ci = ceph_inode(inode);
-	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
-	struct ceph_mds_client *mdsc = fsc->mdsc;
 
 	switch (inode->i_mode & S_IFMT) {
 	case S_IFREG:
-		/* First file open request creates the cookie, we want to keep
-		 * this cookie around for the filetime of the inode as not to
-		 * have to worry about fscache register / revoke / operation
-		 * races.
-		 *
-		 * Also, if we know the operation is going to invalidate data
-		 * (non readonly) just nuke the cache right away.
-		 */
-		ceph_fscache_register_inode_cookie(mdsc->fsc, ci);
-		if ((fmode & CEPH_FILE_MODE_WR))
-			ceph_fscache_invalidate(inode);
+		ceph_fscache_register_inode_cookie(inode);
+		ceph_fscache_file_set_cookie(inode, file);
 	case S_IFDIR:
 		dout("init_file %p %p 0%o (regular)\n", inode, file,
 		     inode->i_mode);
@@ -1407,7 +1395,6 @@ retry_snap:
 			iov_iter_advance(from, written);
 		ceph_put_snap_context(snapc);
 	} else {
-		loff_t old_size = i_size_read(inode);
 		/*
 		 * No need to acquire the i_truncate_mutex. Because
 		 * the MDS revokes Fwb caps before sending truncate
@@ -1418,8 +1405,6 @@ retry_snap:
 		written = generic_perform_write(file, from, pos);
 		if (likely(written >= 0))
 			iocb->ki_pos = pos + written;
-		if (i_size_read(inode) > old_size)
-			ceph_fscache_update_objectsize(inode);
 		inode_unlock(inode);
 	}