On Mon, Jun 21, 2021 at 03:59:58PM +0200, Christoph Hellwig wrote: > What about just setting O_DSYNC in f_flags at open time if a superblock > has SB_SYNCHRONOUS set? That means the flag won't be picked up for open > files when SB_SYNCHRONOUS is set at remount time, but I'm not even sure > we need to support it for remount. i suggested that to viro last night, and he pointed out that ioctl(S_SYNC) would also not be reflected in open files. that's more of a concern, i think. if only we had the revoke() infrastructure (ie could walk currently-open files on an inode).
