On Mon, 12 Jul 2021 at 17:52, Amir Goldstein <amir73il@xxxxxxxxx> wrote:
>
> On Mon, Jul 12, 2021 at 6:51 PM Amir Goldstein <amir73il@xxxxxxxxx> wrote:
> >
> > On Mon, Jul 12, 2021 at 5:20 PM Miklos Szeredi <miklos@xxxxxxxxxx> wrote:
> > >
> > > On Sat, 19 Jun 2021 at 11:26, Amir Goldstein <amir73il@xxxxxxxxx> wrote:
> > > >
> > > > When a lower file has sync/noatime fileattr flags, the behavior of
> > > > overlayfs post copy up is inconsistent.
> > > >
> > > > Immediattely after copy up, ovl inode still has the S_SYNC/S_NOATIME
> > > > inode flags copied from lower inode, so vfs code still treats the ovl
> > > > inode as sync/noatime. After ovl inode evict or mount cycle,
> > > > the ovl inode does not have these inode flags anymore.
> > > >
> > > > To fix this inconsitency, try to copy the fileattr flags on copy up
> > > > if the upper fs supports the fileattr_set() method.
> > > >
> > > > This gives consistent behavior post copy up regardless of inode eviction
> > > > from cache.
> > > >
> > > > We cannot copy up the immutable/append-only inode flags in a similar
> > > > manner, because immutable/append-only inodes cannot be linked and because
> > > > overlayfs will not be able to set overlay.* xattr on the upper inodes.
> > > >
> > > > Those flags will be addressed by a followup patch.
> > > >
> > > > Signed-off-by: Amir Goldstein <amir73il@xxxxxxxxx>
> > > > ---
> > > > fs/overlayfs/copy_up.c | 49 ++++++++++++++++++++++++++++++++++------
> > > > fs/overlayfs/inode.c | 36 ++++++++++++++++++-----------
> > > > fs/overlayfs/overlayfs.h | 14 +++++++++++-
> > > > 3 files changed, 78 insertions(+), 21 deletions(-)
> > > >
> > > > diff --git a/fs/overlayfs/copy_up.c b/fs/overlayfs/copy_up.c
> > > > index 3fa68a5cc16e..a06b423ca5d1 100644
> > > > --- a/fs/overlayfs/copy_up.c
> > > > +++ b/fs/overlayfs/copy_up.c
> > > > @@ -8,6 +8,7 @@
> > > > #include <linux/fs.h>
> > > > #include <linux/slab.h>
> > > > #include <linux/file.h>
> > > > +#include <linux/fileattr.h>
> > > > #include <linux/splice.h>
> > > > #include <linux/xattr.h>
> > > > #include <linux/security.h>
> > > > @@ -130,6 +131,31 @@ int ovl_copy_xattr(struct super_block *sb, struct dentry *old,
> > > > return error;
> > > > }
> > > >
> > > > +static int ovl_copy_fileattr(struct path *old, struct path *new)
> > > > +{
> > > > + struct fileattr oldfa = { .flags_valid = true };
> > > > + struct fileattr newfa = { .flags_valid = true };
> > > > + int err;
> > > > +
> > > > + err = ovl_real_fileattr(old, &oldfa, false);
> > > > + if (err)
> > > > + return err;
> > > > +
> > > > + err = ovl_real_fileattr(new, &newfa, false);
> > > > + if (err)
> > > > + return err;
> > > > +
> > > > + BUILD_BUG_ON(OVL_COPY_FS_FLAGS_MASK & ~FS_COMMON_FL);
> > > > + newfa.flags &= ~OVL_COPY_FS_FLAGS_MASK;
> > > > + newfa.flags |= (oldfa.flags & OVL_COPY_FS_FLAGS_MASK);
> > > > +
> > > > + BUILD_BUG_ON(OVL_COPY_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON);
> > > > + newfa.fsx_xflags &= ~OVL_COPY_FSX_FLAGS_MASK;
> > > > + newfa.fsx_xflags |= (oldfa.fsx_xflags & OVL_COPY_FSX_FLAGS_MASK);
> > > > +
> > > > + return ovl_real_fileattr(new, &newfa, true);
> > > > +}
> > > > +
> > > > static int ovl_copy_up_data(struct ovl_fs *ofs, struct path *old,
> > > > struct path *new, loff_t len)
> > > > {
> > > > @@ -493,20 +519,21 @@ static int ovl_link_up(struct ovl_copy_up_ctx *c)
> > > > static int ovl_copy_up_inode(struct ovl_copy_up_ctx *c, struct dentry *temp)
> > > > {
> > > > struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
> > > > + struct inode *inode = d_inode(c->dentry);
> > > > + struct path upperpath, datapath;
> > > > int err;
> > > >
> > > > + ovl_path_upper(c->dentry, &upperpath);
> > > > + if (WARN_ON(upperpath.dentry != NULL))
> > > > + return -EIO;
> > > > +
> > > > + upperpath.dentry = temp;
> > > > +
> > > > /*
> > > > * Copy up data first and then xattrs. Writing data after
> > > > * xattrs will remove security.capability xattr automatically.
> > > > */
> > > > if (S_ISREG(c->stat.mode) && !c->metacopy) {
> > > > - struct path upperpath, datapath;
> > > > -
> > > > - ovl_path_upper(c->dentry, &upperpath);
> > > > - if (WARN_ON(upperpath.dentry != NULL))
> > > > - return -EIO;
> > > > - upperpath.dentry = temp;
> > > > -
> > > > ovl_path_lowerdata(c->dentry, &datapath);
> > > > err = ovl_copy_up_data(ofs, &datapath, &upperpath,
> > > > c->stat.size);
> > > > @@ -518,6 +545,14 @@ static int ovl_copy_up_inode(struct ovl_copy_up_ctx *c, struct dentry *temp)
> > > > if (err)
> > > > return err;
> > > >
> > > > + if (inode->i_flags & OVL_COPY_I_FLAGS_MASK) {
> > > > + /*
> > > > + * Copy the fileattr inode flags that are the source of already
> > > > + * copied i_flags (best effort).
> > > > + */
> > > > + ovl_copy_fileattr(&c->lowerpath, &upperpath);
> > >
> > > I'm not sure this should be ignoring errors. Was this done to prevent
> > > regressing cases where the upper fs cannot store the flags?
> >
> > Yes.
> >
> > > Do you have a concrete example?
> >
> > Unpriv userns mount??
> >
>
> Upperfs that does not support fileattr (FUSE?)
Okay, so two subcases:
- no xattr on upper
- no fileattr on upper
FUSE is considered "remote" and overlayfs enforces xattr support, but
not fileattr support. So yeah, it seems theoretically these are
possible.
But I think it might be the best to risk it and return the error,
hoping that this is such a rare corner case that there's no existing
use. If a regression is reported, than we need to go for a more
complex solution where this case is detected on startup and handled
accordingly.
Thanks,
Miklos
