generic: add test for detached mount propagation
authorChristian Brauner <christian.brauner@ubuntu.com>
Mon, 12 Apr 2021 11:57:03 +0000 (13:57 +0200)
committerEryu Guan <guaneryu@gmail.com>
Sun, 18 Apr 2021 12:53:53 +0000 (20:53 +0800)
Regression test to verify that creating a series of detached mounts,
attaching them to the filesystem, and unmounting them does not trigger an
integer overflow in ns->mounts causing the kernel to block any new mounts in
count_mounts() and returning ENOSPC because it falsely assumes that the
maximum number of mounts in the mount namespace has been reached, i.e. it
thinks it can't fit the new mounts into the mount namespace anymore.

The test is written in a way that it will leave the host's mount
namespace intact so we are sure to never make the host's mount namespace
unuseable!

Link: https://git.kernel.org/torvalds/c/ee2e3f50629f17b0752b55b2566c15ce8dafb557
Cc: Eryu Guan <guan@eryu.me>
Cc: David Howells <dhowells@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
.gitignore
src/Makefile
src/detached_mounts_propagation.c [new file with mode: 0644]
tests/generic/632 [new file with mode: 0755]
tests/generic/632.out [new file with mode: 0644]
tests/generic/group

index f3bc0a4f444b38640b1ec53dcb76797828955e9f..72bd40a0dfcc966ddb55fd260bed81df6cc11aa8 100644 (file)
@@ -62,6 +62,7 @@
 /src/cloner
 /src/dbtest
 /src/deduperace
+/src/detached_mounts_propagation
 /src/devzero
 /src/dio-interleaved
 /src/dio-invalidate-cache
index 3d729a3411e36a2bae9d4e2a4971292bc902f244..99019e6ef5beedd2eef86c53d72e230f5b04b9ae 100644 (file)
@@ -29,7 +29,8 @@ LINUX_TARGETS = xfsctl bstat t_mtab getdevicesize preallo_rw_pattern_reader \
        attr-list-by-handle-cursor-test listxattr dio-interleaved t_dir_type \
        dio-invalidate-cache stat_test t_encrypted_d_revalidate \
        attr_replace_test swapon mkswap t_attr_corruption t_open_tmpfiles \
-       fscrypt-crypt-util bulkstat_null_ocount splice-test chprojid_fail
+       fscrypt-crypt-util bulkstat_null_ocount splice-test chprojid_fail \
+       detached_mounts_propagation
 
 SUBDIRS = log-writes perf
 
diff --git a/src/detached_mounts_propagation.c b/src/detached_mounts_propagation.c
new file mode 100644 (file)
index 0000000..4a588e4
--- /dev/null
@@ -0,0 +1,252 @@
+/* SPDX-License-Identifier: LGPL-2.1+ */
+/*
+ * Copyright (c) 2021 Christian Brauner <christian.brauner@ubuntu.com>
+ * All Rights Reserved.
+ *
+ * Regression test to verify that creating a series of detached mounts,
+ * attaching them to the filesystem, and unmounting them does not trigger an
+ * integer overflow in ns->mounts causing the kernel to block any new mounts in
+ * count_mounts() and returning ENOSPC because it falsely assumes that the
+ * maximum number of mounts in the mount namespace has been reached, i.e. it
+ * thinks it can't fit the new mounts into the mount namespace anymore.
+ */
+
+#include <errno.h>
+#include <fcntl.h>
+#include <getopt.h>
+#include <limits.h>
+#include <sched.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/mount.h>
+#include <sys/stat.h>
+#include <sys/syscall.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+/* open_tree() */
+#ifndef OPEN_TREE_CLONE
+#define OPEN_TREE_CLONE 1
+#endif
+
+#ifndef OPEN_TREE_CLOEXEC
+#define OPEN_TREE_CLOEXEC O_CLOEXEC
+#endif
+
+#ifndef __NR_open_tree
+       #if defined __alpha__
+               #define __NR_open_tree 538
+       #elif defined _MIPS_SIM
+               #if _MIPS_SIM == _MIPS_SIM_ABI32        /* o32 */
+                       #define __NR_open_tree 4428
+               #endif
+               #if _MIPS_SIM == _MIPS_SIM_NABI32       /* n32 */
+                       #define __NR_open_tree 6428
+               #endif
+               #if _MIPS_SIM == _MIPS_SIM_ABI64        /* n64 */
+                       #define __NR_open_tree 5428
+               #endif
+       #elif defined __ia64__
+               #define __NR_open_tree (428 + 1024)
+       #else
+               #define __NR_open_tree 428
+       #endif
+#endif
+
+/* move_mount() */
+#ifndef MOVE_MOUNT_F_EMPTY_PATH
+#define MOVE_MOUNT_F_EMPTY_PATH 0x00000004 /* Empty from path permitted */
+#endif
+
+#ifndef __NR_move_mount
+       #if defined __alpha__
+               #define __NR_move_mount 539
+       #elif defined _MIPS_SIM
+               #if _MIPS_SIM == _MIPS_SIM_ABI32        /* o32 */
+                       #define __NR_move_mount 4429
+               #endif
+               #if _MIPS_SIM == _MIPS_SIM_NABI32       /* n32 */
+                       #define __NR_move_mount 6429
+               #endif
+               #if _MIPS_SIM == _MIPS_SIM_ABI64        /* n64 */
+                       #define __NR_move_mount 5429
+               #endif
+       #elif defined __ia64__
+               #define __NR_move_mount (428 + 1024)
+       #else
+               #define __NR_move_mount 429
+       #endif
+#endif
+
+static inline int sys_open_tree(int dfd, const char *filename, unsigned int flags)
+{
+       return syscall(__NR_open_tree, dfd, filename, flags);
+}
+
+static inline int sys_move_mount(int from_dfd, const char *from_pathname, int to_dfd,
+                                const char *to_pathname, unsigned int flags)
+{
+       return syscall(__NR_move_mount, from_dfd, from_pathname, to_dfd, to_pathname, flags);
+}
+
+static bool is_shared_mountpoint(const char *path)
+{
+       bool shared = false;
+       FILE *f = NULL;
+       char *line = NULL;
+       int i;
+       size_t len = 0;
+
+       f = fopen("/proc/self/mountinfo", "re");
+       if (!f)
+               return 0;
+
+       while (getline(&line, &len, f) > 0) {
+               char *slider1, *slider2;
+
+               for (slider1 = line, i = 0; slider1 && i < 4; i++)
+                       slider1 = strchr(slider1 + 1, ' ');
+
+               if (!slider1)
+                       continue;
+
+               slider2 = strchr(slider1 + 1, ' ');
+               if (!slider2)
+                       continue;
+
+               *slider2 = '\0';
+               if (strcmp(slider1 + 1, path) == 0) {
+                       /* This is the path. Is it shared? */
+                       slider1 = strchr(slider2 + 1, ' ');
+                       if (slider1 && strstr(slider1, "shared:")) {
+                               shared = true;
+                               break;
+                       }
+               }
+       }
+       fclose(f);
+       free(line);
+
+       return shared;
+}
+
+static void usage(void)
+{
+       const char *text = "mount-new [--recursive] <base-dir>\n";
+       fprintf(stderr, "%s", text);
+       _exit(EXIT_SUCCESS);
+}
+
+#define exit_usage(format, ...)                              \
+       ({                                                   \
+               fprintf(stderr, format "\n", ##__VA_ARGS__); \
+               usage();                                     \
+       })
+
+#define exit_log(format, ...)                                \
+       ({                                                   \
+               fprintf(stderr, format "\n", ##__VA_ARGS__); \
+               exit(EXIT_FAILURE);                          \
+       })
+
+static const struct option longopts[] = {
+       {"help",        no_argument,            0,      'a'},
+       { NULL,         no_argument,            0,       0 },
+};
+
+int main(int argc, char *argv[])
+{
+       int exit_code = EXIT_SUCCESS, index = 0;
+       int dfd, fd_tree, new_argc, ret;
+       char *base_dir;
+       char *const *new_argv;
+       char target[PATH_MAX];
+
+       while ((ret = getopt_long_only(argc, argv, "", longopts, &index)) != -1) {
+               switch (ret) {
+               case 'a':
+                       /* fallthrough */
+               default:
+                       usage();
+               }
+       }
+
+       new_argv = &argv[optind];
+       new_argc = argc - optind;
+       if (new_argc < 1)
+               exit_usage("Missing base directory\n");
+       base_dir = new_argv[0];
+
+       if (*base_dir != '/')
+               exit_log("Please specify an absolute path");
+
+       /* Ensure that target is a shared mountpoint. */
+       if (!is_shared_mountpoint(base_dir))
+               exit_log("Please ensure that \"%s\" is a shared mountpoint", base_dir);
+
+       ret = unshare(CLONE_NEWNS);
+       if (ret < 0)
+               exit_log("%m - Failed to create new mount namespace");
+
+       ret = mount(NULL, base_dir, NULL, MS_REC | MS_SHARED, NULL);
+       if (ret < 0)
+               exit_log("%m - Failed to make base_dir shared mountpoint");
+
+       dfd = open(base_dir, O_RDONLY | O_DIRECTORY | O_CLOEXEC);
+       if (dfd < 0)
+               exit_log("%m - Failed to open base directory \"%s\"", base_dir);
+
+       ret = mkdirat(dfd, "detached-move-mount", 0755);
+       if (ret < 0)
+               exit_log("%m - Failed to create required temporary directories");
+
+       ret = snprintf(target, sizeof(target), "%s/detached-move-mount", base_dir);
+       if (ret < 0 || (size_t)ret >= sizeof(target))
+               exit_log("%m - Failed to assemble target path");
+
+       /*
+        * Having a mount table with 10000 mounts is already quite excessive
+        * and shoult account even for weird test systems.
+        */
+       for (size_t i = 0; i < 10000; i++) {
+               fd_tree = sys_open_tree(dfd, "detached-move-mount",
+                                       OPEN_TREE_CLONE |
+                                       OPEN_TREE_CLOEXEC |
+                                       AT_EMPTY_PATH);
+               if (fd_tree < 0) {
+                       if (errno == ENOSYS) /* New mount API not (fully) supported. */
+                               break;
+
+                       fprintf(stderr, "%m - Failed to open %d(detached-move-mount)", dfd);
+                       exit_code = EXIT_FAILURE;
+                       break;
+               }
+
+               ret = sys_move_mount(fd_tree, "", dfd, "detached-move-mount", MOVE_MOUNT_F_EMPTY_PATH);
+               if (ret < 0) {
+                       if (errno == ENOSPC)
+                               fprintf(stderr, "%m - Buggy mount counting");
+                       else if (errno == ENOSYS) /* New mount API not (fully) supported. */
+                               break;
+                       else
+                               fprintf(stderr, "%m - Failed to attach mount to %d(detached-move-mount)", dfd);
+                       exit_code = EXIT_FAILURE;
+                       break;
+               }
+               close(fd_tree);
+
+               ret = umount2(target, MNT_DETACH);
+               if (ret < 0) {
+                       fprintf(stderr, "%m - Failed to unmount %s", target);
+                       exit_code = EXIT_FAILURE;
+                       break;
+               }
+       }
+
+       (void)unlinkat(dfd, "detached-move-mount", AT_REMOVEDIR);
+       close(dfd);
+
+       exit(exit_code);
+}
diff --git a/tests/generic/632 b/tests/generic/632
new file mode 100755 (executable)
index 0000000..206b6c1
--- /dev/null
@@ -0,0 +1,46 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0+
+#
+# Copyright (c) 2021 Christian Brauner <christian.brauner@ubuntu.com>
+# All Rights Reserved.
+#
+# FS QA Test No. 632
+#
+# Regression test to verify that creating a series of detached mounts,
+# attaching them to the filesystem, and unmounting them does not trigger an
+# integer overflow in ns->mounts causing the kernel to block any new mounts in
+# count_mounts() and returning ENOSPC because it falsely assumes that the
+# maximum number of mounts in the mount namespace has been reached, i.e. it
+# thinks it can't fit the new mounts into the mount namespace anymore.
+#
+# Kernel commit ee2e3f50629f ("mount: fix mounting of detached mounts onto
+# targets that reside on shared mounts") fixed the bug.
+#
+seq=`basename $0`
+seqres=$RESULT_DIR/$seq
+echo "QA output created by $seq"
+
+here=`pwd`
+tmp=/tmp/$$
+status=1       # failure is the default!
+trap "_cleanup; exit \$status" 0 1 2 3 15
+
+_cleanup()
+{
+       cd /
+       rm -f $tmp.*
+}
+
+. ./common/rc
+
+rm -f $seqres.full
+
+_supported_fs generic
+_require_test
+_require_test_program "detached_mounts_propagation"
+
+$here/src/detached_mounts_propagation $TEST_DIR >> $seqres.full
+status=$?
+
+echo silence is golden
+exit
diff --git a/tests/generic/632.out b/tests/generic/632.out
new file mode 100644 (file)
index 0000000..3bba60f
--- /dev/null
@@ -0,0 +1,2 @@
+QA output created by 632
+silence is golden
index ab00cc044f4ca043e348d7994b5c5f9ffafe612d..3c32e0b3412070a087986d6fdb20ccba320905d1 100644 (file)
 629 auto quick rw copy_range
 630 auto quick rw dedupe clone
 631 auto rw overlay rename
+632 auto quick mount