Merge pull request #15566 from poettering/destroy-binfmt

unregister binary formats during shutdown

man/systemd-binfmt.service.xml

@@ -41,6 +41,14 @@
 
   <refsect1><title>Options</title>
     <variablelist>
+
+      <varlistentry>
+        <term><option>--unregister</option></term>
+        <listitem><para>If passed, instead of registering configured binary formats in the kernel, the
+        reverse operation is executed: all currently registered binary formats are unregistered from the
+        kernel.</para></listitem>
+      </varlistentry>
+
       <xi:include href="standard-options.xml" xpointer="cat-config" />
       <xi:include href="standard-options.xml" xpointer="no-pager" />
       <xi:include href="standard-options.xml" xpointer="help" />

src/binfmt/binfmt.c

@@ -10,6 +10,7 @@
 #include <sys/types.h>
 
 #include "alloc-util.h"
+#include "binfmt-util.h"
 #include "conf-files.h"
 #include "def.h"
 #include "fd-util.h"
@@ -24,6 +25,7 @@
 
 static bool arg_cat_config = false;
 static PagerFlags arg_pager_flags = 0;
+static bool arg_unregister = false;
 
 static int delete_rule(const char *rule) {
         _cleanup_free_ char *x = NULL, *fn = NULL;
@@ -32,18 +34,17 @@ static int delete_rule(const char *rule) {
         assert(rule);
         assert(rule[0]);
 
-        x = strdup(rule);
+        e = strchrnul(rule + 1, rule[0]);
+        x = strndup(rule + 1, e - rule - 1);
         if (!x)
                 return log_oom();
 
-        e = strchrnul(x+1, x[0]);
-        *e = 0;
-
-        if (!filename_is_valid(x + 1))
+        if (!filename_is_valid(x) ||
+            STR_IN_SET(x, "register", "status"))
                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
-                                       "Rule file name '%s' is not valid, refusing.", x + 1);
+                                       "Rule file name '%s' is not valid, refusing.", x);
 
-        fn = path_join("/proc/sys/fs/binfmt_misc", x+1);
+        fn = path_join("/proc/sys/fs/binfmt_misc", x);
         if (!fn)
                 return log_oom();
 
@@ -116,6 +117,7 @@ static int help(void) {
                "     --version          Show package version\n"
                "     --cat-config       Show configuration files\n"
                "     --no-pager         Do not pipe output into a pager\n"
+               "     --unregister       Unregister all existing entries\n"
                "\nSee the %s for details.\n"
                , program_invocation_short_name
                , link
@@ -129,6 +131,7 @@ static int parse_argv(int argc, char *argv[]) {
                 ARG_VERSION = 0x100,
                 ARG_CAT_CONFIG,
                 ARG_NO_PAGER,
+                ARG_UNREGISTER,
         };
 
         static const struct option options[] = {
@@ -136,6 +139,7 @@ static int parse_argv(int argc, char *argv[]) {
                 { "version",    no_argument, NULL, ARG_VERSION    },
                 { "cat-config", no_argument, NULL, ARG_CAT_CONFIG },
                 { "no-pager",   no_argument, NULL, ARG_NO_PAGER   },
+                { "unregister", no_argument, NULL, ARG_UNREGISTER },
                 {}
         };
 
@@ -162,6 +166,10 @@ static int parse_argv(int argc, char *argv[]) {
                         arg_pager_flags |= PAGER_DISABLE;
                         break;
 
+                case ARG_UNREGISTER:
+                        arg_unregister = true;
+                        break;
+
                 case '?':
                         return -EINVAL;
 
@@ -169,9 +177,9 @@ static int parse_argv(int argc, char *argv[]) {
                         assert_not_reached("Unhandled option");
                 }
 
-        if (arg_cat_config && argc > optind)
+        if ((arg_unregister || arg_cat_config) && argc > optind)
                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
-                                       "Positional arguments are not allowed with --cat-config");
+                                       "Positional arguments are not allowed with --cat-config or --unregister");
 
         return 1;
 }
@@ -189,6 +197,9 @@ static int run(int argc, char *argv[]) {
 
         r = 0;
 
+        if (arg_unregister)
+                return disable_binfmt();
+
         if (argc > optind) {
                 int i;
 

src/shared/binfmt-util.c

@@ -0,0 +1,33 @@
+#include <sys/stat.h>
+#include <sys/statvfs.h>
+#include <sys/vfs.h>
+
+#include "binfmt-util.h"
+#include "fileio.h"
+#include "missing_magic.h"
+#include "stat-util.h"
+
+int disable_binfmt(void) {
+        int r;
+
+        /* Flush out all rules. This is important during shutdown to cover for rules using "F", since those
+         * might pin a file and thus block us from unmounting stuff cleanly.
+         *
+         * We are a bit careful here, since binfmt_misc might still be an autofs which we don't want to
+         * trigger. */
+
+        r = path_is_fs_type("/proc/sys/fs/binfmt_misc", BINFMTFS_MAGIC);
+        if (r == 0 || r == -ENOENT) {
+                log_debug("binfmt_misc is not mounted, not detaching entries.");
+                return 0;
+        }
+        if (r < 0)
+                return log_warning_errno(r, "Failed to determine whether binfmt_misc is mounted: %m");
+
+        r = write_string_file("/proc/sys/fs/binfmt_misc/status", "-1", WRITE_STRING_FILE_DISABLE_BUFFER);
+        if (r < 0)
+                return log_warning_errno(r, "Failed to unregister binfmt_misc entries: %m");
+
+        log_debug("Unregistered all remaining binfmt_misc entries.");
+        return 0;
+}

src/shared/binfmt-util.h

@@ -0,0 +1,4 @@
+/* SPDX-License-Identifier: LGPL-2.1+ */
+#pragma once
+
+int disable_binfmt(void);

src/shared/meson.build

@@ -12,6 +12,8 @@ shared_sources = files('''
         barrier.h
         base-filesystem.c
         base-filesystem.h
+        binfmt-util.c
+        binfmt-util.h
         bitmap.c
         bitmap.h
         blkid-util.h

src/shutdown/shutdown.c

@@ -16,6 +16,7 @@
 
 #include "alloc-util.h"
 #include "async.h"
+#include "binfmt-util.h"
 #include "cgroup-setup.h"
 #include "cgroup-util.h"
 #include "def.h"
@@ -386,6 +387,7 @@ int main(int argc, char *argv[]) {
                 sync_with_progress();
 
         disable_coredumps();
+        disable_binfmt();
 
         log_info("Sending SIGTERM to remaining processes...");
         broadcast_signal(SIGTERM, true, true, arg_timeout);

units/systemd-binfmt.service.in

@@ -28,4 +28,5 @@ ConditionDirectoryNotEmpty=|/run/binfmt.d
 Type=oneshot
 RemainAfterExit=yes
 ExecStart=@rootlibexecdir@/systemd-binfmt
+ExecStop=@rootlibexecdir@/systemd-binfmt --unregister
 TimeoutSec=90s