Hi guys, Since 2.6.37 (commit 72f98e72), f_op->setlease() is called with a spinlock held (file_lock_lock). This removed BKL but also made distributed filesystems very difficult to implement leasing, which typically needs to sleep for network communications (considering the cifs case). Commit 1c8c601a changes this into inode->i_lock but it's still a spinlock. And since i_lock is widely used. The f_op->setlease() implementer needs to be even more careful to avoid a recursive spinlock in calling other VFS functions. So would you consider to remove the spinlock around f_op->setlease()? I believe here the spinlock is used to protect the returned "struct file_lock **flp". I think it could be acquired in the enter of generic_setlease() but not be released if (and only if) the generic_setlease() returned successfully. This can make the filesystems do anything they want before calling generic_setlease(). Of course the code after generic_setlease() still need to be careful, but it's much easier. Especially, if there are any errors happened, the filesystem need to release the lock explicity and then call __break_lease(). Or we can simply change spinlock into mutex, or any mechanism which is safe for sleeping? Copy to Jeff and Arnd who were working on that lock. Also Miklos, yes, it's still me, the one was worked on fuse adaptive readdir_plus....now I'm working for GPFS.... -- Feng Shuo -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
