* [PATCH] support_become_root: Enable file creation in user namespaces
@ 2017-11-17 14:55 Florian Weimer
2017-11-17 17:20 ` Jonathan Nieder
0 siblings, 1 reply; 2+ messages in thread
From: Florian Weimer @ 2017-11-17 14:55 UTC (permalink / raw)
To: libc-alpha
Without UID/GID maps, file creation will file with EOVERFLOW.
This patch is based on DJ Delorie's work on container testing.
2017-11-17 Florian Weimer <fweimer@redhat.com>
support_become_root: Enable file creation in namespaces.
* support/support_become_root.c (setup_mapping): New function.
(support_become_root): Call it.
diff --git a/support/support_become_root.c b/support/support_become_root.c
index 3fa0bd4ac0..5086570251 100644
--- a/support/support_become_root.c
+++ b/support/support_become_root.c
@@ -18,18 +18,69 @@
#include <support/namespace.h>
+#include <fcntl.h>
#include <sched.h>
#include <stdio.h>
+#include <string.h>
+#include <support/check.h>
+#include <support/xunistd.h>
#include <unistd.h>
+#ifdef CLONE_NEWUSER
+/* The necessary steps to allow file creation in user namespaces. */
+static void
+setup_uid_gid_mapping (uid_t original_uid, gid_t original_gid)
+{
+ int fd = open64 ("/proc/self/uid_map", O_WRONLY);
+ if (fd < 0)
+ {
+ printf ("warning: could not open /proc/self/uid_map: %m\n"
+ "warning: file creation may not be possible\n");
+ return;
+ }
+
+ /* We map our original UID to the same UID in the container so we
+ own our own files normally. Without that, file creation could
+ fail with EOVERFLOW (sic!). */
+ char buf[100];
+ int ret = snprintf (buf, sizeof (buf), "%llu %llu 1\n",
+ (unsigned long long) original_uid,
+ (unsigned long long) original_uid);
+ TEST_VERIFY_EXIT (ret < sizeof (buf));
+ xwrite (fd, buf, ret);
+ xclose (fd);
+
+ /* Disable setgroups before mapping groups, otherwise that would
+ fail with EPERM. */
+ fd = xopen ("/proc/self/setgroups", O_WRONLY, 0);
+ xwrite (fd, "deny\n", strlen ("deny\n"));
+ xclose (fd);
+
+ /* Now map our own GID, like we did for the user ID. */
+ fd = xopen ("/proc/self/gid_map", O_WRONLY, 0);
+ ret = snprintf (buf, sizeof (buf), "%llu %llu 1\n",
+ (unsigned long long) original_gid,
+ (unsigned long long) original_gid);
+ TEST_VERIFY_EXIT (ret < sizeof (buf));
+ xwrite (fd, buf, ret);
+ xclose (fd);
+}
+#endif /* CLONE_NEWUSER */
+
bool
support_become_root (void)
{
#ifdef CLONE_NEWUSER
+ uid_t original_uid = getuid ();
+ gid_t original_gid = getgid ();
+
if (unshare (CLONE_NEWUSER | CLONE_NEWNS) == 0)
- /* Even if we do not have UID zero, we have extended privileges at
- this point. */
- return true;
+ {
+ setup_uid_gid_mapping (original_uid, original_gid);
+ /* Even if we do not have UID zero, we have extended privileges at
+ this point. */
+ return true;
+ }
#endif
if (setuid (0) != 0)
{
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] support_become_root: Enable file creation in user namespaces
2017-11-17 14:55 [PATCH] support_become_root: Enable file creation in user namespaces Florian Weimer
@ 2017-11-17 17:20 ` Jonathan Nieder
0 siblings, 0 replies; 2+ messages in thread
From: Jonathan Nieder @ 2017-11-17 17:20 UTC (permalink / raw)
To: Florian Weimer; +Cc: libc-alpha
Hi,
Florian Weimer wrote:
> Without UID/GID maps, file creation will file with EOVERFLOW.
>
> This patch is based on DJ Delorie's work on container testing.
>
> 2017-11-17 Florian Weimer <fweimer@redhat.com>
>
> support_become_root: Enable file creation in namespaces.
> * support/support_become_root.c (setup_mapping): New function.
> (support_become_root): Call it.
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Thank you.
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2017-11-17 17:20 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-11-17 14:55 [PATCH] support_become_root: Enable file creation in user namespaces Florian Weimer
2017-11-17 17:20 ` Jonathan Nieder
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).