Hi Pat,
sysfs_remove_dir() does not remove the contents of subdirs corresponding
to the attribute groups of a kobject. The following patch fixes this by first
removing the subdir contents and then removing thus emptied subdirs along
with the other attribute files of the kobject and plugs the memory
leakage resulting from orphan dentries.
I tested it by inserting and removing "dummy.o" network module and verifying
that dentires corresponding to "statistics" attribute group are removed.
Please comment.
Thanks
Maneesh
o sysfs_remove_dir() has to remove the files in the subdirs (corresponding
to the attribute groups) and then remove such empty subdirs along with the
other attribute files for the given kobject. The following patch does this
assuming that there are/will be no attribute sub-groups.
fs/sysfs/dir.c | 22 +++++++++++++++++-----
1 files changed, 17 insertions(+), 5 deletions(-)
diff -puN fs/sysfs/dir.c~sysfs_remove_dir-fix fs/sysfs/dir.c
--- linux-2.6.0-test5-mm2/fs/sysfs/dir.c~sysfs_remove_dir-fix 2003-09-15 15:08:45.000000000 +0530
+++ linux-2.6.0-test5-mm2-maneesh/fs/sysfs/dir.c 2003-09-15 15:26:04.000000000 +0530
@@ -119,23 +119,30 @@ void sysfs_remove_dir(struct kobject * k
{
struct list_head * node;
struct dentry * dentry = dget(kobj->dentry);
+ struct dentry * parent;
if (!dentry)
return;
pr_debug("sysfs %s: removing dir\n",dentry->d_name.name);
- down(&dentry->d_inode->i_sem);
+ parent = dentry;
+ down(&parent->d_inode->i_sem);
spin_lock(&dcache_lock);
- node = dentry->d_subdirs.next;
- while (node != &dentry->d_subdirs) {
- struct dentry * d = list_entry(node,struct dentry,d_child);
+repeat:
+ node = parent->d_subdirs.next;
+ while (node != &parent->d_subdirs) {
+ struct dentry * d = list_entry(node, struct dentry, d_child);
list_del_init(node);
pr_debug(" o %s (%d): ",d->d_name.name,atomic_read(&d->d_count));
if (d->d_inode) {
d = dget_locked(d);
pr_debug("removing");
+ if (!list_empty(&d->d_subdirs)) {
+ parent = d;
+ goto repeat;
+ }
/**
* Unlink and unhash.
@@ -147,7 +154,12 @@ void sysfs_remove_dir(struct kobject * k
spin_lock(&dcache_lock);
}
pr_debug(" done\n");
- node = dentry->d_subdirs.next;
+ node = parent->d_subdirs.next;
+ }
+
+ if (!list_empty(&dentry->d_subdirs)) {
+ parent = dentry;
+ goto repeat;
}
list_del_init(&dentry->d_child);
spin_unlock(&dcache_lock);
_
--
Maneesh Soni
Linux Technology Center,
IBM Software Lab, Bangalore, India
email: [email protected]
Phone: 91-80-5044999 Fax: 91-80-5268553
T/L : 9243696
Maneesh, Pat,
Just wanted to let you know I ran this patch
on 2.6.0-test5 with my input class patch and there
were no problems. Still working on my patch though.
Hanna
--On Monday, September 15, 2003 03:51:28 PM +0530 Maneesh Soni <[email protected]> wrote:
>
> Hi Pat,
>
> sysfs_remove_dir() does not remove the contents of subdirs corresponding
> to the attribute groups of a kobject. The following patch fixes this by first
> removing the subdir contents and then removing thus emptied subdirs along
> with the other attribute files of the kobject and plugs the memory
> leakage resulting from orphan dentries.
>
> I tested it by inserting and removing "dummy.o" network module and verifying
> that dentires corresponding to "statistics" attribute group are removed.
>
> Please comment.
>
> Thanks
> Maneesh
>
>
>
> o sysfs_remove_dir() has to remove the files in the subdirs (corresponding
> to the attribute groups) and then remove such empty subdirs along with the
> other attribute files for the given kobject. The following patch does this
> assuming that there are/will be no attribute sub-groups.
>
>
> fs/sysfs/dir.c | 22 +++++++++++++++++-----
> 1 files changed, 17 insertions(+), 5 deletions(-)
>
> diff -puN fs/sysfs/dir.c~sysfs_remove_dir-fix fs/sysfs/dir.c
> --- linux-2.6.0-test5-mm2/fs/sysfs/dir.c~sysfs_remove_dir-fix 2003-09-15 15:08:45.000000000 +0530
> +++ linux-2.6.0-test5-mm2-maneesh/fs/sysfs/dir.c 2003-09-15 15:26:04.000000000 +0530
> @@ -119,23 +119,30 @@ void sysfs_remove_dir(struct kobject * k
> {
> struct list_head * node;
> struct dentry * dentry = dget(kobj->dentry);
> + struct dentry * parent;
>
> if (!dentry)
> return;
>
> pr_debug("sysfs %s: removing dir\n",dentry->d_name.name);
> - down(&dentry->d_inode->i_sem);
>
> + parent = dentry;
> + down(&parent->d_inode->i_sem);
> spin_lock(&dcache_lock);
> - node = dentry->d_subdirs.next;
> - while (node != &dentry->d_subdirs) {
> - struct dentry * d = list_entry(node,struct dentry,d_child);
> +repeat:
> + node = parent->d_subdirs.next;
> + while (node != &parent->d_subdirs) {
> + struct dentry * d = list_entry(node, struct dentry, d_child);
> list_del_init(node);
>
> pr_debug(" o %s (%d): ",d->d_name.name,atomic_read(&d->d_count));
> if (d->d_inode) {
> d = dget_locked(d);
> pr_debug("removing");
> + if (!list_empty(&d->d_subdirs)) {
> + parent = d;
> + goto repeat;
> + }
>
> /**
> * Unlink and unhash.
> @@ -147,7 +154,12 @@ void sysfs_remove_dir(struct kobject * k
> spin_lock(&dcache_lock);
> }
> pr_debug(" done\n");
> - node = dentry->d_subdirs.next;
> + node = parent->d_subdirs.next;
> + }
> +
> + if (!list_empty(&dentry->d_subdirs)) {
> + parent = dentry;
> + goto repeat;
> }
> list_del_init(&dentry->d_child);
> spin_unlock(&dcache_lock);
>
> _
>
> --
> Maneesh Soni
> Linux Technology Center,
> IBM Software Lab, Bangalore, India
> email: [email protected]
> Phone: 91-80-5044999 Fax: 91-80-5268553
> T/L : 9243696
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
>
> sysfs_remove_dir() does not remove the contents of subdirs corresponding
> to the attribute groups of a kobject. The following patch fixes this by first
> removing the subdir contents and then removing thus emptied subdirs along
> with the other attribute files of the kobject and plugs the memory
> leakage resulting from orphan dentries.
>
> I tested it by inserting and removing "dummy.o" network module and verifying
> that dentires corresponding to "statistics" attribute group are removed.
Sorry it's taken a while to reply, but I even now haven't had a chance to
look that deep into this problem. I will say that this is the wrong
approach to the problem.
Upon initial inspection, it looks like it will do the right thing in all
cases, except when a parent is removed before a child is (but we don't
technically support that now).
However, I'd like to move away from the automatic cleanup that sysfs was
doing. attribute groups make it easier to clean up all the files that are
created for an object when the object is removed. I would like to see that
removal call inserted where necessary, instead of adding this complication
to the sysfs core.
I intend to remove the automatic cleanup in sysfs_remove_dir(), but
haven't gotten around to it..
Thanks,
Pat
On Fri, Sep 19, 2003 at 04:02:15PM -0700, Patrick Mochel wrote:
>
> > sysfs_remove_dir() does not remove the contents of subdirs corresponding
> > to the attribute groups of a kobject. The following patch fixes this by first
> > removing the subdir contents and then removing thus emptied subdirs along
> > with the other attribute files of the kobject and plugs the memory
> > leakage resulting from orphan dentries.
> >
> > I tested it by inserting and removing "dummy.o" network module and verifying
> > that dentires corresponding to "statistics" attribute group are removed.
>
> Sorry it's taken a while to reply, but I even now haven't had a chance to
> look that deep into this problem. I will say that this is the wrong
> approach to the problem.
>
> Upon initial inspection, it looks like it will do the right thing in all
> cases, except when a parent is removed before a child is (but we don't
> technically support that now).
>
> However, I'd like to move away from the automatic cleanup that sysfs was
> doing. attribute groups make it easier to clean up all the files that are
> created for an object when the object is removed. I would like to see that
> removal call inserted where necessary, instead of adding this complication
> to the sysfs core.
>
> I intend to remove the automatic cleanup in sysfs_remove_dir(), but
> haven't gotten around to it..
>
ok.. my intention was not to change callers for removing attribute group and
keep the changes within sysfs. But if you prefer otherway (I too :-)), please
see the patch below. As of now I can see only netdev not releasing attribute
group. I guess power subsystem don't need to remove attribute groups created
in pm_init() and pm_disk_init() as these don't unregister from sysfs.
Thanks
Maneesh
o attribute group should be removed while unregistering from sysfs. netdev
does not do this and leaks dentries belonging to the attribute group.
The following patch removes the attribute group when netdev is unregistered.
net/core/dev.c | 3 ++-
net/core/net-sysfs.c | 25 +++++++++++++++++++++++--
2 files changed, 25 insertions(+), 3 deletions(-)
diff -puN net/core/net-sysfs.c~sysfs-dentry-leak-fix net/core/net-sysfs.c
--- linux-2.6.0-test5-mm3/net/core/net-sysfs.c~sysfs-dentry-leak-fix 2003-09-22 09:45:04.000000000 +0530
+++ linux-2.6.0-test5-mm3-maneesh/net/core/net-sysfs.c 2003-09-22 09:45:17.000000000 +0530
@@ -383,6 +383,21 @@ static struct class net_class = {
#endif
};
+void netdev_unregister_sysfs(struct net_device * net)
+{
+ struct class_device * class_dev = &(net->class_dev);
+
+ if (net->get_stats)
+ sysfs_remove_group(&class_dev->kobj, &netstat_group);
+
+#ifdef WIRELESS_EXT
+ if (net->get_wireless_stats)
+ sysfs_remove_group(&class_dev->kobj, &wireless_group);
+#endif
+ class_device_del(class_dev);
+
+}
+
/* Create sysfs entries for network device. */
int netdev_register_sysfs(struct net_device *net)
{
@@ -411,9 +426,15 @@ int netdev_register_sysfs(struct net_dev
#ifdef WIRELESS_EXT
if (net->get_wireless_stats &&
(ret = sysfs_create_group(&class_dev->kobj, &wireless_group)))
- goto out_unreg;
-#endif
+ goto out_cleanup;
+
+ return 0;
+out_cleanup:
+ if (net->get_stats)
+ sysfs_remove_group(&class_dev->kobj, &netstat_group);
+#else
return 0;
+#endif
out_unreg:
printk(KERN_WARNING "%s: sysfs attribute registration failed %d\n",
diff -puN net/core/dev.c~sysfs-dentry-leak-fix net/core/dev.c
--- linux-2.6.0-test5-mm3/net/core/dev.c~sysfs-dentry-leak-fix 2003-09-22 09:45:08.000000000 +0530
+++ linux-2.6.0-test5-mm3-maneesh/net/core/dev.c 2003-09-22 09:45:17.000000000 +0530
@@ -183,6 +183,7 @@ int netdev_fastroute_obstacles;
extern int netdev_sysfs_init(void);
extern int netdev_register_sysfs(struct net_device *);
+extern int netdev_unregister_sysfs(struct net_device *);
#ifdef CONFIG_KGDB
extern int kgdb_net_interrupt(struct sk_buff *skb);
#endif
@@ -2834,7 +2835,7 @@ void netdev_run_todo(void)
break;
case NETREG_UNREGISTERING:
- class_device_del(&dev->class_dev);
+ netdev_unregister_sysfs(dev);
dev->reg_state = NETREG_UNREGISTERED;
netdev_wait_allrefs(dev);
_
Thanks Stephen for letting me know.
Maneesh
Rediffed for 2.6.0-test5.
o attribute group should be removed while unregistering from sysfs. netdev
does not do this and leaks dentries belonging to the attribute group.
The following patch removes the attribute group when netdev is unregistered.
net/core/dev.c | 3 ++-
net/core/net-sysfs.c | 25 +++++++++++++++++++++++--
2 files changed, 25 insertions(+), 3 deletions(-)
diff -puN net/core/net-sysfs.c~sysfs-dentry-leak-fix net/core/net-sysfs.c
--- linux-2.6.0-test5/net/core/net-sysfs.c~sysfs-dentry-leak-fix 2003-09-23 09:32:01.000000000 +0530
+++ linux-2.6.0-test5-maneesh/net/core/net-sysfs.c 2003-09-23 09:32:23.000000000 +0530
@@ -383,6 +383,21 @@ static struct class net_class = {
#endif
};
+void netdev_unregister_sysfs(struct net_device * net)
+{
+ struct class_device * class_dev = &(net->class_dev);
+
+ if (net->get_stats)
+ sysfs_remove_group(&class_dev->kobj, &netstat_group);
+
+#ifdef WIRELESS_EXT
+ if (net->get_wireless_stats)
+ sysfs_remove_group(&class_dev->kobj, &wireless_group);
+#endif
+ class_device_del(class_dev);
+
+}
+
/* Create sysfs entries for network device. */
int netdev_register_sysfs(struct net_device *net)
{
@@ -411,9 +426,15 @@ int netdev_register_sysfs(struct net_dev
#ifdef WIRELESS_EXT
if (net->get_wireless_stats &&
(ret = sysfs_create_group(&class_dev->kobj, &wireless_group)))
- goto out_unreg;
-#endif
+ goto out_cleanup;
+
+ return 0;
+out_cleanup:
+ if (net->get_stats)
+ sysfs_remove_group(&class_dev->kobj, &netstat_group);
+#else
return 0;
+#endif
out_unreg:
printk(KERN_WARNING "%s: sysfs attribute registration failed %d\n",
diff -puN net/core/dev.c~sysfs-dentry-leak-fix net/core/dev.c
--- linux-2.6.0-test5/net/core/dev.c~sysfs-dentry-leak-fix 2003-09-23 09:31:21.000000000 +0530
+++ linux-2.6.0-test5-maneesh/net/core/dev.c 2003-09-23 09:33:04.000000000 +0530
@@ -183,6 +183,7 @@ int netdev_fastroute_obstacles;
extern int netdev_sysfs_init(void);
extern int netdev_register_sysfs(struct net_device *);
+extern int netdev_unregister_sysfs(struct net_device *);
/*******************************************************************************
@@ -2819,7 +2820,7 @@ void netdev_run_todo(void)
break;
case NETREG_UNREGISTERING:
- class_device_del(&dev->class_dev);
+ netdev_unregister_sysfs(dev);
dev->reg_state = NETREG_UNREGISTERED;
netdev_wait_allrefs(dev);
_
--
Maneesh Soni
Linux Technology Center,
IBM Software Lab, Bangalore, India
email: [email protected]
Phone: 91-80-5044999 Fax: 91-80-5268553
T/L : 9243696