On Tue, Jun 08, 2021 at 04:02:20PM -0700, Roman Gushchin wrote:
> isw_nr_in_flight is used do determine whether the inode switch queue
> should be flushed from the umount path. Currently it's increased
> after grabbing an inode and even scheduling the switch work. It means
> the umount path can be walked past cleanup_offline_cgwb() with active
> inode references, which can result in a "Busy inodes after unmount."
> message and use-after-free issues (with inode->i_sb which gets freed).
>
> Fix it by incrementing isw_nr_in_flight before doing anything with
> the inode and decrementing in the case when switching wasn't scheduled.
>
> The problem hasn't yet been seen in the real life and was discovered
> by Jan Kara by looking into the code.
>
> Suggested-by: Jan Kara <[email protected]>
> Signed-off-by: Roman Gushchin <[email protected]>
> Reviewed-by: Jan Kara <[email protected]>
> ---
> fs/fs-writeback.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
> index b6fc13a4962d..4413e005c28c 100644
> --- a/fs/fs-writeback.c
> +++ b/fs/fs-writeback.c
> @@ -505,6 +505,8 @@ static void inode_switch_wbs(struct inode *inode, int new_wb_id)
> if (!isw)
> return;
>
> + atomic_inc(&isw_nr_in_flight);
smp_mb() may be required for ordering the WRITE in 'atomic_inc(&isw_nr_in_flight)'
and the following READ on 'inode->i_sb->s_flags & SB_ACTIVE'. Otherwise,
cgroup_writeback_umount() may observe zero of 'isw_nr_in_flight' because of
re-order of the two OPs, then miss the flush_workqueue().
Also this barrier should serve as pair of the one added in cgroup_writeback_umount(),
so maybe this patch should be merged with 2/8.
Thanks,
Ming
On Wed, Jun 09, 2021 at 11:32:44AM +0800, Ming Lei wrote:
> On Tue, Jun 08, 2021 at 04:02:20PM -0700, Roman Gushchin wrote:
> > isw_nr_in_flight is used do determine whether the inode switch queue
> > should be flushed from the umount path. Currently it's increased
> > after grabbing an inode and even scheduling the switch work. It means
> > the umount path can be walked past cleanup_offline_cgwb() with active
> > inode references, which can result in a "Busy inodes after unmount."
> > message and use-after-free issues (with inode->i_sb which gets freed).
> >
> > Fix it by incrementing isw_nr_in_flight before doing anything with
> > the inode and decrementing in the case when switching wasn't scheduled.
> >
> > The problem hasn't yet been seen in the real life and was discovered
> > by Jan Kara by looking into the code.
> >
> > Suggested-by: Jan Kara <[email protected]>
> > Signed-off-by: Roman Gushchin <[email protected]>
> > Reviewed-by: Jan Kara <[email protected]>
> > ---
> > fs/fs-writeback.c | 5 +++--
> > 1 file changed, 3 insertions(+), 2 deletions(-)
> >
> > diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
> > index b6fc13a4962d..4413e005c28c 100644
> > --- a/fs/fs-writeback.c
> > +++ b/fs/fs-writeback.c
> > @@ -505,6 +505,8 @@ static void inode_switch_wbs(struct inode *inode, int new_wb_id)
> > if (!isw)
> > return;
> >
> > + atomic_inc(&isw_nr_in_flight);
>
> smp_mb() may be required for ordering the WRITE in 'atomic_inc(&isw_nr_in_flight)'
> and the following READ on 'inode->i_sb->s_flags & SB_ACTIVE'. Otherwise,
> cgroup_writeback_umount() may observe zero of 'isw_nr_in_flight' because of
> re-order of the two OPs, then miss the flush_workqueue().
>
> Also this barrier should serve as pair of the one added in cgroup_writeback_umount(),
> so maybe this patch should be merged with 2/8.
Hi Ming!
Good point, I agree. How about a patch below?
Thanks!
--
From 282861286074c47907759d80c01419f0d0630dae Mon Sep 17 00:00:00 2001
From: Roman Gushchin <[email protected]>
Date: Wed, 9 Jun 2021 14:14:26 -0700
Subject: [PATCH] cgroup, writeback: add smp_mb() to inode_prepare_wbs_switch()
Add a memory barrier between incrementing isw_nr_in_flight
and checking the sb's SB_ACTIVE flag and grabbing an inode in
inode_prepare_wbs_switch(). It's required to prevent grabbing
an inode before incrementing isw_nr_in_flight, otherwise
0 can be obtained as isw_nr_in_flight in cgroup_writeback_umount()
and isw_wq will not be flushed, potentially leading to a memory
corruption.
Added smp_mb() will work in pair with smp_mb() in
cgroup_writeback_umount().
Suggested-by: Ming Lei <[email protected]>
Signed-off-by: Roman Gushchin <[email protected]>
---
fs/fs-writeback.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
index 545fce68e919..6332b86ca4ed 100644
--- a/fs/fs-writeback.c
+++ b/fs/fs-writeback.c
@@ -513,6 +513,14 @@ static void inode_switch_wbs_work_fn(struct work_struct *work)
static bool inode_prepare_wbs_switch(struct inode *inode,
struct bdi_writeback *new_wb)
{
+ /*
+ * Paired with smp_mb() in cgroup_writeback_umount().
+ * isw_nr_in_flight must be increased before checking SB_ACTIVE and
+ * grabbing an inode, otherwise isw_nr_in_flight can be observed as 0
+ * in cgroup_writeback_umount() and the isw_wq will be not flushed.
+ */
+ smp_mb();
+
/* while holding I_WB_SWITCH, no one else can update the association */
spin_lock(&inode->i_lock);
if (!(inode->i_sb->s_flags & SB_ACTIVE) ||
--
2.31.1
On Wed, Jun 09, 2021 at 05:21:14PM -0700, Roman Gushchin wrote:
> On Wed, Jun 09, 2021 at 11:32:44AM +0800, Ming Lei wrote:
> > On Tue, Jun 08, 2021 at 04:02:20PM -0700, Roman Gushchin wrote:
> > > isw_nr_in_flight is used do determine whether the inode switch queue
> > > should be flushed from the umount path. Currently it's increased
> > > after grabbing an inode and even scheduling the switch work. It means
> > > the umount path can be walked past cleanup_offline_cgwb() with active
> > > inode references, which can result in a "Busy inodes after unmount."
> > > message and use-after-free issues (with inode->i_sb which gets freed).
> > >
> > > Fix it by incrementing isw_nr_in_flight before doing anything with
> > > the inode and decrementing in the case when switching wasn't scheduled.
> > >
> > > The problem hasn't yet been seen in the real life and was discovered
> > > by Jan Kara by looking into the code.
> > >
> > > Suggested-by: Jan Kara <[email protected]>
> > > Signed-off-by: Roman Gushchin <[email protected]>
> > > Reviewed-by: Jan Kara <[email protected]>
> > > ---
> > > fs/fs-writeback.c | 5 +++--
> > > 1 file changed, 3 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
> > > index b6fc13a4962d..4413e005c28c 100644
> > > --- a/fs/fs-writeback.c
> > > +++ b/fs/fs-writeback.c
> > > @@ -505,6 +505,8 @@ static void inode_switch_wbs(struct inode *inode, int new_wb_id)
> > > if (!isw)
> > > return;
> > >
> > > + atomic_inc(&isw_nr_in_flight);
> >
> > smp_mb() may be required for ordering the WRITE in 'atomic_inc(&isw_nr_in_flight)'
> > and the following READ on 'inode->i_sb->s_flags & SB_ACTIVE'. Otherwise,
> > cgroup_writeback_umount() may observe zero of 'isw_nr_in_flight' because of
> > re-order of the two OPs, then miss the flush_workqueue().
> >
> > Also this barrier should serve as pair of the one added in cgroup_writeback_umount(),
> > so maybe this patch should be merged with 2/8.
>
> Hi Ming!
>
> Good point, I agree. How about a patch below?
>
> Thanks!
>
> --
>
> From 282861286074c47907759d80c01419f0d0630dae Mon Sep 17 00:00:00 2001
> From: Roman Gushchin <[email protected]>
> Date: Wed, 9 Jun 2021 14:14:26 -0700
> Subject: [PATCH] cgroup, writeback: add smp_mb() to inode_prepare_wbs_switch()
>
> Add a memory barrier between incrementing isw_nr_in_flight
> and checking the sb's SB_ACTIVE flag and grabbing an inode in
> inode_prepare_wbs_switch(). It's required to prevent grabbing
> an inode before incrementing isw_nr_in_flight, otherwise
> 0 can be obtained as isw_nr_in_flight in cgroup_writeback_umount()
> and isw_wq will not be flushed, potentially leading to a memory
> corruption.
>
> Added smp_mb() will work in pair with smp_mb() in
> cgroup_writeback_umount().
>
> Suggested-by: Ming Lei <[email protected]>
> Signed-off-by: Roman Gushchin <[email protected]>
> ---
> fs/fs-writeback.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
> index 545fce68e919..6332b86ca4ed 100644
> --- a/fs/fs-writeback.c
> +++ b/fs/fs-writeback.c
> @@ -513,6 +513,14 @@ static void inode_switch_wbs_work_fn(struct work_struct *work)
> static bool inode_prepare_wbs_switch(struct inode *inode,
> struct bdi_writeback *new_wb)
> {
> + /*
> + * Paired with smp_mb() in cgroup_writeback_umount().
> + * isw_nr_in_flight must be increased before checking SB_ACTIVE and
> + * grabbing an inode, otherwise isw_nr_in_flight can be observed as 0
> + * in cgroup_writeback_umount() and the isw_wq will be not flushed.
> + */
> + smp_mb();
> +
> /* while holding I_WB_SWITCH, no one else can update the association */
> spin_lock(&inode->i_lock);
> if (!(inode->i_sb->s_flags & SB_ACTIVE) ||
Looks fine, you may have to merge this one with 2/8 & 3/8, so the memory
barrier use can be correct & intact for avoiding the race between switching
cgwb and generic_shutdown_super().
Thanks,
Ming