ALT Linux Team development discussions
 help / color / mirror / Atom feed
From: "Kirill A. Shutemov" <kirill@shutemov.name>
To: ALT Devel discussion list <devel@lists.altlinux.org>
Subject: Re: [devel] [partially solved] Q: exit_group slowness
Date: Fri, 8 Jun 2012 15:46:29 +0300
Message-ID: <20120608124629.GA5947@shutemov.name> (raw)
In-Reply-To: <20120606013127.GA29780@altlinux.org>

On Wed, Jun 06, 2012 at 05:31:27AM +0400, Dmitry V. Levin wrote:
> On Tue, Jun 05, 2012 at 04:10:35PM +0300, Kirill A. Shutemov wrote:
> > On Tue, Jun 05, 2012 at 04:46:03PM +0400, Dmitry V. Levin wrote:
> > > On Tue, Jun 05, 2012 at 03:37:19PM +0300, Kirill A. Shutemov wrote:
> > > > On Tue, Jun 05, 2012 at 01:08:28PM +0400, Dmitry V. Levin wrote:
> > > > > On Tue, Jun 05, 2012 at 09:06:22AM +0300, Kirill A. Shutemov wrote:
> > > > > > Похоже вот на это:
> > > > > > 
> > > > > > http://lkml.org/lkml/2012/6/3/34
> > > > > > 
> > > > > > Ниже по трэду Пол предлагает возможное решение.
> > > > > 
> > > > > Спасибо, откат на 3.2.14-std-def-alt1 радикально улучшил картинку:
> > > > > # env -i strace -r -eprocess /usr/bin/unshare -i /bin/true
> > > > >      0.000000 execve("/usr/bin/unshare", ["/usr/bin/unshare", "-i", "/bin/true"], [/* 0 vars */]) = 0
> > > > >      0.001357 arch_prctl(ARCH_SET_FS, 0x7f49add73700) = 0
> > > > >      0.000457 unshare(CLONE_NEWIPC)     = 0
> > > > >      0.000200 execve("/bin/true", ["/bin/true"], [/* 0 vars */]) = 0
> > > > >      0.000350 exit_group(0)             = ?
> > > > >      0.156461 +++ exited with 0 +++
> > > > > 
> > > > > Хотя до идеала еще далеко:
> > > > > # env -i strace -r /bin/true
> > > > >      0.000000 execve("/bin/true", ["/bin/true"], [/* 0 vars */]) = 0
> > > > >      0.000628 exit_group(0)             = ?
> > > > >      0.000329 +++ exited with 0 +++
> > > > 
> > > > exit_group() дорогой только для последнего процесса в namespace, так?
> > > 
> > > Да, конечно.
> > > 
> > > > Думаю, это цена которую можно заплатить.
> > > 
> > > 0.15 секунды это очень долго, то же самое ядро на более простых серверах
> > > завершает последний процесс в namespace на 1-2 порядка быстрее.  Так что
> > > вопрос, что не так и как с этим бороться, пока остается.
> > 
> > RCU_FAST_NO_HZ не предназначена для больших машин. О чём в явном виде
> > написано в описании опции. Так что CONFIG_RCU_FAST_NO_HZ=y было ошибкой
> > для дистрибутивного ядра.
> > 
> > Когда моя большая машинка (40 ядер + ht) освободится я попробую
> > поэксперементировать с этим немного. Есть подозрение, что отмонирование
> > файловой системы (внутриядерной, в случае CLONE_NEWIPC) плохо
> > масштабируется на большое количество процессоров даже без RCU_FAST_NO_HZ.
> 
> Для справки: на той же машинке (где "rpm --eval %__nprocs" показывает 32) с
> 3.3.8-std-def-alt1 (с отключенным CONFIG_RCU_FAST_NO_HZ) время завершения
> "unshare -i /bin/true" уменьшилось примерно в полтора раза по сравнению с
> 3.2.14-std-def-alt1 (с 0.15 до 0.10 секунды).

Дима, можешь проверить поможет ли патчик, прежде чем я покажу его
апстриму:

>From b76600918d021ec9836659dbb991f4d9a92de795 Mon Sep 17 00:00:00 2001
From: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Date: Fri, 8 Jun 2012 15:30:31 +0300
Subject: [PATCH] fs: push rcu_barrier() from deactivate_locked_super() to
 filesystems

There's no reason to call rcu_barrier() on every deactivate_locked_super().
We only need to make sure that all delayed rcu free inodes are flushed
before we destroy related cache.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
---
 fs/9p/v9fs.c             |    5 +++++
 fs/adfs/super.c          |    5 +++++
 fs/affs/super.c          |    5 +++++
 fs/afs/super.c           |    5 +++++
 fs/befs/linuxvfs.c       |    5 +++++
 fs/bfs/inode.c           |    5 +++++
 fs/btrfs/extent_io.c     |    6 ++++++
 fs/btrfs/inode.c         |    5 +++++
 fs/ceph/super.c          |    5 +++++
 fs/cifs/cifsfs.c         |    5 +++++
 fs/coda/inode.c          |    5 +++++
 fs/ecryptfs/main.c       |    6 ++++++
 fs/efs/super.c           |    5 +++++
 fs/exofs/super.c         |    5 +++++
 fs/ext2/super.c          |    5 +++++
 fs/ext3/super.c          |    5 +++++
 fs/ext4/super.c          |    5 +++++
 fs/fat/inode.c           |    5 +++++
 fs/freevxfs/vxfs_super.c |    5 +++++
 fs/fuse/inode.c          |    6 ++++++
 fs/hfs/super.c           |    6 ++++++
 fs/hfsplus/super.c       |    6 ++++++
 fs/hpfs/super.c          |    5 +++++
 fs/hugetlbfs/inode.c     |    5 +++++
 fs/isofs/inode.c         |    5 +++++
 fs/jffs2/super.c         |    6 ++++++
 fs/jfs/super.c           |    6 ++++++
 fs/logfs/inode.c         |    5 +++++
 fs/minix/inode.c         |    5 +++++
 fs/ncpfs/inode.c         |    5 +++++
 fs/nfs/inode.c           |    5 +++++
 fs/nilfs2/super.c        |    6 ++++++
 fs/ntfs/super.c          |    6 ++++++
 fs/ocfs2/dlmfs/dlmfs.c   |    5 +++++
 fs/ocfs2/super.c         |    5 +++++
 fs/openpromfs/inode.c    |    5 +++++
 fs/qnx4/inode.c          |    5 +++++
 fs/qnx6/inode.c          |    5 +++++
 fs/reiserfs/super.c      |    5 +++++
 fs/romfs/super.c         |    5 +++++
 fs/squashfs/super.c      |    5 +++++
 fs/super.c               |    6 ------
 fs/sysv/inode.c          |    5 +++++
 fs/ubifs/super.c         |    6 ++++++
 fs/udf/super.c           |    5 +++++
 fs/ufs/super.c           |    5 +++++
 fs/xfs/xfs_super.c       |    5 +++++
 47 files changed, 240 insertions(+), 6 deletions(-)

diff --git a/fs/9p/v9fs.c b/fs/9p/v9fs.c
index b85efa7..392c5da 100644
--- a/fs/9p/v9fs.c
+++ b/fs/9p/v9fs.c
@@ -560,6 +560,11 @@ static int v9fs_init_inode_cache(void)
  */
 static void v9fs_destroy_inode_cache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(v9fs_inode_cache);
 }
 
diff --git a/fs/adfs/super.c b/fs/adfs/super.c
index 06fdcc9..f137165 100644
--- a/fs/adfs/super.c
+++ b/fs/adfs/super.c
@@ -276,6 +276,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(adfs_inode_cachep);
 }
 
diff --git a/fs/affs/super.c b/fs/affs/super.c
index 0782653..4fe18a8 100644
--- a/fs/affs/super.c
+++ b/fs/affs/super.c
@@ -129,6 +129,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(affs_inode_cachep);
 }
 
diff --git a/fs/afs/super.c b/fs/afs/super.c
index f02b31e..af69050 100644
--- a/fs/afs/super.c
+++ b/fs/afs/super.c
@@ -123,6 +123,11 @@ void __exit afs_fs_exit(void)
 		BUG();
 	}
 
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(afs_inode_cachep);
 	_leave("");
 }
diff --git a/fs/befs/linuxvfs.c b/fs/befs/linuxvfs.c
index e18da23..02f3331 100644
--- a/fs/befs/linuxvfs.c
+++ b/fs/befs/linuxvfs.c
@@ -454,6 +454,11 @@ befs_init_inodecache(void)
 static void
 befs_destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(befs_inode_cachep);
 }
 
diff --git a/fs/bfs/inode.c b/fs/bfs/inode.c
index 9870417..d5fc598 100644
--- a/fs/bfs/inode.c
+++ b/fs/bfs/inode.c
@@ -280,6 +280,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(bfs_inode_cachep);
 }
 
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 2c8f7b2..010623a 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -106,6 +106,12 @@ void extent_io_exit(void)
 		list_del(&eb->leak_list);
 		kmem_cache_free(extent_buffer_cache, eb);
 	}
+
+	/*
+	 * Make sure all delayed rcu free are flushed before we
+	 * destroy caches.
+	 */
+	rcu_barrier();
 	if (extent_state_cache)
 		kmem_cache_destroy(extent_state_cache);
 	if (extent_buffer_cache)
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index f6ab6f5..971d222 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -6976,6 +6976,11 @@ static void init_once(void *foo)
 
 void btrfs_destroy_cachep(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	if (btrfs_inode_cachep)
 		kmem_cache_destroy(btrfs_inode_cachep);
 	if (btrfs_trans_handle_cachep)
diff --git a/fs/ceph/super.c b/fs/ceph/super.c
index 1e67dd7..ecc251e 100644
--- a/fs/ceph/super.c
+++ b/fs/ceph/super.c
@@ -602,6 +602,11 @@ bad_cap:
 
 static void destroy_caches(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(ceph_inode_cachep);
 	kmem_cache_destroy(ceph_cap_cachep);
 	kmem_cache_destroy(ceph_dentry_cachep);
diff --git a/fs/cifs/cifsfs.c b/fs/cifs/cifsfs.c
index 8b6e344..93ce9b3f 100644
--- a/fs/cifs/cifsfs.c
+++ b/fs/cifs/cifsfs.c
@@ -975,6 +975,11 @@ cifs_init_inodecache(void)
 static void
 cifs_destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(cifs_inode_cachep);
 }
 
diff --git a/fs/coda/inode.c b/fs/coda/inode.c
index f181312..0c70460 100644
--- a/fs/coda/inode.c
+++ b/fs/coda/inode.c
@@ -85,6 +85,11 @@ int coda_init_inodecache(void)
 
 void coda_destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(coda_inode_cachep);
 }
 
diff --git a/fs/ecryptfs/main.c b/fs/ecryptfs/main.c
index 6895493..f6acf8a 100644
--- a/fs/ecryptfs/main.c
+++ b/fs/ecryptfs/main.c
@@ -693,6 +693,12 @@ static void ecryptfs_free_kmem_caches(void)
 {
 	int i;
 
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
+
 	for (i = 0; i < ARRAY_SIZE(ecryptfs_cache_infos); i++) {
 		struct ecryptfs_cache_info *info;
 
diff --git a/fs/efs/super.c b/fs/efs/super.c
index e755ec7..2002431 100644
--- a/fs/efs/super.c
+++ b/fs/efs/super.c
@@ -96,6 +96,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(efs_inode_cachep);
 }
 
diff --git a/fs/exofs/super.c b/fs/exofs/super.c
index 4337836..d271f96 100644
--- a/fs/exofs/super.c
+++ b/fs/exofs/super.c
@@ -206,6 +206,11 @@ static int init_inodecache(void)
  */
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(exofs_inode_cachep);
 }
 
diff --git a/fs/ext2/super.c b/fs/ext2/super.c
index b3621cb..de2051c 100644
--- a/fs/ext2/super.c
+++ b/fs/ext2/super.c
@@ -204,6 +204,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(ext2_inode_cachep);
 }
 
diff --git a/fs/ext3/super.c b/fs/ext3/super.c
index 8c3a44b..bd60e08 100644
--- a/fs/ext3/super.c
+++ b/fs/ext3/super.c
@@ -543,6 +543,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(ext3_inode_cachep);
 }
 
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index eb7aa3e..4e2aacb 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -1046,6 +1046,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(ext4_inode_cachep);
 }
 
diff --git a/fs/fat/inode.c b/fs/fat/inode.c
index a3d81eb..35113bc 100644
--- a/fs/fat/inode.c
+++ b/fs/fat/inode.c
@@ -526,6 +526,11 @@ static int __init fat_init_inodecache(void)
 
 static void __exit fat_destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(fat_inode_cachep);
 }
 
diff --git a/fs/freevxfs/vxfs_super.c b/fs/freevxfs/vxfs_super.c
index d4fabd2..fed2c8a 100644
--- a/fs/freevxfs/vxfs_super.c
+++ b/fs/freevxfs/vxfs_super.c
@@ -279,6 +279,11 @@ static void __exit
 vxfs_cleanup(void)
 {
 	unregister_filesystem(&vxfs_fs_type);
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(vxfs_inode_cachep);
 }
 
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index 1cd6165..9078173 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -1171,6 +1171,12 @@ static void fuse_fs_cleanup(void)
 {
 	unregister_filesystem(&fuse_fs_type);
 	unregister_fuseblk();
+
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(fuse_inode_cachep);
 }
 
diff --git a/fs/hfs/super.c b/fs/hfs/super.c
index 7b4c537..373a9e5 100644
--- a/fs/hfs/super.c
+++ b/fs/hfs/super.c
@@ -483,6 +483,12 @@ static int __init init_hfs_fs(void)
 static void __exit exit_hfs_fs(void)
 {
 	unregister_filesystem(&hfs_fs_type);
+
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(hfs_inode_cachep);
 }
 
diff --git a/fs/hfsplus/super.c b/fs/hfsplus/super.c
index a9bca4b..bde9ecf 100644
--- a/fs/hfsplus/super.c
+++ b/fs/hfsplus/super.c
@@ -615,6 +615,12 @@ static int __init init_hfsplus_fs(void)
 static void __exit exit_hfsplus_fs(void)
 {
 	unregister_filesystem(&hfsplus_fs_type);
+
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(hfsplus_inode_cachep);
 }
 
diff --git a/fs/hpfs/super.c b/fs/hpfs/super.c
index 706a12c..3cb1da5 100644
--- a/fs/hpfs/super.c
+++ b/fs/hpfs/super.c
@@ -210,6 +210,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(hpfs_inode_cachep);
 }
 
diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index cc9281b..b482175 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -1042,6 +1042,11 @@ static int __init init_hugetlbfs_fs(void)
 
 static void __exit exit_hugetlbfs_fs(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(hugetlbfs_inode_cachep);
 	kern_unmount(hugetlbfs_vfsmount);
 	unregister_filesystem(&hugetlbfs_fs_type);
diff --git a/fs/isofs/inode.c b/fs/isofs/inode.c
index 29037c3..f94cde4 100644
--- a/fs/isofs/inode.c
+++ b/fs/isofs/inode.c
@@ -114,6 +114,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(isofs_inode_cachep);
 }
 
diff --git a/fs/jffs2/super.c b/fs/jffs2/super.c
index 61ea413..ff48795 100644
--- a/fs/jffs2/super.c
+++ b/fs/jffs2/super.c
@@ -418,6 +418,12 @@ static void __exit exit_jffs2_fs(void)
 	unregister_filesystem(&jffs2_fs_type);
 	jffs2_destroy_slab_caches();
 	jffs2_compressors_exit();
+
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(jffs2_inode_cachep);
 }
 
diff --git a/fs/jfs/super.c b/fs/jfs/super.c
index 4a82950..2fc7602 100644
--- a/fs/jfs/super.c
+++ b/fs/jfs/super.c
@@ -898,6 +898,12 @@ static void __exit exit_jfs_fs(void)
 	jfs_proc_clean();
 #endif
 	unregister_filesystem(&jfs_fs_type);
+
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(jfs_inode_cachep);
 }
 
diff --git a/fs/logfs/inode.c b/fs/logfs/inode.c
index a422f42..3d1b9fa 100644
--- a/fs/logfs/inode.c
+++ b/fs/logfs/inode.c
@@ -401,5 +401,10 @@ int logfs_init_inode_cache(void)
 
 void logfs_destroy_inode_cache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(logfs_inode_cache);
 }
diff --git a/fs/minix/inode.c b/fs/minix/inode.c
index 2a503ad..dc8d362 100644
--- a/fs/minix/inode.c
+++ b/fs/minix/inode.c
@@ -100,6 +100,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(minix_inode_cachep);
 }
 
diff --git a/fs/ncpfs/inode.c b/fs/ncpfs/inode.c
index 333df07..0c62c55 100644
--- a/fs/ncpfs/inode.c
+++ b/fs/ncpfs/inode.c
@@ -89,6 +89,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(ncp_inode_cachep);
 }
 
diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index e605d69..39aae2e 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -1566,6 +1566,11 @@ static int __init nfs_init_inodecache(void)
 
 static void nfs_destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(nfs_inode_cachep);
 }
 
diff --git a/fs/nilfs2/super.c b/fs/nilfs2/super.c
index 1099a76..956e5a4 100644
--- a/fs/nilfs2/super.c
+++ b/fs/nilfs2/super.c
@@ -1386,6 +1386,12 @@ static void nilfs_segbuf_init_once(void *obj)
 
 static void nilfs_destroy_cachep(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
+
 	if (nilfs_inode_cachep)
 		kmem_cache_destroy(nilfs_inode_cachep);
 	if (nilfs_transaction_cachep)
diff --git a/fs/ntfs/super.c b/fs/ntfs/super.c
index b341492..ecc8625 100644
--- a/fs/ntfs/super.c
+++ b/fs/ntfs/super.c
@@ -3185,6 +3185,12 @@ static void __exit exit_ntfs_fs(void)
 	ntfs_debug("Unregistering NTFS driver.");
 
 	unregister_filesystem(&ntfs_fs_type);
+
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(ntfs_big_inode_cache);
 	kmem_cache_destroy(ntfs_inode_cache);
 	kmem_cache_destroy(ntfs_name_cache);
diff --git a/fs/ocfs2/dlmfs/dlmfs.c b/fs/ocfs2/dlmfs/dlmfs.c
index e31d6ae..6934459 100644
--- a/fs/ocfs2/dlmfs/dlmfs.c
+++ b/fs/ocfs2/dlmfs/dlmfs.c
@@ -691,6 +691,11 @@ static void __exit exit_dlmfs_fs(void)
 	flush_workqueue(user_dlm_worker);
 	destroy_workqueue(user_dlm_worker);
 
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(dlmfs_inode_cache);
 
 	bdi_destroy(&dlmfs_backing_dev_info);
diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
index 68f4541..0e91ec2 100644
--- a/fs/ocfs2/super.c
+++ b/fs/ocfs2/super.c
@@ -1818,6 +1818,11 @@ static int ocfs2_initialize_mem_caches(void)
 
 static void ocfs2_free_mem_caches(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	if (ocfs2_inode_cachep)
 		kmem_cache_destroy(ocfs2_inode_cachep);
 	ocfs2_inode_cachep = NULL;
diff --git a/fs/openpromfs/inode.c b/fs/openpromfs/inode.c
index bc49c97..1b76370 100644
--- a/fs/openpromfs/inode.c
+++ b/fs/openpromfs/inode.c
@@ -463,6 +463,11 @@ static int __init init_openprom_fs(void)
 static void __exit exit_openprom_fs(void)
 {
 	unregister_filesystem(&openprom_fs_type);
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(op_inode_cachep);
 }
 
diff --git a/fs/qnx4/inode.c b/fs/qnx4/inode.c
index 552e994..9534b4f 100644
--- a/fs/qnx4/inode.c
+++ b/fs/qnx4/inode.c
@@ -391,6 +391,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(qnx4_inode_cachep);
 }
 
diff --git a/fs/qnx6/inode.c b/fs/qnx6/inode.c
index e44012d..3cba5d2 100644
--- a/fs/qnx6/inode.c
+++ b/fs/qnx6/inode.c
@@ -652,6 +652,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(qnx6_inode_cachep);
 }
 
diff --git a/fs/reiserfs/super.c b/fs/reiserfs/super.c
index 651ce76..ad39139 100644
--- a/fs/reiserfs/super.c
+++ b/fs/reiserfs/super.c
@@ -603,6 +603,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(reiserfs_inode_cachep);
 }
 
diff --git a/fs/romfs/super.c b/fs/romfs/super.c
index e64f6b5..f901eaf 100644
--- a/fs/romfs/super.c
+++ b/fs/romfs/super.c
@@ -648,6 +648,11 @@ error_register:
 static void __exit exit_romfs_fs(void)
 {
 	unregister_filesystem(&romfs_fs_type);
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(romfs_inode_cachep);
 }
 
diff --git a/fs/squashfs/super.c b/fs/squashfs/super.c
index 29cd014..260e392 100644
--- a/fs/squashfs/super.c
+++ b/fs/squashfs/super.c
@@ -425,6 +425,11 @@ static int __init init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(squashfs_inode_cachep);
 }
 
diff --git a/fs/super.c b/fs/super.c
index cf00177..2d962aa 100644
--- a/fs/super.c
+++ b/fs/super.c
@@ -256,12 +256,6 @@ void deactivate_locked_super(struct super_block *s)
 
 		/* caches are now gone, we can safely kill the shrinker now */
 		unregister_shrinker(&s->s_shrink);
-
-		/*
-		 * We need to call rcu_barrier so all the delayed rcu free
-		 * inodes are flushed before we release the fs module.
-		 */
-		rcu_barrier();
 		put_filesystem(fs);
 		put_super(s);
 	} else {
diff --git a/fs/sysv/inode.c b/fs/sysv/inode.c
index 08d0b25..a08341c 100644
--- a/fs/sysv/inode.c
+++ b/fs/sysv/inode.c
@@ -376,5 +376,10 @@ int __init sysv_init_icache(void)
 
 void sysv_destroy_icache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(sysv_inode_cachep);
 }
diff --git a/fs/ubifs/super.c b/fs/ubifs/super.c
index 5862dd9..e8ae748 100644
--- a/fs/ubifs/super.c
+++ b/fs/ubifs/super.c
@@ -2302,6 +2302,12 @@ static void __exit ubifs_exit(void)
 	dbg_debugfs_exit();
 	ubifs_compressors_exit();
 	unregister_shrinker(&ubifs_shrinker_info);
+
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(ubifs_inode_slab);
 	unregister_filesystem(&ubifs_fs_type);
 }
diff --git a/fs/udf/super.c b/fs/udf/super.c
index ac8a348..ebe9f52 100644
--- a/fs/udf/super.c
+++ b/fs/udf/super.c
@@ -170,6 +170,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(udf_inode_cachep);
 }
 
diff --git a/fs/ufs/super.c b/fs/ufs/super.c
index 302f340..f655c1d 100644
--- a/fs/ufs/super.c
+++ b/fs/ufs/super.c
@@ -1449,6 +1449,11 @@ static int init_inodecache(void)
 
 static void destroy_inodecache(void)
 {
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
 	kmem_cache_destroy(ufs_inode_cachep);
 }
 
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index 0d9de41..bbcbf2a 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -1582,6 +1582,11 @@ xfs_init_zones(void)
 STATIC void
 xfs_destroy_zones(void)
 {
+	/*
+	 * Make sure all delayed rcu free are flushed before we
+	 * destroy caches.
+	 */
+	rcu_barrier();
 	kmem_zone_destroy(xfs_ili_zone);
 	kmem_zone_destroy(xfs_inode_zone);
 	kmem_zone_destroy(xfs_efi_zone);
-- 
 Kirill A. Shutemov


  parent reply	other threads:[~2012-06-08 12:46 UTC|newest]

Thread overview: 20+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2012-06-04 21:11 [devel] " Dmitry V. Levin
2012-06-04 22:30 ` Kirill A. Shutemov
2012-06-05  0:20   ` Dmitry V. Levin
2012-06-05  6:06     ` Kirill A. Shutemov
2012-06-05  9:08       ` [devel] [partially solved] " Dmitry V. Levin
2012-06-05 12:37         ` Kirill A. Shutemov
2012-06-05 12:46           ` Dmitry V. Levin
2012-06-05 13:10             ` Kirill A. Shutemov
2012-06-06  1:31               ` Dmitry V. Levin
2012-06-06 18:40                 ` Kirill A. Shutemov
2012-06-08 12:46                 ` Kirill A. Shutemov [this message]
2012-06-08 15:31                   ` Kirill A. Shutemov
2012-06-08 15:56                   ` Dmitry V. Levin
2012-06-12  7:00                     ` Kirill A. Shutemov
2012-06-13  0:32                       ` Dmitry V. Levin
2012-08-10 19:41                         ` Michael Shigorin
2012-06-04 22:35 ` [devel] " Led
2012-06-04 22:39   ` Kirill A. Shutemov
2012-06-04 22:47     ` Led
2012-06-06  8:43 ` Michael Shigorin

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20120608124629.GA5947@shutemov.name \
    --to=kirill@shutemov.name \
    --cc=devel@lists.altlinux.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

ALT Linux Team development discussions

This inbox may be cloned and mirrored by anyone:

	git clone --mirror http://lore.altlinux.org/devel/0 devel/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 devel devel/ http://lore.altlinux.org/devel \
		devel@altlinux.org devel@altlinux.ru devel@lists.altlinux.org devel@lists.altlinux.ru devel@linux.iplabs.ru mandrake-russian@linuxteam.iplabs.ru sisyphus@linuxteam.iplabs.ru
	public-inbox-index devel

Example config snippet for mirrors.
Newsgroup available over NNTP:
	nntp://lore.altlinux.org/org.altlinux.lists.devel


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git