Hi, There is a long-standing demand for synchronous behaviour of fuse_release:
http://sourceforge.net/mailarchive/message.php?msg_id=19343889 http://sourceforge.net/mailarchive/message.php?msg_id=29814693 A year ago Avati and me explained why such a feature would be useful: http://sourceforge.net/mailarchive/message.php?msg_id=29889055 http://sourceforge.net/mailarchive/message.php?msg_id=29867423 In short, the problem is that fuse_release (that's usually called on last user close(2)) sends FUSE_RELEASE to userspace and returns without waiting for ACK from userspace. Consequently, there is a gap when user regards the file released while userspace fuse is still working on it. An attempt to access the file from another node leads to complicated synchronization problems because the first node still "holds" the file. The patch-set resolves the problem by making fuse_release synchronous: wait for ACK from userspace for FUSE_RELEASE if the feature is ON. It must be emphasized that even if the feature is enabled (i.e. fuse_release is synchronous), nothing guarantees that fuse_release() will be called in the context of close(2). In fact, it may happen later, on last fput(). However, there are still a lot of use cases when close(2) is synchronous, so the feature must be regarded as an optimization maximizing chances of synchronous behaviour of close(2). To keep single-threaded userspace implementations happy the patch-set ensures that by the time fuse_release_common calls fuse_file_put, no more in-flight I/O exists. Asynchronous fuse callbacks (like fuse_readpages_end) cannot trigger FUSE_RELEASE anymore. Hence, we'll never block in contexts other than close(). Changed in v2: - improved comments, commented spin_unlock_wait out according to Brian' suggestions. - rebased on v3.15-rc8 tag of Linus' tree. Changed in v3: - changed patch-set title from "close file synchronously" to "handle release synchronously" - renamed CLOSE_WAIT to SYNC_RELEASE to convey the essence of the feature ("synchronous release" instead of "wait on close") - enabled feature on per file basis (instead of global fuse_conn parameter) - added "disable_sync_release" mount option - rebased on v3.17-rc1 tag of Linus' tree. Changed in v4: - removed redundant locking around __fuse_file_put() - removed a patch making FUSE_RELEASE request uninterruptible. As Miklos pointed out, if the request is interrupted, then we should possibly allow that, effectively backgrounding the request. However, such a backgrounding is not easy to implement without breaking locking expectations of the userspace filesystem. Given the problem has existed for fuseblk mount for long time and there is no reasonable solution now, I put it aside for the future. Thanks, Maxim --- Maxim Patlasov (5): fuse: add FOPEN_SYNC_RELEASE flag to ff->open_flags fuse: cosmetic rework of fuse_send_readpages fuse: wait for end of IO on release fuse: add mount option to disable synchronous release fuse: enable close_wait synchronous release fs/fuse/file.c | 97 ++++++++++++++++++++++++++++++++++++++------- fs/fuse/fuse_i.h | 3 + fs/fuse/inode.c | 8 ++++ include/uapi/linux/fuse.h | 3 + 4 files changed, 95 insertions(+), 16 deletions(-) -- Signature -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
