[PATCH 0/2 v2] Fix deadlock with suspend & quotas

[PATCH 0/2 v2] Fix deadlock with suspend & quotas

[Jan Kara]

  Hello Al,

  This is my attempt to fix deadlock in quota code when filesystem is
suspended and someone tries to modify quota e.g. via Q_SETQUOTA command. I
added a VFS function to get a superblock reference and return only after
the filesystem is thawed.

Changes since v1:
* wait for frozen fs really only when requested
* updated some comments & names

								Honza

[PATCH 1/2] vfs: Provide function to get superblock and wait for it to thaw

[Jan Kara]

In quota code we need to find a superblock corresponding to a device and wait
for superblock to be unfrozen. However this waiting has to happen without
s_umount semaphore because that is required for superblock to thaw. So provide
a function in VFS for this to keep dances with s_umount where they belong.

Signed-off-by: Jan Kara <jack@suse.cz>
---
 fs/super.c         |   50 ++++++++++++++++++++++++++++++++++++++------------
 include/linux/fs.h |    1 +
 2 files changed, 39 insertions(+), 12 deletions(-)

diff --git a/fs/super.c b/fs/super.c
index 6015c02..1f657d6 100644
--- a/fs/super.c
+++ b/fs/super.c
@@ -593,15 +593,8 @@ void iterate_supers_type(struct file_system_type *type,
 
 EXPORT_SYMBOL(iterate_supers_type);
 
-/**
- *	get_super - get the superblock of a device
- *	@bdev: device to get the superblock for
- *	
- *	Scans the superblock list and finds the superblock of the file system
- *	mounted on the device given. %NULL is returned if no match is found.
- */
-
-struct super_block *get_super(struct block_device *bdev)
+static struct super_block *__get_super(struct block_device *bdev,
+				       bool wait_thaw)
 {
 	struct super_block *sb;
 
@@ -618,9 +611,13 @@ rescan:
 			spin_unlock(&sb_lock);
 			down_read(&sb->s_umount);
 			/* still alive? */
-			if (sb->s_root && (sb->s_flags & MS_BORN))
-				return sb;
-			up_read(&sb->s_umount);
+			if (sb->s_root && (sb->s_flags & MS_BORN)) {
+				if (!wait_thaw || sb->s_frozen == SB_UNFROZEN)
+					return sb;
+				up_read(&sb->s_umount);
+				vfs_check_frozen(sb, SB_FREEZE_WRITE);
+			} else
+				up_read(&sb->s_umount);
 			/* nope, got unmounted */
 			spin_lock(&sb_lock);
 			__put_super(sb);
@@ -631,9 +628,38 @@ rescan:
 	return NULL;
 }
 
+/**
+ *	get_super - get the superblock of a device
+ *	@bdev: device to get the superblock for
+ *
+ *	Scans the superblock list and finds the superblock of the file system
+ *	mounted on the device given. %NULL is returned if no match is found.
+ *	Note that the returned superblock may be frozen.
+ */
+
+struct super_block *get_super(struct block_device *bdev)
+{
+	return __get_super(bdev, false);
+}
 EXPORT_SYMBOL(get_super);
 
 /**
+ *	get_super_thawed - get thawed superblock of a device
+ *	@bdev: device to get the superblock for
+ *
+ *	Scans the superblock list and finds the superblock of the file system
+ *	mounted on the device. The superblock is returned once it is thawed
+ *	(or immediately if it was not frozen). %NULL is returned if no match
+ *	is found.
+ */
+
+struct super_block *get_super_thawed(struct block_device *bdev)
+{
+	return __get_super(bdev, true);
+}
+EXPORT_SYMBOL(get_super_thawed);
+
+/**
  * get_active_super - get an active reference to the superblock of a device
  * @bdev: device to get the superblock for
  *
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 386da09..69cd5bb 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -2496,6 +2496,7 @@ extern void get_filesystem(struct file_system_type *fs);
 extern void put_filesystem(struct file_system_type *fs);
 extern struct file_system_type *get_fs_type(const char *name);
 extern struct super_block *get_super(struct block_device *);
+extern struct super_block *get_super_thawed(struct block_device *);
 extern struct super_block *get_active_super(struct block_device *bdev);
 extern void drop_super(struct super_block *sb);
 extern void iterate_supers(void (*)(struct super_block *, void *), void *);
-- 
1.7.1

[PATCH 2/2] quota: Fix deadlock with suspend and quotas

[Jan Kara]

This script causes a kernel deadlock:
set -e
DEVICE=/dev/vg1/linear
lvchange -ay $DEVICE
mkfs.ext3 $DEVICE
mount -t ext3 -o usrquota,grpquota $DEVICE /mnt/test
quotacheck -gu /mnt/test
umount /mnt/test
mount -t ext3 -o usrquota,grpquota $DEVICE /mnt/test
quotaon /mnt/test
dmsetup suspend $DEVICE
setquota -u root 1 2 3 4 /mnt/test &
sleep 1
dmsetup resume $DEVICE

setquota acquired semaphore s_umount for read and then tried to perform a
transaction (and waits because the device is suspended).  dmsetup resume tries
to acquire s_umount for write before resuming the device (and waits for
setquota).

Fix the deadlock by grabbing a thawed superblock for quota commands which need
it.

Reported-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Jan Kara <jack@suse.cz>
---
 fs/quota/quota.c |   24 +++++++++++++++++++++---
 1 files changed, 21 insertions(+), 3 deletions(-)

diff --git a/fs/quota/quota.c b/fs/quota/quota.c
index 7898cd6..fc2c438 100644
--- a/fs/quota/quota.c
+++ b/fs/quota/quota.c
@@ -292,11 +292,26 @@ static int do_quotactl(struct super_block *sb, int type, int cmd, qid_t id,
 	}
 }
 
+/* Return 1 if 'cmd' will block on frozen filesystem */
+static int quotactl_cmd_write(int cmd)
+{
+	switch (cmd) {
+	case Q_GETFMT:
+	case Q_GETINFO:
+	case Q_SYNC:
+	case Q_XGETQSTAT:
+	case Q_XGETQUOTA:
+	case Q_XQUOTASYNC:
+		return 0;
+	}
+	return 1;
+}
+
 /*
  * look up a superblock on which quota ops will be performed
  * - use the name of a block device to find the superblock thereon
  */
-static struct super_block *quotactl_block(const char __user *special)
+static struct super_block *quotactl_block(const char __user *special, int cmd)
 {
 #ifdef CONFIG_BLOCK
 	struct block_device *bdev;
@@ -309,7 +324,10 @@ static struct super_block *quotactl_block(const char __user *special)
 	putname(tmp);
 	if (IS_ERR(bdev))
 		return ERR_CAST(bdev);
-	sb = get_super(bdev);
+	if (quotactl_cmd_write(cmd))
+		sb = get_super_thawed(bdev);
+	else
+		sb = get_super(bdev);
 	bdput(bdev);
 	if (!sb)
 		return ERR_PTR(-ENODEV);
@@ -361,7 +379,7 @@ SYSCALL_DEFINE4(quotactl, unsigned int, cmd, const char __user *, special,
 			pathp = &path;
 	}
 
-	sb = quotactl_block(special);
+	sb = quotactl_block(special, cmds);
 	if (IS_ERR(sb)) {
 		ret = PTR_ERR(sb);
 		goto out;
-- 
1.7.1

Re: [PATCH 1/2] vfs: Provide function to get superblock and wait for it to thaw

[Al Viro]

On Fri, Feb 10, 2012 at 11:03:00AM +0100, Jan Kara wrote:
> In quota code we need to find a superblock corresponding to a device and wait
> for superblock to be unfrozen. However this waiting has to happen without
> s_umount semaphore because that is required for superblock to thaw. So provide
> a function in VFS for this to keep dances with s_umount where they belong.

Eww...  All it takes is
struct super_block *get_super_thawed(struct block_device *bdev)
{
        while (1) {
                struct super_block *s = get_super(bdev);
                if (!s || s->s_frozen == SB_UNFROZEN)
                        return s;
		up_read(&s->s_umount);
		vfs_check_frozen(s, SB_FREEZE_WRITE);
		put_super(s);
        }
}
and there's no need of extra arguments, etc. whatsoever.  Both patches
applied, with implementation of get_super_thawed() done as above.

Re: [PATCH 1/2] vfs: Provide function to get superblock and wait for it to thaw

[Jan Kara]

On Sun 12-02-12 21:13:25, Al Viro wrote:
> On Fri, Feb 10, 2012 at 11:03:00AM +0100, Jan Kara wrote:
> > In quota code we need to find a superblock corresponding to a device and wait
> > for superblock to be unfrozen. However this waiting has to happen without
> > s_umount semaphore because that is required for superblock to thaw. So provide
> > a function in VFS for this to keep dances with s_umount where they belong.
> 
> Eww...  All it takes is
> struct super_block *get_super_thawed(struct block_device *bdev)
> {
>         while (1) {
>                 struct super_block *s = get_super(bdev);
>                 if (!s || s->s_frozen == SB_UNFROZEN)
>                         return s;
> 		up_read(&s->s_umount);
> 		vfs_check_frozen(s, SB_FREEZE_WRITE);
> 		put_super(s);
>         }
> }
> and there's no need of extra arguments, etc. whatsoever.  Both patches
> applied, with implementation of get_super_thawed() done as above.
  Yeah, this looks better. Thanks!

									Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR

Re: [PATCH 2/2] quota: Fix deadlock with suspend and quotas

[Eric Sandeen]

On 2/7/12 4:37 PM, Jan Kara wrote:
> This script causes a kernel deadlock:
> set -e
> DEVICE=/dev/vg1/linear
> lvchange -ay $DEVICE
> mkfs.ext3 $DEVICE
> mount -t ext3 -o usrquota,grpquota $DEVICE /mnt/test
> quotacheck -gu /mnt/test
> umount /mnt/test
> mount -t ext3 -o usrquota,grpquota $DEVICE /mnt/test
> quotaon /mnt/test
> dmsetup suspend $DEVICE
> setquota -u root 1 2 3 4 /mnt/test &
> sleep 1
> dmsetup resume $DEVICE
> 
> setquota acquired semaphore s_umount for read and then tried to perform a
> transaction (and waits because the device is suspended).  dmsetup resume tries
> to acquire s_umount for write before resuming the device (and waits for
> setquota).
> 
> Fix the deadlock by grabbing a thawed superblock for quota commands which need
> it.
> 
> Reported-by: Mikulas Patocka <mpatocka@redhat.com>
> Signed-off-by: Jan Kara <jack@suse.cz>

Makes sense to me, thanks.

Reviewed-by: Eric Sandeen <sandeen@redhat.com>

> ---
>  fs/quota/quota.c |   24 +++++++++++++++++++++---
>  1 files changed, 21 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/quota/quota.c b/fs/quota/quota.c
> index 7898cd6..fc2c438 100644
> --- a/fs/quota/quota.c
> +++ b/fs/quota/quota.c
> @@ -292,11 +292,26 @@ static int do_quotactl(struct super_block *sb, int type, int cmd, qid_t id,
>  	}
>  }
>  
> +/* Return 1 if 'cmd' will block on frozen filesystem */
> +static int quotactl_cmd_write(int cmd)
> +{
> +	switch (cmd) {
> +	case Q_GETFMT:
> +	case Q_GETINFO:
> +	case Q_SYNC:
> +	case Q_XGETQSTAT:
> +	case Q_XGETQUOTA:
> +	case Q_XQUOTASYNC:
> +		return 0;
> +	}
> +	return 1;
> +}
> +
>  /*
>   * look up a superblock on which quota ops will be performed
>   * - use the name of a block device to find the superblock thereon
>   */
> -static struct super_block *quotactl_block(const char __user *special)
> +static struct super_block *quotactl_block(const char __user *special, int cmd)
>  {
>  #ifdef CONFIG_BLOCK
>  	struct block_device *bdev;
> @@ -309,7 +324,10 @@ static struct super_block *quotactl_block(const char __user *special)
>  	putname(tmp);
>  	if (IS_ERR(bdev))
>  		return ERR_CAST(bdev);
> -	sb = get_super(bdev);
> +	if (quotactl_cmd_write(cmd))
> +		sb = get_super_thawed(bdev);
> +	else
> +		sb = get_super(bdev);
>  	bdput(bdev);
>  	if (!sb)
>  		return ERR_PTR(-ENODEV);
> @@ -361,7 +379,7 @@ SYSCALL_DEFINE4(quotactl, unsigned int, cmd, const char __user *, special,
>  			pathp = &path;
>  	}
>  
> -	sb = quotactl_block(special);
> +	sb = quotactl_block(special, cmds);
>  	if (IS_ERR(sb)) {
>  		ret = PTR_ERR(sb);
>  		goto out;

[PATCH 2/2] quota: Fix deadlock with suspend and quotas

[Jan Kara]

This script causes a kernel deadlock:
set -e
DEVICE=/dev/vg1/linear
lvchange -ay $DEVICE
mkfs.ext3 $DEVICE
mount -t ext3 -o usrquota,grpquota $DEVICE /mnt/test
quotacheck -gu /mnt/test
umount /mnt/test
mount -t ext3 -o usrquota,grpquota $DEVICE /mnt/test
quotaon /mnt/test
dmsetup suspend $DEVICE
setquota -u root 1 2 3 4 /mnt/test &
sleep 1
dmsetup resume $DEVICE

setquota acquired semaphore s_umount for read and then tried to perform a
transaction (and waits because the device is suspended).  dmsetup resume tries
to acquire s_umount for write before resuming the device (and waits for
setquota).

Fix the deadlock by grabbing a thawed superblock for quota commands which need
it.

Reported-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Jan Kara <jack@suse.cz>
---
 fs/quota/quota.c |   24 +++++++++++++++++++++---
 1 files changed, 21 insertions(+), 3 deletions(-)

diff --git a/fs/quota/quota.c b/fs/quota/quota.c
index 7898cd6..fc2c438 100644
--- a/fs/quota/quota.c
+++ b/fs/quota/quota.c
@@ -292,11 +292,26 @@ static int do_quotactl(struct super_block *sb, int type, int cmd, qid_t id,
 	}
 }
 
+/* Return 1 if 'cmd' will block on frozen filesystem */
+static int quotactl_cmd_write(int cmd)
+{
+	switch (cmd) {
+	case Q_GETFMT:
+	case Q_GETINFO:
+	case Q_SYNC:
+	case Q_XGETQSTAT:
+	case Q_XGETQUOTA:
+	case Q_XQUOTASYNC:
+		return 0;
+	}
+	return 1;
+}
+
 /*
  * look up a superblock on which quota ops will be performed
  * - use the name of a block device to find the superblock thereon
  */
-static struct super_block *quotactl_block(const char __user *special)
+static struct super_block *quotactl_block(const char __user *special, int cmd)
 {
 #ifdef CONFIG_BLOCK
 	struct block_device *bdev;
@@ -309,7 +324,10 @@ static struct super_block *quotactl_block(const char __user *special)
 	putname(tmp);
 	if (IS_ERR(bdev))
 		return ERR_CAST(bdev);
-	sb = get_super(bdev);
+	if (quotactl_cmd_write(cmd))
+		sb = get_super_thawed(bdev);
+	else
+		sb = get_super(bdev);
 	bdput(bdev);
 	if (!sb)
 		return ERR_PTR(-ENODEV);
@@ -361,7 +379,7 @@ SYSCALL_DEFINE4(quotactl, unsigned int, cmd, const char __user *, special,
 			pathp = &path;
 	}
 
-	sb = quotactl_block(special);
+	sb = quotactl_block(special, cmds);
 	if (IS_ERR(sb)) {
 		ret = PTR_ERR(sb);
 		goto out;
-- 
1.7.1