On Sun, Jun 11, 2023 at 7:55 PM Amir Goldstein <amir73il@xxxxxxxxx> wrote: > > On Sun, Jun 11, 2023 at 5:23 PM Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > > > > On Sun, 11 Jun 2023 at 15:27, Amir Goldstein <amir73il@xxxxxxxxx> wrote: > > > > > > Miklos, > > > > > > The first solution that we discussed for removing FMODE_NONOTIFY > > > from overlayfs real files using file_fake container got complicated. > > > > > > This alternative solution is less intrusive to vfs and all the vfs > > > code should remian unaffected expect for the special fsnotify case > > > that we want to fix. > > > > > > Thanks, > > > Amir. > > > > > > Changes since v1: > > > - Drop the file_fake container > > > > Why? > > See my question on v1. > The fake file objects are used both as vm_file and in read/write iter > how do we know which path to use in low level functions? > > If we allocate file_fake container and still store the fake path > in f_path, then we have no need to store also the real path > because ovl knows how to get from fake path to real path. > > This is what v2 does. > > What am I missing? > Is it because getting f_real_path() and file_dentry() via d_real() is more expensive? and caching this information in file_fake container would be more efficient? I will restore the file_fake container and post v3. Thanks, Amir.
