commit 2fe64148a81f0d78050c302f34a6853d21f7cae4

Author: DJ Delorie <dj@redhat.com>

Date:   Mon Mar 28 23:53:33 2022 -0400

    Allow for unpriviledged nested containers

    If the build itself is run in a container, we may not be able to

    fully set up a nested container for test-container testing.

    Notably is the mounting of /proc, since it's critical that it

    be mounted from within the same PID namespace as its users, and

    thus cannot be bind mounted from outside the container like other

    mounts.

    This patch defaults to using the parent's PID namespace instead of

    creating a new one, as this is more likely to be allowed.

    If the test needs an isolated PID namespace, it should add the "pidns"

    command to its init script.

    Reviewed-by: Carlos O'Donell <carlos@redhat.com>

Conflicts:

	nss/tst-reload2.c

          (not in RHEL-8)

	support/Makefile

          (RHEL-8 missing some routines in libsupport-routines)

diff --git a/elf/tst-pldd.c b/elf/tst-pldd.c

index f381cb0fa7e6b93d..45ac033a0f897088 100644

--- a/elf/tst-pldd.c

+++ b/elf/tst-pldd.c

@@ -85,6 +85,8 @@ in_str_list (const char *libname, const char *const strlist[])

 static int

 do_test (void)

 {

+  support_need_proc ("needs /proc/sys/kernel/yama/ptrace_scope and /proc/$child");

+

   /* Check if our subprocess can be debugged with ptrace.  */

   {

     int ptrace_scope = support_ptrace_scope ();

diff --git a/nptl/tst-pthread-getattr.c b/nptl/tst-pthread-getattr.c

index 273b6073abe9cb60..f1c0b39f3a27724c 100644

--- a/nptl/tst-pthread-getattr.c

+++ b/nptl/tst-pthread-getattr.c

@@ -28,6 +28,8 @@

 #include <unistd.h>

 #include <inttypes.h>

+#include <support/support.h>

+

 /* There is an obscure bug in the kernel due to which RLIMIT_STACK is sometimes

    returned as unlimited when it is not, which may cause this test to fail.

    There is also the other case where RLIMIT_STACK is intentionally set as

@@ -152,6 +154,8 @@ check_stack_top (void)

 static int

 do_test (void)

 {

+  support_need_proc ("Reads /proc/self/maps to get stack size.");

+

   pagesize = sysconf (_SC_PAGESIZE);

   return check_stack_top ();

 }

diff --git a/support/Makefile b/support/Makefile

index 636d69c4f8e7e139..e184fccbe7d2310c 100644

--- a/support/Makefile

+++ b/support/Makefile

@@ -59,6 +59,7 @@ libsupport-routines = \

   support_format_hostent \

   support_format_netent \

   support_isolate_in_subprocess \

+  support_need_proc \

   support_process_state \

   support_ptrace \

   support_openpty \

diff --git a/support/support.h b/support/support.h

index 96833bd4e992e6d3..1466eb29f840fa59 100644

--- a/support/support.h

+++ b/support/support.h

@@ -81,6 +81,11 @@ char *support_quote_string (const char *);

    regular file open for writing, and initially empty.  */

 int support_descriptor_supports_holes (int fd);

+/* Predicates that a test requires a working /proc filesystem.  This

+   call will exit with UNSUPPORTED if /proc is not available, printing

+   WHY_MSG as part of the diagnostic.  */

+void support_need_proc (const char *why_msg);

+

 /* Error-checking wrapper functions which terminate the process on

    error.  */

diff --git a/support/support_need_proc.c b/support/support_need_proc.c

new file mode 100644

index 0000000000000000..9b4eab7539b2d6c3

--- /dev/null

+++ b/support/support_need_proc.c

@@ -0,0 +1,35 @@

+/* Indicate that a test requires a working /proc.

+   Copyright (C) 2022 Free Software Foundation, Inc.

+   This file is part of the GNU C Library.

+

+   The GNU C Library is free software; you can redistribute it and/or

+   modify it under the terms of the GNU Lesser General Public

+   License as published by the Free Software Foundation; either

+   version 2.1 of the License, or (at your option) any later version.

+

+   The GNU C Library is distributed in the hope that it will be useful,

+   but WITHOUT ANY WARRANTY; without even the implied warranty of

+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

+   Lesser General Public License for more details.

+

+   You should have received a copy of the GNU Lesser General Public

+   License along with the GNU C Library; if not, see

+   <https://www.gnu.org/licenses/>.  */

+

+#include <unistd.h>

+#include <support/check.h>

+#include <support/support.h>

+

+/* We test for /proc/self/maps since that's one of the files that one

+   of our tests actually uses, but the general idea is if Linux's

+   /proc/ (procfs) filesystem is mounted.  If not, the process exits

+   with an UNSUPPORTED result code.  */

+

+void

+support_need_proc (const char *why_msg)

+{

+#ifdef __linux__

+  if (access ("/proc/self/maps", R_OK))

+    FAIL_UNSUPPORTED ("/proc is not available, %s", why_msg);

+#endif

+}

diff --git a/support/test-container.c b/support/test-container.c

index 9975c8cb7bc9a955..2bce4db841ff7668 100644

--- a/support/test-container.c

+++ b/support/test-container.c

@@ -95,6 +95,7 @@ int verbose = 0;

    * mytest.root/mytest.script has a list of "commands" to run:

        syntax:

          # comment

+	 pidns <comment>

          su

          mv FILE FILE

 	 cp FILE FILE

@@ -120,6 +121,8 @@ int verbose = 0;

        details:

          - '#': A comment.

+	 - 'pidns': Require a separate PID namespace, prints comment if it can't

+	    (default is a shared pid namespace)

          - 'su': Enables running test as root in the container.

          - 'mv': A minimal move files command.

          - 'cp': A minimal copy files command.

@@ -143,7 +146,7 @@ int verbose = 0;

    * Simple, easy to review code (i.e. prefer simple naive code over

      complex efficient code)

-   * The current implementation ist parallel-make-safe, but only in

+   * The current implementation is parallel-make-safe, but only in

      that it uses a lock to prevent parallel access to the testroot.  */

@@ -222,11 +225,37 @@ concat (const char *str, ...)

   return bufs[n];

 }

+/* Like the above, but put spaces between words.  Caller frees.  */

+static char *

+concat_words (char **words, int num_words)

+{

+  int len = 0;

+  int i;

+  char *rv, *p;

+

+  for (i = 0; i < num_words; i ++)

+    {

+      len += strlen (words[i]);

+      len ++;

+    }

+

+  p = rv = (char *) xmalloc (len);

+

+  for (i = 0; i < num_words; i ++)

+    {

+      if (i > 0)

+	p = stpcpy (p, " ");

+      p = stpcpy (p, words[i]);

+    }

+

+  return rv;

+}

+

 /* Try to mount SRC onto DEST.  */

 static void

 trymount (const char *src, const char *dest)

 {

-  if (mount (src, dest, "", MS_BIND, NULL) < 0)

+  if (mount (src, dest, "", MS_BIND | MS_REC, NULL) < 0)

     FAIL_EXIT1 ("can't mount %s onto %s\n", src, dest);

 }

@@ -709,6 +738,9 @@ main (int argc, char **argv)

   gid_t original_gid;

   /* If set, the test runs as root instead of the user running the testsuite.  */

   int be_su = 0;

+  int require_pidns = 0;

+  const char *pidns_comment = NULL;

+  int do_proc_mounts = 0;

   int UMAP;

   int GMAP;

   /* Used for "%lld %lld 1" so need not be large.  */

@@ -991,6 +1023,12 @@ main (int argc, char **argv)

 	      {

 		be_su = 1;

 	      }

+	    else if (nt >= 1 && strcmp (the_words[0], "pidns") == 0)

+	      {

+		require_pidns = 1;

+		if (nt > 1)

+		  pidns_comment = concat_words (the_words + 1, nt - 1);

+	      }

 	    else if (nt == 3 && strcmp (the_words[0], "mkdirp") == 0)

 	      {

 		long int m;

@@ -1048,7 +1086,8 @@ main (int argc, char **argv)

 #ifdef CLONE_NEWNS

   /* The unshare here gives us our own spaces and capabilities.  */

-  if (unshare (CLONE_NEWUSER | CLONE_NEWPID | CLONE_NEWNS) < 0)

+  if (unshare (CLONE_NEWUSER | CLONE_NEWNS

+	       | (require_pidns ? CLONE_NEWPID : 0)) < 0)

     {

       /* Older kernels may not support all the options, or security

 	 policy may block this call.  */

@@ -1059,6 +1098,11 @@ main (int argc, char **argv)

 	    check_for_unshare_hints ();

 	  FAIL_UNSUPPORTED ("unable to unshare user/fs: %s", strerror (saved_errno));

 	}

+      /* We're about to exit anyway, it's "safe" to call unshare again

+	 just to see if the CLONE_NEWPID caused the error.  */

+      else if (require_pidns && unshare (CLONE_NEWUSER | CLONE_NEWNS) >= 0)

+	FAIL_EXIT1 ("unable to unshare pid ns: %s : %s", strerror (errno),

+		    pidns_comment ? pidns_comment : "required by test");

       else

 	FAIL_EXIT1 ("unable to unshare user/fs: %s", strerror (errno));

     }

@@ -1074,6 +1118,15 @@ main (int argc, char **argv)

   trymount (support_srcdir_root, new_srcdir_path);

   trymount (support_objdir_root, new_objdir_path);

+  /* It may not be possible to mount /proc directly.  */

+  if (! require_pidns)

+  {

+    char *new_proc = concat (new_root_path, "/proc", NULL);

+    xmkdirp (new_proc, 0755);

+    trymount ("/proc", new_proc);

+    do_proc_mounts = 1;

+  }

+

   xmkdirp (concat (new_root_path, "/dev", NULL), 0755);

   devmount (new_root_path, "null");

   devmount (new_root_path, "zero");

@@ -1136,42 +1189,60 @@ main (int argc, char **argv)

   maybe_xmkdir ("/tmp", 0755);

-  /* Now that we're pid 1 (effectively "root") we can mount /proc  */

-  maybe_xmkdir ("/proc", 0777);

-  if (mount ("proc", "/proc", "proc", 0, NULL) < 0)

-    FAIL_EXIT1 ("Unable to mount /proc: ");

-

-  /* We map our original UID to the same UID in the container so we

-     can own our own files normally.  */

-  UMAP = open ("/proc/self/uid_map", O_WRONLY);

-  if (UMAP < 0)

-    FAIL_EXIT1 ("can't write to /proc/self/uid_map\n");

-

-  sprintf (tmp, "%lld %lld 1\n",

-	   (long long) (be_su ? 0 : original_uid), (long long) original_uid);

-  write (UMAP, tmp, strlen (tmp));

-  xclose (UMAP);

-

-  /* We must disable setgroups () before we can map our groups, else we

-     get EPERM.  */

-  GMAP = open ("/proc/self/setgroups", O_WRONLY);

-  if (GMAP >= 0)

+  if (require_pidns)

     {

-      /* We support kernels old enough to not have this.  */

-      write (GMAP, "deny\n", 5);

-      xclose (GMAP);

+      /* Now that we're pid 1 (effectively "root") we can mount /proc  */

+      maybe_xmkdir ("/proc", 0777);

+      if (mount ("proc", "/proc", "proc", 0, NULL) != 0)

+	{

+	  /* This happens if we're trying to create a nested container,

+	     like if the build is running under podman, and we lack

+	     priviledges.

+

+	     Ideally we would WARN here, but that would just add noise to

+	     *every* test-container test, and the ones that care should

+	     have their own relevent diagnostics.

+

+	     FAIL_EXIT1 ("Unable to mount /proc: ");  */

+	}

+      else

+	do_proc_mounts = 1;

     }

-  /* We map our original GID to the same GID in the container so we

-     can own our own files normally.  */

-  GMAP = open ("/proc/self/gid_map", O_WRONLY);

-  if (GMAP < 0)

-    FAIL_EXIT1 ("can't write to /proc/self/gid_map\n");

+  if (do_proc_mounts)

+    {

+      /* We map our original UID to the same UID in the container so we

+	 can own our own files normally.  */

+      UMAP = open ("/proc/self/uid_map", O_WRONLY);

+      if (UMAP < 0)

+	FAIL_EXIT1 ("can't write to /proc/self/uid_map\n");

+

+      sprintf (tmp, "%lld %lld 1\n",

+	       (long long) (be_su ? 0 : original_uid), (long long) original_uid);

+      write (UMAP, tmp, strlen (tmp));

+      xclose (UMAP);

+

+      /* We must disable setgroups () before we can map our groups, else we

+	 get EPERM.  */

+      GMAP = open ("/proc/self/setgroups", O_WRONLY);

+      if (GMAP >= 0)

+	{

+	  /* We support kernels old enough to not have this.  */

+	  write (GMAP, "deny\n", 5);

+	  xclose (GMAP);

+	}

-  sprintf (tmp, "%lld %lld 1\n",

-	   (long long) (be_su ? 0 : original_gid), (long long) original_gid);

-  write (GMAP, tmp, strlen (tmp));

-  xclose (GMAP);

+      /* We map our original GID to the same GID in the container so we

+	 can own our own files normally.  */

+      GMAP = open ("/proc/self/gid_map", O_WRONLY);

+      if (GMAP < 0)

+	FAIL_EXIT1 ("can't write to /proc/self/gid_map\n");

+

+      sprintf (tmp, "%lld %lld 1\n",

+	       (long long) (be_su ? 0 : original_gid), (long long) original_gid);

+      write (GMAP, tmp, strlen (tmp));

+      xclose (GMAP);

+    }

   if (change_cwd)

     {