| Message ID | 20180816073407.12567-1-avagin@openvz.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [dhowells/mount-api] proc: set a proper user namespace for fs_context | expand |
Hi David,
I reported this problem about a month ago when patches were not in the
linux-next. Now they are there and we can't run CRIU tests, because it
is impossible to mount /proc in a container.
[root@fc24 ~]# strace unshare -Urnm --mount-proc true
unshare(CLONE_NEWNS|CLONE_NEWNET|CLONE_NEWUSER) = 0
openat(AT_FDCWD, "/proc/self/setgroups", O_WRONLY) = 3
write(3, "deny", 4) = 4
close(3) = 0
openat(AT_FDCWD, "/proc/self/uid_map", O_WRONLY) = 3
write(3, "0 0 1", 5) = 5
close(3) = 0
openat(AT_FDCWD, "/proc/self/gid_map", O_WRONLY) = 3
write(3, "0 0 1", 5) = 5
close(3) = 0
mount("none", "/", NULL, MS_REC|MS_PRIVATE, NULL) = 0
mount("none", "/proc", NULL, MS_REC|MS_PRIVATE, NULL) = 0
mount("proc", "/proc", "proc", MS_NOSUID|MS_NODEV|MS_NOEXEC, NULL) =
-1 EBUSY (Device or resource busy)
https://travis-ci.org/avagin/linux/jobs/416641093
Thanks,
Andrei
On Thu, Aug 16, 2018 at 12:34 AM Andrei Vagin <avagin@openvz.org> wrote:
>
> A user namespace should be taken from a pidns for which a procfs is created.
>
> Signed-off-by: Andrei Vagin <avagin@gmail.com>
> ---
> fs/proc/root.c | 5 +++++
> 1 file changed, 5 insertions(+)
>
> diff --git a/fs/proc/root.c b/fs/proc/root.c
> index 1d6e5bfa30cc..1419b48a89ab 100644
> --- a/fs/proc/root.c
> +++ b/fs/proc/root.c
> @@ -315,6 +315,11 @@ int pid_ns_prepare_proc(struct pid_namespace *ns)
> if (IS_ERR(fc))
> return PTR_ERR(fc);
>
> + if (fc->user_ns != ns->user_ns) {
> + put_user_ns(fc->user_ns);
> + fc->user_ns = get_user_ns(ns->user_ns);
> + }
> +
> ctx = fc->fs_private;
> if (ctx->pid_ns != ns) {
> put_pid_ns(ctx->pid_ns);
> --
> 2.17.1
>
Andrei Vagin <avagin@openvz.org> wrote: > A user namespace should be taken from a pidns for which a procfs is created. That would seem wrong. Shouldn't the superblock user_ns be from the mounter? Adding Al and Eric to the list to get their opinion. David > Signed-off-by: Andrei Vagin <avagin@gmail.com> > --- > fs/proc/root.c | 5 +++++ > 1 file changed, 5 insertions(+) > > diff --git a/fs/proc/root.c b/fs/proc/root.c > index 1d6e5bfa30cc..1419b48a89ab 100644 > --- a/fs/proc/root.c > +++ b/fs/proc/root.c > @@ -315,6 +315,11 @@ int pid_ns_prepare_proc(struct pid_namespace *ns) > if (IS_ERR(fc)) > return PTR_ERR(fc); > > + if (fc->user_ns != ns->user_ns) { > + put_user_ns(fc->user_ns); > + fc->user_ns = get_user_ns(ns->user_ns); > + } > + > ctx = fc->fs_private; > if (ctx->pid_ns != ns) { > put_pid_ns(ctx->pid_ns);
David Howells <dhowells@redhat.com> wrote: > > A user namespace should be taken from a pidns for which a procfs is created. > > That would seem wrong. Shouldn't the superblock user_ns be from the mounter? Ah, no. The change is correct. What the patch description doesn't mention is that this is when a new pid namespace is setting up its own proc filesystem. I'll change the subject and patch body to: proc: Set correct userns for new proc super created by a new pid_namespace Fix the setting up a new proc superblock for a new pid_namespace such that the user_ns for that proc superblock needs to be taken from the new pid_namespace and not the active process. which I think describes it better. David
diff --git a/fs/proc/root.c b/fs/proc/root.c index 1d6e5bfa30cc..1419b48a89ab 100644 --- a/fs/proc/root.c +++ b/fs/proc/root.c @@ -315,6 +315,11 @@ int pid_ns_prepare_proc(struct pid_namespace *ns) if (IS_ERR(fc)) return PTR_ERR(fc); + if (fc->user_ns != ns->user_ns) { + put_user_ns(fc->user_ns); + fc->user_ns = get_user_ns(ns->user_ns); + } + ctx = fc->fs_private; if (ctx->pid_ns != ns) { put_pid_ns(ctx->pid_ns);
A user namespace should be taken from a pidns for which a procfs is created. Signed-off-by: Andrei Vagin <avagin@gmail.com> --- fs/proc/root.c | 5 +++++ 1 file changed, 5 insertions(+)