From: sashiko-bot@kernel.org
To: "Jarkko Sakkinen" <jarkko@kernel.org>
Cc: linux-mm@kvack.org, linux-security-module@vger.kernel.org,
"Alexander Mikhalitsyn" <alex@amutable.com>,
"Alexander Viro" <viro@zeniv.linux.org.uk>,
"Andy Lutomirski" <luto@amacapital.net>,
"Christian Brauner" <brauner@kernel.org>,
"David Howells" <dhowells@redhat.com>, "Jan Kara" <jack@suse.cz>,
"Jonathan Corbet" <corbet@lwn.net>,
"Josef Bacik" <josef@toxicpanda.com>,
"Kees Cook" <kees@kernel.org>, "Neal Gompa" <neal@gompa.dev>,
"Paul Moore" <paul@paul-moore.com>,
"Serge Hallyn" <sergeh@kernel.org>,
"Zygmunt Krynicki" <zygmunt.krynicki@canonical.com>
Subject: Re: [RFC PATCH v5 02/25] userns: Add userns_set_idmap() system call
Date: Sun, 22 Nov 2026 00:10:52 +0000 [thread overview]
Message-ID: <sashiko-outbox-227537@kernel.org> (raw)
In-Reply-To: <20261121000844.1863324-3-jarkko@kernel.org>
Thank you for your contribution! Sashiko AI review found 3 potential
issues to consider:
- [High] userns: missing target CAP_SYS_ADMIN and caller checks in
sys_userns_set_idmap()
- [Medium] userns: unbounded size passed to copy_struct_from_user() in
sys_userns_set_idmap()
- [Medium] userns: transient ns->flags mutation in
sys_userns_set_idmap() before validation
--
commit cd58304550496b7580b53ae3de9d67b848eccae2
Author: Jarkko Sakkinen <jarkko@kernel.org>
userns: Add userns_set_idmap() system call
This commit introduces the userns_set_idmap() system call to atomically
configure setgroups policy and UID, GID, and project ID mappings directly
on a user namespace file descriptor.
> diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
> index 6b9673dcdc670..d7f2cad388c8c 100644
> --- a/kernel/user_namespace.c
> +++ b/kernel/user_namespace.c
[ ... ]
> +static bool caller_idmap_permitted(struct user_namespace *ns, int cap_setid,
> + const struct uid_gid_map *new_map)
> +{
> + const struct cred *cred = current_cred();
> +
> + if (cap_setid == CAP_SETUID && idmap_maps_root(new_map)) {
> + if (cred->user_ns == ns) {
> + if (!ns->parent_could_setfcap)
> + return false;
> + } else {
> + if (!ns_capable(ns->parent, CAP_SETFCAP))
> + return false;
> + }
> + }
> +
> + if (new_map->nr_extents == 1 && new_map->extent[0].count == 1 &&
> + uid_eq(ns->owner, cred->euid)) {
> + u32 id = new_map->extent[0].lower_first;
> +
> + if (cap_setid == CAP_SETUID) {
> + kuid_t uid = make_kuid(ns->parent, id);
> +
> + if (uid_eq(uid, cred->euid))
> + return true;
> + } else if (cap_setid == CAP_SETGID) {
> + kgid_t gid = make_kgid(ns->parent, id);
> +
> + if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
> + gid_eq(gid, cred->egid))
> + return true;
> + }
> + }
> +
> + if (!cap_valid(cap_setid))
> + return true;
> +
> + return ns_capable(ns->parent, cap_setid);
> +}
[ ... ]
> +SYSCALL_DEFINE4(userns_set_idmap, int, fd,
> + const struct userns_idmap __user *, map,
> + size_t, size, unsigned int, flags)
> +{
[ ... ]
> + if (!map || size < USERNS_IDMAP_SIZE_VER0)
> + return -EINVAL;
> +
> + ret = copy_struct_from_user(&req, sizeof(req), map, size);
> + if (ret)
> + return ret;
[Severity: Medium]
Should sys_userns_set_idmap() reject size > PAGE_SIZE with -E2BIG before
calling copy_struct_from_user()? When size exceeds sizeof(req),
copy_struct_from_user() calls check_zeroed_user() on the trailing bytes,
which scans the user range word-by-word inside user_read_access_begin()
without rescheduling or checking for pending signals. An unprivileged
caller can pass a multi-gigabyte zero-filled mapping and stall a CPU in
the kernel.
> +
> + if (!req.flags || (req.flags & ~USERNS_IDMAP_ALL_FLAGS))
> + return -EINVAL;
> + if (!(req.flags & USERNS_IDMAP_UID) &&
> + (req.nr_uid_extents || req.uid_extents))
> + return -EINVAL;
> + if (!(req.flags & USERNS_IDMAP_GID) &&
> + (req.nr_gid_extents || req.gid_extents))
> + return -EINVAL;
> + if (!(req.flags & USERNS_IDMAP_PROJID) &&
> + (req.nr_projid_extents || req.projid_extents))
> + return -EINVAL;
> +
> + if (!ns->parent)
> + return -EPERM;
> + if (current_user_ns() != ns && current_user_ns() != ns->parent)
> + return -EPERM;
> +
> + if (req.flags & (USERNS_IDMAP_UID | USERNS_IDMAP_GID |
> + USERNS_SETGROUPS_DENY)) {
> + bool file_ok = (filp->f_cred->user_ns == ns ||
> + filp->f_cred->user_ns == ns->parent) &&
> + file_ns_capable(filp, ns, CAP_SYS_ADMIN);
> +
> + if (!ns_capable(ns, CAP_SYS_ADMIN) && !file_ok)
> + return -EPERM;
> + }
[Severity: High]
Does omitting USERNS_IDMAP_PROJID from this CAP_SYS_ADMIN check allow any
unprivileged process holding a read-only userns fd to write ns->projid_map?
For /proc/<pid>/projid_map, writing requires opening the procfs file
(mode 0644, owned by the target task) for write, so VFS permission checks
block unrelated users in ns->parent and unprivileged non-owner tasks inside
ns. By contrast, nsfs files have mode 0444 and are always opened O_RDONLY,
and caller_idmap_permitted() returns true unconditionally when cap_setid is
-1. As a result, an unprivileged task inside ns that dropped capabilities,
or an unrelated user in ns->parent that obtained an nsfs fd (for example via
NS_GET_USERNS), can call userns_set_idmap() with USERNS_IDMAP_PROJID and
lock in the write-once projid_map.
In addition, when file_ok is true on a delegated nsfs fd and the caller
lacks ns_capable(ns, CAP_SYS_ADMIN), neither USERNS_SETGROUPS_DENY nor
USERNS_IDMAP_PROJID verifies any privilege on current_cred() in ns->parent,
allowing an unprivileged recipient of a read-only userns fd to disable
setgroups(2) or configure projid_map on ns.
[ ... ]
> + if (req.flags & USERNS_SETGROUPS_DENY)
> + ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
> +
> + if (req.flags & USERNS_IDMAP_UID) {
> + ret = prepare_caller_idmap(ns, CAP_SETUID, &new_uid,
> + &ns->parent->uid_map);
> + if (ret < 0)
> + goto out_restore;
> + }
> + if (req.flags & USERNS_IDMAP_GID) {
> + ret = prepare_caller_idmap(ns, CAP_SETGID, &new_gid,
> + &ns->parent->gid_map);
> + if (ret < 0)
> + goto out_restore;
> + }
> + if (req.flags & USERNS_IDMAP_PROJID) {
> + ret = prepare_caller_idmap(ns, -1, &new_projid,
> + &ns->parent->projid_map);
> + if (ret < 0)
> + goto out_restore;
> + }
> +
> + if (req.flags & USERNS_IDMAP_UID)
> + commit_new_idmap(&ns->uid_map, &new_uid);
> + if (req.flags & USERNS_IDMAP_GID)
> + commit_new_idmap(&ns->gid_map, &new_gid);
> + if (req.flags & USERNS_IDMAP_PROJID)
> + commit_new_idmap(&ns->projid_map, &new_projid);
> +
> + mutex_unlock(&userns_state_mutex);
> + return 0;
> +
> +out_restore:
> + ns->flags = old_flags;
[Severity: Medium]
Can clearing USERNS_SETGROUPS_ALLOWED in ns->flags before
prepare_caller_idmap() expose a transient "deny" state to concurrent
readers if preparation fails?
proc_setgroups_show() reads READ_ONCE(ns->flags) without holding
userns_state_mutex:
kernel/user_namespace.c:proc_setgroups_show() {
struct user_namespace *ns = seq->private;
unsigned long userns_flags = READ_ONCE(ns->flags);
seq_printf(seq, "%s\n",
(userns_flags & USERNS_SETGROUPS_ALLOWED) ?
"allow" : "deny");
return 0;
}
While prepare_caller_idmap() needs USERNS_SETGROUPS_ALLOWED cleared when
checking an unprivileged caller's GID mapping, mutating ns->flags in place
and restoring old_flags on failure lets a concurrent reader of
/proc/<pid>/setgroups observe "deny" before it reverts to "allow". Could
caller_idmap_permitted() check the effective flags value and leave the
update to ns->flags until the commit phase after all prepare_caller_idmap()
calls succeed?
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261121000844.1863324-1-jarkko@kernel.org?part=2
next prev parent reply other threads:[~2026-11-22 0:10 UTC|newest]
Thread overview: 52+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-21 0:08 [RFC PATCH v5 00/25] rootns: fd-bound process closures Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 01/25] userns: Factor out idmap validation and commit helpers Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 02/25] userns: Add userns_set_idmap() system call Jarkko Sakkinen
2026-11-21 16:09 ` Jarkko Sakkinen
2026-11-21 16:49 ` Jarkko Sakkinen
2026-11-21 16:53 ` Jarkko Sakkinen
2026-11-22 0:10 ` sashiko-bot [this message]
2026-11-21 0:08 ` [RFC PATCH v5 03/25] arch: Wire up " Jarkko Sakkinen
2026-11-21 15:58 ` Jarkko Sakkinen
2026-11-21 18:12 ` Jarkko Sakkinen
2026-11-22 0:10 ` sashiko-bot
2026-11-21 0:08 ` [RFC PATCH v5 04/25] mount: Add mnt_ns_get_root() Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 05/25] mount: Enforce mount_too_revealing() for foreign mntns Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 06/25] mount: Support MOVE_MOUNT_T_MNTNS in move_mount() Jarkko Sakkinen
2026-11-22 0:10 ` sashiko-bot
2026-11-21 0:08 ` [RFC PATCH v5 07/25] mount: Support MOVE_MOUNT_CLONE with MOVE_MOUNT_T_MNTNS Jarkko Sakkinen
2026-11-22 0:10 ` sashiko-bot
2026-11-21 0:08 ` [RFC PATCH v5 08/25] userns: Factor out install_user_ns_cred() Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 09/25] nsfs: Add nsproxy_open_namespace() Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 10/25] nsproxy: Pass nsproxy to create_new_namespaces() Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 11/25] rootns: Add core infrastructure for root namespaces Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 12/25] rootns: Track init lifecycle via copy_process() and pid_namespace Jarkko Sakkinen
2026-11-21 1:32 ` Jarkko Sakkinen
2026-11-21 18:08 ` Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 13/25] rootns: Add rootns() system call Jarkko Sakkinen
2026-11-21 19:10 ` Jarkko Sakkinen
2026-11-21 19:12 ` Jarkko Sakkinen
2026-11-21 19:54 ` Jarkko Sakkinen
2026-11-21 20:16 ` Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 14/25] rootns: Add ROOTNS_WAIT ioctl Jarkko Sakkinen
2026-11-21 0:17 ` Jarkko Sakkinen
2026-11-21 0:18 ` Jarkko Sakkinen
2026-11-21 1:09 ` Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 15/25] rootns: Add ROOTNS_KILL ioctl Jarkko Sakkinen
2026-11-21 1:19 ` Jarkko Sakkinen
2026-11-21 11:24 ` Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 16/25] fork: Support CLONE_INTO_ROOTNS in clone3() Jarkko Sakkinen
2026-11-22 0:10 ` sashiko-bot
2026-11-21 0:08 ` [RFC PATCH v5 17/25] fs: Support FSCONFIG_SET_ROOTNS in fsconfig() Jarkko Sakkinen
2026-11-22 0:10 ` sashiko-bot
2026-11-21 0:08 ` [RFC PATCH v5 18/25] proc: Implement ->set_rootns() fs_context operation Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 19/25] ipc/mqueue: " Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 20/25] security: Add rootns LSM hooks Jarkko Sakkinen
2026-11-21 11:47 ` Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 21/25] bpf: Register " Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 22/25] arch: Wire up rootns() system call Jarkko Sakkinen
2026-11-22 0:11 ` sashiko-bot
2026-11-21 0:08 ` [RFC PATCH v5 23/25] docs: Document root namespace uAPI Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 24/25] selftests/rootns: Add root namespace test suite Jarkko Sakkinen
2026-11-21 0:08 ` [RFC PATCH v5 25/25] MAINTAINERS: Add entry for ROOT NAMESPACE Jarkko Sakkinen
2026-11-21 17:21 ` Jarkko Sakkinen
2026-11-22 0:11 ` [RFC PATCH v5 00/25] rootns: fd-bound process closures sashiko-bot
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=sashiko-outbox-227537@kernel.org \
--to=sashiko-bot@kernel.org \
--cc=alex@amutable.com \
--cc=brauner@kernel.org \
--cc=corbet@lwn.net \
--cc=dhowells@redhat.com \
--cc=jack@suse.cz \
--cc=jarkko@kernel.org \
--cc=josef@toxicpanda.com \
--cc=kees@kernel.org \
--cc=linux-mm@kvack.org \
--cc=linux-security-module@vger.kernel.org \
--cc=luto@amacapital.net \
--cc=neal@gompa.dev \
--cc=paul@paul-moore.com \
--cc=sashiko-reviews@lists.linux.dev \
--cc=sergeh@kernel.org \
--cc=viro@zeniv.linux.org.uk \
--cc=zygmunt.krynicki@canonical.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox