Re: [PATCH v3] vfs: introduce UMOUNT_WAIT to wait for delayed_fput/mntput completion

From: Jaegeuk Kim
Date: Thu Sep 21 2017 - 14:20:33 EST


This patch introduces UMOUNT_WAIT flag for umount(2) which let user wait
for the completion of delayed fput()/mntput() explicitly.

Espcially, it can be used for Android reboot procedure, and resolve the below
issue where a kernel panic happens when a living filesystem tries to access
dead block device after device_shutdown done by kernel_restart.

Term: namespace(mnt_get_count())

1. create_new_namespaces() creates ns1 and ns2,

/data(1) ns1(1) ns2(1)
| | |
---------------------
|
sb->s_active = 3

2. after binder_proc_clear_zombies() for ns2 and ns1 triggers
- delayed_fput()
- delayed_mntput_work(ns2)

/data(1) ns1(1)
| |
----------
|
sb->s_active = 2

3. umount() for /data is successed.

ns1(1)
|
sb->s_active = 1

4. device_shutdown() by init

5. - delayed_mntput_work(ns1)
- put_super(), since sb->s_active = 0
- -EIO

Cc: Al Viro <viro@xxxxxxxxxxxxxxxxxx>
Signed-off-by: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
---
fs/file_table.c | 6 ++++++
fs/namespace.c | 26 +++++++++++++++++++++++++-
include/linux/file.h | 1 +
include/linux/fs.h | 1 +
4 files changed, 33 insertions(+), 1 deletion(-)

diff --git a/fs/file_table.c b/fs/file_table.c
index 72e861a35a7f..35b32ffdb934 100644
--- a/fs/file_table.c
+++ b/fs/file_table.c
@@ -263,6 +263,12 @@ void flush_delayed_fput(void)

static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);

+void flush_delayed_fput_wait(void)
+{
+ delayed_fput(NULL);
+ flush_delayed_work(&delayed_fput_work);
+}
+
void fput(struct file *file)
{
if (atomic_long_dec_and_test(&file->f_count)) {
diff --git a/fs/namespace.c b/fs/namespace.c
index f8893dc6a989..e2586a38c83c 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -21,6 +21,7 @@
#include <linux/fs_struct.h> /* get_fs_root et.al. */
#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
#include <linux/uaccess.h>
+#include <linux/file.h>
#include <linux/proc_ns.h>
#include <linux/magic.h>
#include <linux/bootmem.h>
@@ -1133,6 +1134,12 @@ static void delayed_mntput(struct work_struct *unused)
}
static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);

+void flush_delayed_mntput_wait(void)
+{
+ delayed_mntput(NULL);
+ flush_delayed_work(&delayed_mntput_work);
+}
+
static void mntput_no_expire(struct mount *mnt)
{
rcu_read_lock();
@@ -1629,7 +1636,8 @@ SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
int retval;
int lookup_flags = 0;

- if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
+ if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW |
+ UMOUNT_WAIT))
return -EINVAL;

if (!may_mount())
@@ -1653,11 +1661,27 @@ SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
goto dput_and_out;

+ /* flush delayed_fput to put mnt_count */
+ if (flags & UMOUNT_WAIT)
+ flush_delayed_fput_wait();
+
retval = do_umount(mnt, flags);
dput_and_out:
/* we mustn't call path_put() as that would clear mnt_expiry_mark */
dput(path.dentry);
mntput_no_expire(mnt);
+
+ if (!retval && (flags & UMOUNT_WAIT)) {
+ /*
+ * If the last delayed_fput() is called during do_umount()
+ * and makes mnt_count zero, we need to guarantee to register
+ * delayed_mntput by waiting for delayed_fput work again.
+ */
+ flush_delayed_fput_wait();
+
+ /* flush delayed_mntput_work to put sb->s_active */
+ flush_delayed_mntput_wait();
+ }
out:
return retval;
}
diff --git a/include/linux/file.h b/include/linux/file.h
index 61eb82cbafba..ffb4236cde39 100644
--- a/include/linux/file.h
+++ b/include/linux/file.h
@@ -84,6 +84,7 @@ extern void put_unused_fd(unsigned int fd);
extern void fd_install(unsigned int fd, struct file *file);

extern void flush_delayed_fput(void);
+extern void flush_delayed_fput_wait(void);
extern void __fput_sync(struct file *);

#endif /* __LINUX_FILE_H */
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 6e1fd5d21248..69f0fd53c9c7 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -1278,6 +1278,7 @@ struct mm_struct;
#define MNT_DETACH 0x00000002 /* Just detach from the tree */
#define MNT_EXPIRE 0x00000004 /* Mark for expiry */
#define UMOUNT_NOFOLLOW 0x00000008 /* Don't follow symlink on umount */
+#define UMOUNT_WAIT 0x00000010 /* Wait to unmount completely */
#define UMOUNT_UNUSED 0x80000000 /* Flag guaranteed to be unused */

/* sb->s_iflags */
--
2.14.0.rc1.383.gd1ce394fe2-goog