[PATCH-block 0/3] blk-cgroup: Fix potential UAF & miscellaneous cleanup

[PATCH-block 0/3] blk-cgroup: Fix potential UAF & miscellaneous cleanup

[Waiman Long]

It was found that blkcg_destroy_blkgs() may be called with all blkcg
references gone. This may potentially cause user-after-free and so
should be fixed. The last 2 patches are miscellaneous cleanups of
commit 3b8cc6298724 ("blk-cgroup: Optimize blkcg_rstat_flush()").

Waiman Long (3):
  bdi, blk-cgroup: Fix potential UAF of blkcg
  blk-cgroup: Don't flush a blkg if destroyed
  blk-cgroup: Flush stats at blkgs destruction path

 block/blk-cgroup.c     | 26 ++++++++++++++++++++++++++
 include/linux/cgroup.h |  1 +
 kernel/cgroup/rstat.c  | 20 ++++++++++++++++++++
 mm/backing-dev.c       |  8 ++++++--
 4 files changed, 53 insertions(+), 2 deletions(-)

-- 
2.31.1

[PATCH-block 1/3] bdi, blk-cgroup: Fix potential UAF of blkcg

[Waiman Long]

Commit 59b57717fff8 ("blkcg: delay blkg destruction until after
writeback has finished") delayed call to blkcg_destroy_blkgs() to
cgwb_release_workfn(). However, it is done after a css_put() of blkcg
which may be the final put that causes the blkcg to be freed as RCU
read lock isn't held.

Another place where blkcg_destroy_blkgs() can be called indirectly via
blkcg_unpin_online() is from the offline_css() function called from
css_killed_work_fn(). Over there, the potentially final css_put() call
is issued after offline_css().

By adding a css_tryget() into blkcg_destroy_blkgs() and warning its
failure, the following stack trace was produced in a test system on
bootup.

[   34.254240] RIP: 0010:blkcg_destroy_blkgs+0x16a/0x1a0
      :
[   34.339943] Call Trace:
[   34.344510]  blkcg_unpin_online+0x38/0x60
[   34.348523]  cgwb_release_workfn+0x6a/0x200
[   34.352708]  process_one_work+0x1e5/0x3b0
[   34.360758]  worker_thread+0x50/0x3a0
[   34.368447]  kthread+0xd9/0x100
[   34.376386]  ret_from_fork+0x22/0x30

This confirms that a potential UAF situation can really happen in
cgwb_release_workfn().

Fix that by delaying the css_put() until after the blkcg_unpin_online()
call. Also use css_tryget() in blkcg_destroy_blkgs() and issue a warning
if css_tryget() fails.

The reproducing system can no longer produce a warning with this patch.
All the runnable block/0* tests including block/027 were run successfully
without failure.

Fixes: 59b57717fff8 ("blkcg: delay blkg destruction until after writeback has finished")
Suggested-by: Michal Koutný <mkoutny@suse.com>
Reported-by: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Waiman Long <longman@redhat.com>
---
 block/blk-cgroup.c | 8 ++++++++
 mm/backing-dev.c   | 8 ++++++--
 2 files changed, 14 insertions(+), 2 deletions(-)

diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
index 1bb939d3b793..21cc88349f21 100644
--- a/block/blk-cgroup.c
+++ b/block/blk-cgroup.c
@@ -1084,6 +1084,13 @@ struct list_head *blkcg_get_cgwb_list(struct cgroup_subsys_state *css)
  */
 static void blkcg_destroy_blkgs(struct blkcg *blkcg)
 {
+	/*
+	 * blkcg_destroy_blkgs() shouldn't be called with all the blkcg
+	 * references gone.
+	 */
+	if (WARN_ON_ONCE(!css_tryget(&blkcg->css)))
+		return;
+
 	might_sleep();
 
 	spin_lock_irq(&blkcg->lock);
@@ -1110,6 +1117,7 @@ static void blkcg_destroy_blkgs(struct blkcg *blkcg)
 	}
 
 	spin_unlock_irq(&blkcg->lock);
+	css_put(&blkcg->css);
 }
 
 /**
diff --git a/mm/backing-dev.c b/mm/backing-dev.c
index c30419a5e119..36f75b072325 100644
--- a/mm/backing-dev.c
+++ b/mm/backing-dev.c
@@ -390,11 +390,15 @@ static void cgwb_release_workfn(struct work_struct *work)
 	wb_shutdown(wb);
 
 	css_put(wb->memcg_css);
-	css_put(wb->blkcg_css);
 	mutex_unlock(&wb->bdi->cgwb_release_mutex);
 
-	/* triggers blkg destruction if no online users left */
+	/*
+	 * Triggers blkg destruction if no online users left
+	 * The final blkcg css_put() has to be done after blkcg_unpin_online()
+	 * to avoid use-after-free.
+	 */
 	blkcg_unpin_online(wb->blkcg_css);
+	css_put(wb->blkcg_css);
 
 	fprop_local_destroy_percpu(&wb->memcg_completions);
 
-- 
2.31.1

[PATCH-block 2/3] blk-cgroup: Don't flush a blkg if destroyed

[Waiman Long]

Before commit 3b8cc6298724 ("blk-cgroup: Optimize blkcg_rstat_flush()"),
blkg's stats is only flushed if they are online. In addition, the
stat flushing of blkgs in blkcg_rstat_flush() includes propagating
the rstat data to its parent. However, if a blkg has been destroyed
(offline), the validity of its parent may be questionable. For safety,
revert back to the old behavior by ignoring offline blkg's.

Signed-off-by: Waiman Long <longman@redhat.com>
---
 block/blk-cgroup.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
index 21cc88349f21..c466aef0d467 100644
--- a/block/blk-cgroup.c
+++ b/block/blk-cgroup.c
@@ -885,6 +885,12 @@ static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu)
 
 		WRITE_ONCE(bisc->lqueued, false);
 
+		/* Don't flush its stats if blkg is offline */
+		if (unlikely(!blkg->online)) {
+			percpu_ref_put(&blkg->refcnt);
+			continue;
+		}
+
 		/* fetch the current per-cpu values */
 		do {
 			seq = u64_stats_fetch_begin(&bisc->sync);
-- 
2.31.1

[PATCH-block 3/3] blk-cgroup: Flush stats at blkgs destruction path

[Waiman Long]

As noted by Michal, the blkg_iostat_set's in the lockless list
hold reference to blkg's to protect against their removal. Those
blkg's hold reference to blkcg. When a cgroup is being destroyed,
cgroup_rstat_flush() is only called at css_release_work_fn() which is
called when the blkcg reference count reaches 0. This circular dependency
will prevent blkcg from being freed until some other events cause
cgroup_rstat_flush() to be called to flush out the pending blkcg stats.

To prevent this delayed blkcg removal, add a new cgroup_rstat_css_flush()
function to flush stats for a given css and cpu and call it at the blkgs
destruction path, blkcg_destroy_blkgs(), whenever there are still some
pending stats to be flushed. This will ensure that blkcg reference
count can reach 0 ASAP.

Signed-off-by: Waiman Long <longman@redhat.com>
---
 block/blk-cgroup.c     | 12 ++++++++++++
 include/linux/cgroup.h |  1 +
 kernel/cgroup/rstat.c  | 20 ++++++++++++++++++++
 3 files changed, 33 insertions(+)

diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
index c466aef0d467..534f3baeb84a 100644
--- a/block/blk-cgroup.c
+++ b/block/blk-cgroup.c
@@ -1090,6 +1090,8 @@ struct list_head *blkcg_get_cgwb_list(struct cgroup_subsys_state *css)
  */
 static void blkcg_destroy_blkgs(struct blkcg *blkcg)
 {
+	int cpu;
+
 	/*
 	 * blkcg_destroy_blkgs() shouldn't be called with all the blkcg
 	 * references gone.
@@ -1099,6 +1101,16 @@ static void blkcg_destroy_blkgs(struct blkcg *blkcg)
 
 	might_sleep();
 
+	/*
+	 * Flush all the non-empty percpu lockless lists.
+	 */
+	for_each_possible_cpu(cpu) {
+		struct llist_head *lhead = per_cpu_ptr(blkcg->lhead, cpu);
+
+		if (!llist_empty(lhead))
+			cgroup_rstat_css_cpu_flush(&blkcg->css, cpu);
+	}
+
 	spin_lock_irq(&blkcg->lock);
 
 	while (!hlist_empty(&blkcg->blkg_list)) {
diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h
index 528bd44b59e2..6c4e66b3fa84 100644
--- a/include/linux/cgroup.h
+++ b/include/linux/cgroup.h
@@ -766,6 +766,7 @@ void cgroup_rstat_flush(struct cgroup *cgrp);
 void cgroup_rstat_flush_irqsafe(struct cgroup *cgrp);
 void cgroup_rstat_flush_hold(struct cgroup *cgrp);
 void cgroup_rstat_flush_release(void);
+void cgroup_rstat_css_cpu_flush(struct cgroup_subsys_state *css, int cpu);
 
 /*
  * Basic resource stats.
diff --git a/kernel/cgroup/rstat.c b/kernel/cgroup/rstat.c
index 793ecff29038..910e633869b0 100644
--- a/kernel/cgroup/rstat.c
+++ b/kernel/cgroup/rstat.c
@@ -281,6 +281,26 @@ void cgroup_rstat_flush_release(void)
 	spin_unlock_irq(&cgroup_rstat_lock);
 }
 
+/**
+ * cgroup_rstat_css_cpu_flush - flush stats for the given css and cpu
+ * @css: target css to be flush
+ * @cpu: the cpu that holds the stats to be flush
+ *
+ * A lightweight rstat flush operation for a given css and cpu.
+ * Only the cpu_lock is being held for mutual exclusion, the cgroup_rstat_lock
+ * isn't used.
+ */
+void cgroup_rstat_css_cpu_flush(struct cgroup_subsys_state *css, int cpu)
+{
+	raw_spinlock_t *cpu_lock = per_cpu_ptr(&cgroup_rstat_cpu_lock, cpu);
+
+	raw_spin_lock_irq(cpu_lock);
+	rcu_read_lock();
+	css->ss->css_rstat_flush(css, cpu);
+	rcu_read_unlock();
+	raw_spin_unlock_irq(cpu_lock);
+}
+
 int cgroup_rstat_init(struct cgroup *cgrp)
 {
 	int cpu;
-- 
2.31.1

Re: [PATCH-block 3/3] blk-cgroup: Flush stats at blkgs destruction path

[Jens Axboe]

On 12/8/22 3:01?PM, Waiman Long wrote:
> diff --git a/kernel/cgroup/rstat.c b/kernel/cgroup/rstat.c
> index 793ecff29038..910e633869b0 100644
> --- a/kernel/cgroup/rstat.c
> +++ b/kernel/cgroup/rstat.c
> @@ -281,6 +281,26 @@ void cgroup_rstat_flush_release(void)
>  	spin_unlock_irq(&cgroup_rstat_lock);
>  }
>  
> +/**
> + * cgroup_rstat_css_cpu_flush - flush stats for the given css and cpu
> + * @css: target css to be flush
> + * @cpu: the cpu that holds the stats to be flush
> + *
> + * A lightweight rstat flush operation for a given css and cpu.
> + * Only the cpu_lock is being held for mutual exclusion, the cgroup_rstat_lock
> + * isn't used.
> + */
> +void cgroup_rstat_css_cpu_flush(struct cgroup_subsys_state *css, int cpu)
> +{
> +	raw_spinlock_t *cpu_lock = per_cpu_ptr(&cgroup_rstat_cpu_lock, cpu);
> +
> +	raw_spin_lock_irq(cpu_lock);
> +	rcu_read_lock();
> +	css->ss->css_rstat_flush(css, cpu);
> +	rcu_read_unlock();
> +	raw_spin_unlock_irq(cpu_lock);
> +}
> +
>  int cgroup_rstat_init(struct cgroup *cgrp)
>  {
>  	int cpu;

As I mentioned last time, raw_spin_lock_irq() will be equivalent to an
RCU protected section anyway, so you don't need to do both. Just add a
comment on why rcu_read_lock()/rcu_read_unlock() isn't needed inside the
raw irq safe lock.

-- 
Jens Axboe

Re: [PATCH-block 3/3] blk-cgroup: Flush stats at blkgs destruction path

[Waiman Long]

On 12/8/22 18:00, Jens Axboe wrote:
> On 12/8/22 3:01?PM, Waiman Long wrote:
>> diff --git a/kernel/cgroup/rstat.c b/kernel/cgroup/rstat.c
>> index 793ecff29038..910e633869b0 100644
>> --- a/kernel/cgroup/rstat.c
>> +++ b/kernel/cgroup/rstat.c
>> @@ -281,6 +281,26 @@ void cgroup_rstat_flush_release(void)
>>   	spin_unlock_irq(&cgroup_rstat_lock);
>>   }
>>   
>> +/**
>> + * cgroup_rstat_css_cpu_flush - flush stats for the given css and cpu
>> + * @css: target css to be flush
>> + * @cpu: the cpu that holds the stats to be flush
>> + *
>> + * A lightweight rstat flush operation for a given css and cpu.
>> + * Only the cpu_lock is being held for mutual exclusion, the cgroup_rstat_lock
>> + * isn't used.
>> + */
>> +void cgroup_rstat_css_cpu_flush(struct cgroup_subsys_state *css, int cpu)
>> +{
>> +	raw_spinlock_t *cpu_lock = per_cpu_ptr(&cgroup_rstat_cpu_lock, cpu);
>> +
>> +	raw_spin_lock_irq(cpu_lock);
>> +	rcu_read_lock();
>> +	css->ss->css_rstat_flush(css, cpu);
>> +	rcu_read_unlock();
>> +	raw_spin_unlock_irq(cpu_lock);
>> +}
>> +
>>   int cgroup_rstat_init(struct cgroup *cgrp)
>>   {
>>   	int cpu;
> As I mentioned last time, raw_spin_lock_irq() will be equivalent to an
> RCU protected section anyway, so you don't need to do both. Just add a
> comment on why rcu_read_lock()/rcu_read_unlock() isn't needed inside the
> raw irq safe lock.

Yes, you are right.  We don't need rcu_read_lock() here. I put it there 
to follow the locking pattern in cgroup_rstat_flush_locked(). I will 
remove it in the next version.

Cheers,
Longman