Re: [pseudo] [PATCH v3 1/6] ports/linux/guts: Add open_tree wrapper

[Paul Barker <paul@pbarker.dev>]

Hi Dmitry,

Apologies for the slow replies here, we're all busy and pseudo is
under-maintained.

On Mon, 2026-09-21 at 14:12 +0300, Dmitry Sakhonchik wrote:
> пн, 21 сент. 2026 г. в 12:28, Paul Barker <paul@pbarker.dev>:
> 
> > On Mon, 2026-09-07 at 14:18 +0300, frezidok1@gmail.com wrote:
> > > From: Dmitry Sakhonchik <frezidok1@gmail.com>
> > > 
> > > [YOCTO #16379]
> > > 
> > > OPEN_TREE_CLONE is not handled since it requires CAP_SYS_ADMIN and is
> > outside the common pseudo use case.
> > 
> > We should at least log an error if OPEN_TREE_CLONE is seen instead of
> > silently ignoring it.
> > 
> > > 
> > > Signed-off-by: Dmitry Sakhonchik <frezidok1@gmail.com>
> > > ---
> > >  ports/linux/guts/open_tree.c | 46 ++++++++++++++++++++++++++++++++++++
> > >  ports/linux/portdefs.h       |  9 +++++++
> > >  ports/linux/wrapfuncs.in     |  1 +
> > >  3 files changed, 56 insertions(+)
> > >  create mode 100644 ports/linux/guts/open_tree.c
> > > 
> > > diff --git a/ports/linux/guts/open_tree.c b/ports/linux/guts/open_tree.c
> > > new file mode 100644
> > > index 0000000..6c2aad9
> > > --- /dev/null
> > > +++ b/ports/linux/guts/open_tree.c
> > > @@ -0,0 +1,46 @@
> > > +/*
> > > + * Copyright (c) 2026 Yocto Project
> > > + * guts/COPYRIGHT for information.
> > > + *
> > > + * SPDX-License-Identifier: LGPL-2.1-only
> > > + *
> > > + * int open_tree(int dirfd, const char *path, unsigned int flags)
> > > + *   int rc = -1;
> > > + */
> > > +
> > > +     struct stat64 buf;
> > > +     int save_errno;
> > > +     char *pseudo_path;
> > > +
> > > +     if (real_open_tree) {
> > > +             pseudo_debug(PDBGF_SYSCALL, "open_tree, calling
> > open_tree.\n");
> > > +             rc = real_open_tree(dirfd, path, flags);
> > > +     } else {
> > > +             pseudo_debug(PDBGF_SYSCALL, "open_tree, calling
> > syscall.\n");
> > > +             rc = real_syscall(SYS_open_tree, dirfd, path, flags);
> > > +     }
> > 
> > The commit message or a comment should briefly explain why we need to
> > handle both ways of invoking the real open_tree syscall.
> > 
> > > +
> > > +     pseudo_path = pseudo_root_path(__func__, __LINE__, dirfd, path, 0);
> > 
> > This should be after the (rc != -1) conditional, as this call could
> > clobber errno.
> > 
> > If we use preserve_path=1, then the absolute path must go through
> pseudo_root_path before the real open_tree call so that a possible chroot
> is handled correctly.
> So I think these two cases need to be somehow handled separately rather
> than moving pseudo_root_path() entirely below the rc != -1 check.
> By the way, for some reason the openat2 wrapper is ok with the
> pseudo_root_path outside of conditional.

The problem is that the real open_tree() call could set errno if it
fails, and calling pseudo_root_path() after this before returning the
error could overwrite errno. It is safe to call pseudo_root_path()
before calling the real open_tree(), or to only call it if the real
open_tree() call succeeds.

Best regards,

-- 
Paul Barker