When a dentry exists for the path argument to an open with O_CREAT or
a negative dentry exists for the path argument to any open, make a
lookup request to the user-space daemon to verify the
existence/non-existence of the path.
This improves file open behavior for a FUSE filesystem where changes
may be made without going through the mount point, such as a
distributed filesystem accessed concurrently from multiple hosts.
Specifically:
- For an open with O_CREAT of a path with a cached dentry, the
user-space daemon is able to report a recent unlink of a file
allowing it to be re-created rather than either the open failing
with EEXIST (when O_EXCL is used) or a FUSE open request causing the
open to fail with ENOENT (when O_EXCL is not used).
- For an open of a path with a cached negative dentry, the user-space
daemon is able to report the recent creation of a file allowing it
to be opened rather than the open failing with ENOENT.
This is intended to be functionally equivalent to behavior in the NFS
client which re-validates a cached dentry on file open.
Signed-off-by: Kenneth C Schalk <kschalk@xxxxxxxxxx>
---
fs/fuse/dir.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index c4a0129..b46f5e8 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -204,7 +204,8 @@ static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
if (inode && is_bad_inode(inode))
goto invalid;
else if (time_before64(fuse_dentry_time(entry), get_jiffies_64()) ||
- (flags & LOOKUP_REVAL)) {
+ (!inode && (flags & LOOKUP_OPEN)) ||
+ (flags & (LOOKUP_CREATE | LOOKUP_REVAL))) {
struct fuse_entry_out outarg;
FUSE_ARGS(args);
struct fuse_forget_link *forget;
--
2.9.2
