---- 在 星期四, 2021-04-01 19:31:12 Miklos Szeredi <miklos@xxxxxxxxxx> 撰写 ---- > On Thu, Apr 1, 2021 at 1:15 PM Chengguang Xu <cgxu519@xxxxxxxxxxxx> wrote: > > > > ---- 在 星期一, 2021-03-29 23:13:52 Miklos Szeredi <miklos@xxxxxxxxxx> 撰写 ---- > > > On Mon, Mar 8, 2021 at 12:17 PM Chengguang Xu <cgxu519@xxxxxxxxxxxx> wrote: > > > > > > > > Currently copy-up will copy whole lower file to upper > > > > regardless of the data range which is needed for further > > > > operation. This patch avoids unnecessary copy when truncate > > > > size is smaller than the file size. > > > > > > This doesn't look right. If copy up succeeds, resulting in a > > > truncated file, then we should return success there and then. Doing > > > the truncate again and failing (unlikely, but I wouldn't think it > > > impossible) wouldn't be nice. > > > > Hi Miklos > > > > I noticed a problem here, if we just return success after copy-up then mtime > > keeps the same as lower file. I think doing the truncate again would be better > > than manually updating the upper file's mtime. What do you think for this? > > Let's simplify instead: skip the mtime restore on copy-up. Not sure > how that's handled on O_TRUNC opens, maybe it's relevant to that case > too. > Currently on O_TRUNC open, copy-up(zero size) triggered by vfs_open() then do the truncate(zero size) operation by handle_truncate() afterwards. mtime will be updated after calling handle_truncate(). Thanks, Chengguang
