From: "Joe Jin" <joe.jin@oracle.com>
To: Jan Beulich <JBeulich@novell.com>
Cc: deepak.patel@oracle.com, Keir Fraser <keir.fraser@citrix.com>,
xen-devel@lists.xensource.com, Joe Jin <joe.jin@oracle.com>,
greg.marsden@oracle.com
Subject: Re: [PATCH] Fix blkback/blktap sysfs read bug.
Date: Wed, 20 Jan 2010 18:51:36 +0800 [thread overview]
Message-ID: <20100120105136.GA6801@joejin-pc.cn.oracle.com> (raw)
In-Reply-To: <4B56C2DB020000780002AE45@vpn.id2.novell.com>
On 2010-01-20 07:46, Jan Beulich wrote:
> >>> "Joe Jin" <joe.jin@oracle.com> 20.01.10 03:06 >>>
> >sysfs did not provide lock to handle this, not sure if developer
> >think it is not necessary or they'd like to caller to handled it.
>
> A lock is probably not the usual way to deal with this; ref-counting
> would seem more common. Nevertheless I think adding a lock will
> take care of the issue here.
Add refcnt could not fix the race for file open is sysfs file operations
and will not touch xenbus/blk{back,tap} struct, so could not increase
the refcnt.
The root cause is after open sysfs file, vbd was been removed by other.
For the lock type is read/write lock, so the lock almost will not slowdwon
performance, isn't it?
>
> >--- a/drivers/xen/blktap/xenbus.c Fri Jan 08 13:07:17 2010 +0000
> >+++ b/drivers/xen/blktap/xenbus.c Wed Jan 20 10:00:53 2010 +0800
> >...
> >@@ -122,10 +123,15 @@
> > struct device_attribute *attr, \
> > char *buf) \
> > { \
> >+ ssize_t ret = -ENODEV; \
> > struct xenbus_device *dev = to_xenbus_device(_dev); \
>
> The use of to_xenbus_device() here makes ...
>
> >- struct backend_info *be = dev->dev.driver_data; \
> >+ struct backend_info *be; \
> > \
> >- return sprintf(buf, format, ##args); \
> >+ read_lock(&sysfs_read_lock); \
> >+ if (dev && (be = dev->dev.driver_data) && be->blkif) \
>
> ... the checking of dev here useless (and the blkback part of the patch
> doesn't do the same).
Sorry I forgot get rid of it :)
>
> >+ ret = sprintf(buf, format, ##args); \
> >+ read_unlock(&sysfs_read_lock); \
> >+ return ret; \
> > } \
> > static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
> >
>
> And btw., in both cases with the lock added there's no need to check
> both 'be' and 'be->blkif', since be->blkif can't be NULL when be is
> non-NULL.
No we must check it, as I have methioned, root cause is vbd was been
freed after open sysfs file.
Below is the new patch, please review.
diff -r 0bec29c94ce9 drivers/xen/blkback/xenbus.c
--- a/drivers/xen/blkback/xenbus.c Mon Jan 18 14:50:43 2010 +0000
+++ b/drivers/xen/blkback/xenbus.c Wed Jan 20 18:49:50 2010 +0800
@@ -26,6 +26,8 @@
#define DPRINTK(fmt, args...) \
pr_debug("blkback/xenbus (%s:%d) " fmt ".\n", \
__FUNCTION__, __LINE__, ##args)
+
+static rwlock_t sysfs_read_lock = RW_LOCK_UNLOCKED;
struct backend_info
{
@@ -104,10 +106,19 @@
struct device_attribute *attr, \
char *buf) \
{ \
- struct xenbus_device *dev = to_xenbus_device(_dev); \
- struct backend_info *be = dev->dev.driver_data; \
+ ssize_t ret = -ENODEV; \
+ struct xenbus_device *dev; \
+ struct backend_info *be; \
\
- return sprintf(buf, format, ##args); \
+ if (!get_device(_dev)) \
+ return ret; \
+ dev = to_xenbus_device(_dev); \
+ read_lock(&sysfs_read_lock); \
+ if ((be = dev->dev.driver_data) && be->blkif) \
+ ret = sprintf(buf, format, ##args); \
+ read_unlock(&sysfs_read_lock); \
+ put_device(_dev); \
+ return ret; \
} \
static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
@@ -173,6 +184,7 @@
DPRINTK("");
+ write_lock(&sysfs_read_lock);
if (be->major || be->minor)
xenvbd_sysfs_delif(dev);
@@ -191,6 +203,7 @@
kfree(be);
dev->dev.driver_data = NULL;
+ write_unlock(&sysfs_read_lock);
return 0;
}
diff -r 0bec29c94ce9 drivers/xen/blktap/xenbus.c
--- a/drivers/xen/blktap/xenbus.c Mon Jan 18 14:50:43 2010 +0000
+++ b/drivers/xen/blktap/xenbus.c Wed Jan 20 18:49:50 2010 +0800
@@ -50,6 +50,7 @@
int group_added;
};
+static rwlock_t sysfs_read_lock = RW_LOCK_UNLOCKED;
static void connect(struct backend_info *);
static int connect_ring(struct backend_info *);
@@ -122,10 +123,19 @@
struct device_attribute *attr, \
char *buf) \
{ \
- struct xenbus_device *dev = to_xenbus_device(_dev); \
- struct backend_info *be = dev->dev.driver_data; \
+ ssize_t ret = -ENODEV; \
+ struct xenbus_device *dev; \
+ struct backend_info *be; \
\
- return sprintf(buf, format, ##args); \
+ if (!get_device(_dev)) \
+ return ret; \
+ dev = to_xenbus_device(_dev); \
+ read_lock(&sysfs_read_lock); \
+ if ((be = dev->dev.driver_data) && be->blkif) \
+ ret = sprintf(buf, format, ##args); \
+ read_unlock(&sysfs_read_lock); \
+ put_device(_dev); \
+ return ret; \
} \
static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
@@ -170,6 +180,7 @@
{
struct backend_info *be = dev->dev.driver_data;
+ write_lock(&sysfs_read_lock);
if (be->group_added)
xentap_sysfs_delif(be->dev);
if (be->backend_watch.node) {
@@ -187,6 +198,7 @@
}
kfree(be);
dev->dev.driver_data = NULL;
+ write_unlock(&sysfs_read_lock);
return 0;
}
next prev parent reply other threads:[~2010-01-20 10:51 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-01-19 9:52 [PATCH] Fix blkback/blktap sysfs read bug Joe Jin
2010-01-19 10:25 ` Jan Beulich
2010-01-19 11:32 ` Joe Jin
2010-01-19 12:06 ` Jan Beulich
2010-01-19 14:13 ` Joe Jin
2010-01-19 16:20 ` Jan Beulich
2010-01-20 2:06 ` Joe Jin
2010-01-20 7:46 ` Jan Beulich
2010-01-20 10:51 ` Joe Jin [this message]
2010-01-20 11:06 ` Jan Beulich
2010-01-20 11:45 ` Joe Jin
2010-01-20 20:25 ` Keir Fraser
2010-01-21 2:16 ` Daniel Stodden
2010-01-21 3:13 ` Joe Jin
2010-01-21 7:26 ` Daniel Stodden
2010-01-21 7:49 ` Joe Jin
2010-01-21 18:01 ` Daniel Stodden
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=20100120105136.GA6801@joejin-pc.cn.oracle.com \
--to=joe.jin@oracle.com \
--cc=JBeulich@novell.com \
--cc=deepak.patel@oracle.com \
--cc=greg.marsden@oracle.com \
--cc=keir.fraser@citrix.com \
--cc=xen-devel@lists.xensource.com \
/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
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).