On Wed, Sep 20, 2017 at 08:22:54AM +0800, Qu Wenruo wrote: > The costly part will be tracking the filesystems of subvolumes. > We must do it for each subvolume and batch them to address the final > transaction commit for each fs. > > I didn't see any easy ioctl to get the UUID from fd, meaning (if didn't > miss anything) we need to iterate the path until reaching the mount > boundary and then refer to mountinfo to find the fs.
The FS_INFO ioctl will tell you the fsid from a fd. Iterating the paths will not work reliably, due to some potentially very creative use of mounts that could build up the path components. > Not to mention that the possible softlink in the path may make things > more complex. > > Yes, this may be fixed with tons of code, but I don't think the > complexity worthy for a minor feature. So what if we fix it like that: - iterate over subvolumes - check if the fsid is same as for the previous subvolume - if it is, continue - if not, do the sync IOW, sync when we leave subvolumes of one filesystem. In the degenerate case, we can have subvolumes interleaved that we'd sync after each, effectively --commit-each. The simple tracking should avoid keeping all the filedescriptors open and I think this won't need tons of code to implement. The typical could be 'btrfs subvol delete -c path/*', ie. deleting from just one filesystem. If we happen to cross more filesystems, each of them will get the sync, but possibly more than one. I don't think this is too serious. -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
