[PATCH 0/4] block: fix partition use-after-free and optimization

[PATCH 0/4] block: fix partition use-after-free and optimization

[Ming Lei]

Hi,

The 1st patch fixes one use-after-free on cached last_lookup partition.

The other 3 patches optimizes partition uses in IO path.



Ming Lei (4):
  block: fix use-after-free on cached last_lookup partition
  block: only define 'nr_sects_seq' in hd_part for 32bit SMP
  block: re-organize fields of 'struct hd_part'
  block: don't hold part0's refcount in IO path

 block/blk-core.c          | 15 ++-------------
 block/genhd.c             |  7 +++++--
 block/partition-generic.c | 12 ++++++++++--
 include/linux/genhd.h     | 25 ++++++++++++++++++-------
 4 files changed, 35 insertions(+), 24 deletions(-)

Cc: Yufen Yu <yuyufen@huawei.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hou Tao <houtao1@huawei.com>

-- 
2.20.1

[PATCH 1/4] block: fix use-after-free on cached last_lookup partition

[Ming Lei]

delete_partition() clears the cached last_lookup partition. However
the .last_lookup cache may be overwritten by one IO path after
it is cleared from delete_partition(). Then another IO path may
use the cached deleting partition after __delete_partition() is
called, then use-after-free is triggered on the cached partition.

Fixes the issue by the following approach:

1) always get the partition's refcount via hd_struct_try_get() before
setting .last_lookup

2) move clearing .last_lookup from delete_partition() to
__delete_partition() which is release handle of the partition's
percpu-refcount, so that no IO path can overwrite .last_lookup after it
is cleared in __delete_partition().

It is one candidate approach of Yufen's patch[1] which adds overhead
in fast path by indirect lookup which may introduce one extra cacheline
in IO path. Also this patch relies on percpu-refcount's protection, and
it is easier to understand and verify.

[1] https://lore.kernel.org/linux-block/20200109013551.GB9655@ming.t460p/T/#t

Reported-by: Yufen Yu <yuyufen@huawei.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hou Tao <houtao1@huawei.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 block/blk-core.c          | 12 ------------
 block/genhd.c             |  6 +++++-
 block/partition-generic.c | 10 +++++++++-
 include/linux/genhd.h     |  1 +
 4 files changed, 15 insertions(+), 14 deletions(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index 089e890ab208..79599f5fd5b7 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -1365,18 +1365,6 @@ void blk_account_io_start(struct request *rq, bool new_io)
 		part_stat_inc(part, merges[rw]);
 	} else {
 		part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
-		if (!hd_struct_try_get(part)) {
-			/*
-			 * The partition is already being removed,
-			 * the request will be accounted on the disk only
-			 *
-			 * We take a reference on disk->part0 although that
-			 * partition will never be deleted, so we can treat
-			 * it as any other partition.
-			 */
-			part = &rq->rq_disk->part0;
-			hd_struct_get(part);
-		}
 		part_inc_in_flight(rq->q, part, rw);
 		rq->part = part;
 	}
diff --git a/block/genhd.c b/block/genhd.c
index ff6268970ddc..6029c94510f0 100644
--- a/block/genhd.c
+++ b/block/genhd.c
@@ -286,17 +286,21 @@ struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
 	ptbl = rcu_dereference(disk->part_tbl);
 
 	part = rcu_dereference(ptbl->last_lookup);
-	if (part && sector_in_part(part, sector))
+	if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
 		return part;
 
 	for (i = 1; i < ptbl->len; i++) {
 		part = rcu_dereference(ptbl->part[i]);
 
 		if (part && sector_in_part(part, sector)) {
+			if (!hd_struct_try_get(part))
+				goto exit;
 			rcu_assign_pointer(ptbl->last_lookup, part);
 			return part;
 		}
 	}
+ exit:
+	hd_struct_get(&disk->part0);
 	return &disk->part0;
 }
 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
diff --git a/block/partition-generic.c b/block/partition-generic.c
index 1d20c9cf213f..1739f750dbf2 100644
--- a/block/partition-generic.c
+++ b/block/partition-generic.c
@@ -262,6 +262,12 @@ static void delete_partition_work_fn(struct work_struct *work)
 void __delete_partition(struct percpu_ref *ref)
 {
 	struct hd_struct *part = container_of(ref, struct hd_struct, ref);
+	struct disk_part_tbl *ptbl =
+		rcu_dereference_protected(part->disk->part_tbl, 1);
+
+	rcu_assign_pointer(ptbl->last_lookup, NULL);
+	put_device(disk_to_dev(part->disk));
+
 	INIT_RCU_WORK(&part->rcu_work, delete_partition_work_fn);
 	queue_rcu_work(system_wq, &part->rcu_work);
 }
@@ -283,8 +289,9 @@ void delete_partition(struct gendisk *disk, int partno)
 	if (!part)
 		return;
 
+	get_device(disk_to_dev(disk));
 	rcu_assign_pointer(ptbl->part[partno], NULL);
-	rcu_assign_pointer(ptbl->last_lookup, NULL);
+
 	kobject_put(part->holder_dir);
 	device_del(part_to_dev(part));
 
@@ -349,6 +356,7 @@ struct hd_struct *add_partition(struct gendisk *disk, int partno,
 	p->nr_sects = len;
 	p->partno = partno;
 	p->policy = get_disk_ro(disk);
+	p->disk = disk;
 
 	if (info) {
 		struct partition_meta_info *pinfo = alloc_part_info(disk);
diff --git a/include/linux/genhd.h b/include/linux/genhd.h
index 8bb63027e4d6..1b09cfe00aa3 100644
--- a/include/linux/genhd.h
+++ b/include/linux/genhd.h
@@ -130,6 +130,7 @@ struct hd_struct {
 	struct disk_stats dkstats;
 #endif
 	struct percpu_ref ref;
+	struct gendisk *disk;
 	struct rcu_work rcu_work;
 };
 
-- 
2.20.1

[PATCH 2/4] block: only define 'nr_sects_seq' in hd_part for 32bit SMP

[Ming Lei]

The seqcount of 'nr_sects_seq' is only needed in case of 32bit SMP,
so define it just for 32bit SMP.

Cc: Yufen Yu <yuyufen@huawei.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hou Tao <houtao1@huawei.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 block/genhd.c             | 2 +-
 block/partition-generic.c | 2 +-
 include/linux/genhd.h     | 9 +++++++++
 3 files changed, 11 insertions(+), 2 deletions(-)

diff --git a/block/genhd.c b/block/genhd.c
index 6029c94510f0..bfc4148ec341 100644
--- a/block/genhd.c
+++ b/block/genhd.c
@@ -1506,7 +1506,7 @@ struct gendisk *__alloc_disk_node(int minors, int node_id)
 		 * TODO: Ideally set_capacity() and get_capacity() should be
 		 * converted to make use of bd_mutex and sequence counters.
 		 */
-		seqcount_init(&disk->part0.nr_sects_seq);
+		hd_sects_seq_init(&disk->part0);
 		if (hd_ref_init(&disk->part0)) {
 			hd_free_part(&disk->part0);
 			kfree(disk);
diff --git a/block/partition-generic.c b/block/partition-generic.c
index 1739f750dbf2..eb606c2c3d6c 100644
--- a/block/partition-generic.c
+++ b/block/partition-generic.c
@@ -345,7 +345,7 @@ struct hd_struct *add_partition(struct gendisk *disk, int partno,
 		goto out_free;
 	}
 
-	seqcount_init(&p->nr_sects_seq);
+	hd_sects_seq_init(p);
 	pdev = part_to_dev(p);
 
 	p->start_sect = start;
diff --git a/include/linux/genhd.h b/include/linux/genhd.h
index 1b09cfe00aa3..5f5718ce5e86 100644
--- a/include/linux/genhd.h
+++ b/include/linux/genhd.h
@@ -113,7 +113,9 @@ struct hd_struct {
 	 * can be non-atomic on 32bit machines with 64bit sector_t.
 	 */
 	sector_t nr_sects;
+#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
 	seqcount_t nr_sects_seq;
+#endif
 	sector_t alignment_offset;
 	unsigned int discard_alignment;
 	struct device __dev;
@@ -713,6 +715,13 @@ static inline void hd_free_part(struct hd_struct *part)
 	percpu_ref_exit(&part->ref);
 }
 
+static inline void hd_sects_seq_init(struct hd_struct *p)
+{
+#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
+	seqcount_init(&p->nr_sects_seq);
+#endif
+}
+
 /*
  * Any access of part->nr_sects which is not protected by partition
  * bd_mutex or gendisk bdev bd_mutex, should be done using this
-- 
2.20.1

[PATCH 3/4] block: re-organize fields of 'struct hd_part'

[Ming Lei]

Put all fields accessed in IO path together at the beginning
of the struct, so that all can be fetched in single cacheline.

Cc: Yufen Yu <yuyufen@huawei.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hou Tao <houtao1@huawei.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 include/linux/genhd.h | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/include/linux/genhd.h b/include/linux/genhd.h
index 5f5718ce5e86..4c6a998a0c56 100644
--- a/include/linux/genhd.h
+++ b/include/linux/genhd.h
@@ -116,6 +116,14 @@ struct hd_struct {
 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
 	seqcount_t nr_sects_seq;
 #endif
+	unsigned long stamp;
+#ifdef	CONFIG_SMP
+	struct disk_stats __percpu *dkstats;
+#else
+	struct disk_stats dkstats;
+#endif
+	struct percpu_ref ref;
+
 	sector_t alignment_offset;
 	unsigned int discard_alignment;
 	struct device __dev;
@@ -125,13 +133,6 @@ struct hd_struct {
 #ifdef CONFIG_FAIL_MAKE_REQUEST
 	int make_it_fail;
 #endif
-	unsigned long stamp;
-#ifdef	CONFIG_SMP
-	struct disk_stats __percpu *dkstats;
-#else
-	struct disk_stats dkstats;
-#endif
-	struct percpu_ref ref;
 	struct gendisk *disk;
 	struct rcu_work rcu_work;
 };
-- 
2.20.1

[PATCH 4/4] block: don't hold part0's refcount in IO path

[Ming Lei]

gendisk can't be gone when there is IO activity, so not hold
part0's refcount in IO path.

Cc: Yufen Yu <yuyufen@huawei.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hou Tao <houtao1@huawei.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 block/blk-core.c | 3 ++-
 block/genhd.c    | 1 -
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index 79599f5fd5b7..a1184a59819a 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -1345,7 +1345,8 @@ void blk_account_io_done(struct request *req, u64 now)
 		part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns));
 		part_dec_in_flight(req->q, part, rq_data_dir(req));
 
-		hd_struct_put(part);
+		if (part->partno)
+			hd_struct_put(part);
 		part_stat_unlock();
 	}
 }
diff --git a/block/genhd.c b/block/genhd.c
index bfc4148ec341..8c7c32b5dcc0 100644
--- a/block/genhd.c
+++ b/block/genhd.c
@@ -300,7 +300,6 @@ struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
 		}
 	}
  exit:
-	hd_struct_get(&disk->part0);
 	return &disk->part0;
 }
 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
-- 
2.20.1

Re: [PATCH 1/4] block: fix use-after-free on cached last_lookup partition

[Yufen Yu]

Hi Guys,

Did this patch have been forgotten? So, ping...

Thanks,
Yufen


On 2020/1/9 14:21, Ming Lei wrote:
> delete_partition() clears the cached last_lookup partition. However
> the .last_lookup cache may be overwritten by one IO path after
> it is cleared from delete_partition(). Then another IO path may
> use the cached deleting partition after __delete_partition() is
> called, then use-after-free is triggered on the cached partition.
> 
> Fixes the issue by the following approach:
> 
> 1) always get the partition's refcount via hd_struct_try_get() before
> setting .last_lookup
> 
> 2) move clearing .last_lookup from delete_partition() to
> __delete_partition() which is release handle of the partition's
> percpu-refcount, so that no IO path can overwrite .last_lookup after it
> is cleared in __delete_partition().
> 
> It is one candidate approach of Yufen's patch[1] which adds overhead
> in fast path by indirect lookup which may introduce one extra cacheline
> in IO path. Also this patch relies on percpu-refcount's protection, and
> it is easier to understand and verify.
> 
> [1] https://lore.kernel.org/linux-block/20200109013551.GB9655@ming.t460p/T/#t
> 
> Reported-by: Yufen Yu <yuyufen@huawei.com>
> Cc: Christoph Hellwig <hch@infradead.org>
> Cc: Hou Tao <houtao1@huawei.com>
> Signed-off-by: Ming Lei <ming.lei@redhat.com>
> ---
>   block/blk-core.c          | 12 ------------
>   block/genhd.c             |  6 +++++-
>   block/partition-generic.c | 10 +++++++++-
>   include/linux/genhd.h     |  1 +
>   4 files changed, 15 insertions(+), 14 deletions(-)
> 
> diff --git a/block/blk-core.c b/block/blk-core.c
> index 089e890ab208..79599f5fd5b7 100644
> --- a/block/blk-core.c
> +++ b/block/blk-core.c
> @@ -1365,18 +1365,6 @@ void blk_account_io_start(struct request *rq, bool new_io)
>   		part_stat_inc(part, merges[rw]);
>   	} else {
>   		part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
> -		if (!hd_struct_try_get(part)) {
> -			/*
> -			 * The partition is already being removed,
> -			 * the request will be accounted on the disk only
> -			 *
> -			 * We take a reference on disk->part0 although that
> -			 * partition will never be deleted, so we can treat
> -			 * it as any other partition.
> -			 */
> -			part = &rq->rq_disk->part0;
> -			hd_struct_get(part);
> -		}
>   		part_inc_in_flight(rq->q, part, rw);
>   		rq->part = part;
>   	}
> diff --git a/block/genhd.c b/block/genhd.c
> index ff6268970ddc..6029c94510f0 100644
> --- a/block/genhd.c
> +++ b/block/genhd.c
> @@ -286,17 +286,21 @@ struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
>   	ptbl = rcu_dereference(disk->part_tbl);
>   
>   	part = rcu_dereference(ptbl->last_lookup);
> -	if (part && sector_in_part(part, sector))
> +	if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
>   		return part;
>   
>   	for (i = 1; i < ptbl->len; i++) {
>   		part = rcu_dereference(ptbl->part[i]);
>   
>   		if (part && sector_in_part(part, sector)) {
> +			if (!hd_struct_try_get(part))
> +				goto exit;
>   			rcu_assign_pointer(ptbl->last_lookup, part);
>   			return part;
>   		}
>   	}
> + exit:
> +	hd_struct_get(&disk->part0);
>   	return &disk->part0;
>   }
>   EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
> diff --git a/block/partition-generic.c b/block/partition-generic.c
> index 1d20c9cf213f..1739f750dbf2 100644
> --- a/block/partition-generic.c
> +++ b/block/partition-generic.c
> @@ -262,6 +262,12 @@ static void delete_partition_work_fn(struct work_struct *work)
>   void __delete_partition(struct percpu_ref *ref)
>   {
>   	struct hd_struct *part = container_of(ref, struct hd_struct, ref);
> +	struct disk_part_tbl *ptbl =
> +		rcu_dereference_protected(part->disk->part_tbl, 1);
> +
> +	rcu_assign_pointer(ptbl->last_lookup, NULL);
> +	put_device(disk_to_dev(part->disk));
> +
>   	INIT_RCU_WORK(&part->rcu_work, delete_partition_work_fn);
>   	queue_rcu_work(system_wq, &part->rcu_work);
>   }
> @@ -283,8 +289,9 @@ void delete_partition(struct gendisk *disk, int partno)
>   	if (!part)
>   		return;
>   
> +	get_device(disk_to_dev(disk));
>   	rcu_assign_pointer(ptbl->part[partno], NULL);
> -	rcu_assign_pointer(ptbl->last_lookup, NULL);
> +
>   	kobject_put(part->holder_dir);
>   	device_del(part_to_dev(part));
>   
> @@ -349,6 +356,7 @@ struct hd_struct *add_partition(struct gendisk *disk, int partno,
>   	p->nr_sects = len;
>   	p->partno = partno;
>   	p->policy = get_disk_ro(disk);
> +	p->disk = disk;
>   
>   	if (info) {
>   		struct partition_meta_info *pinfo = alloc_part_info(disk);
> diff --git a/include/linux/genhd.h b/include/linux/genhd.h
> index 8bb63027e4d6..1b09cfe00aa3 100644
> --- a/include/linux/genhd.h
> +++ b/include/linux/genhd.h
> @@ -130,6 +130,7 @@ struct hd_struct {
>   	struct disk_stats dkstats;
>   #endif
>   	struct percpu_ref ref;
> +	struct gendisk *disk;
>   	struct rcu_work rcu_work;
>   };
>   
> 

Re: [PATCH 1/4] block: fix use-after-free on cached last_lookup partition

[Bart Van Assche]

On 2020-01-08 22:21, Ming Lei wrote:
> delete_partition() clears the cached last_lookup partition. However
> the .last_lookup cache may be overwritten by one IO path after
> it is cleared from delete_partition(). Then another IO path may
> use the cached deleting partition after __delete_partition() is
> called, then use-after-free is triggered on the cached partition.
> 
> Fixes the issue by the following approach:
> 
> 1) always get the partition's refcount via hd_struct_try_get() before
> setting .last_lookup
> 
> 2) move clearing .last_lookup from delete_partition() to
> __delete_partition() which is release handle of the partition's
> percpu-refcount, so that no IO path can overwrite .last_lookup after it
> is cleared in __delete_partition().
> 
> It is one candidate approach of Yufen's patch[1] which adds overhead
> in fast path by indirect lookup which may introduce one extra cacheline
> in IO path. Also this patch relies on percpu-refcount's protection, and
> it is easier to understand and verify.
> 
> [1] https://lore.kernel.org/linux-block/20200109013551.GB9655@ming.t460p/T/#t

Hi Ming,

disk_map_sector_rcu() is called from the I/O path only and hence with
q->q_usage_counter > 0. Has it been considered to freeze disk->queue
from delete_partition() before deleting a partition and unfreezing
disk->queue after partition deletion has finished? Would that approach
allow to eliminate partition reference counting and thereby improve the
performance of the hot path?

Thanks,

Bart.

Re: [PATCH 1/4] block: fix use-after-free on cached last_lookup partition

[Ming Lei]

On Thu, Feb 20, 2020 at 08:03:52PM -0800, Bart Van Assche wrote:
> On 2020-01-08 22:21, Ming Lei wrote:
> > delete_partition() clears the cached last_lookup partition. However
> > the .last_lookup cache may be overwritten by one IO path after
> > it is cleared from delete_partition(). Then another IO path may
> > use the cached deleting partition after __delete_partition() is
> > called, then use-after-free is triggered on the cached partition.
> > 
> > Fixes the issue by the following approach:
> > 
> > 1) always get the partition's refcount via hd_struct_try_get() before
> > setting .last_lookup
> > 
> > 2) move clearing .last_lookup from delete_partition() to
> > __delete_partition() which is release handle of the partition's
> > percpu-refcount, so that no IO path can overwrite .last_lookup after it
> > is cleared in __delete_partition().
> > 
> > It is one candidate approach of Yufen's patch[1] which adds overhead
> > in fast path by indirect lookup which may introduce one extra cacheline
> > in IO path. Also this patch relies on percpu-refcount's protection, and
> > it is easier to understand and verify.
> > 
> > [1] https://lore.kernel.org/linux-block/20200109013551.GB9655@ming.t460p/T/#t
> 
> Hi Ming,
> 
> disk_map_sector_rcu() is called from the I/O path only and hence with
> q->q_usage_counter > 0. Has it been considered to freeze disk->queue
> from delete_partition() before deleting a partition and unfreezing
> disk->queue after partition deletion has finished? Would that approach
> allow to eliminate partition reference counting and thereby improve the
> performance of the hot path?

Hi Bart,

I did consider that approach, but this way may cause performance
regression, given deleting any partition drops IO performance a lot
on other un-related partitions.



Thanks,
Ming

Re: [PATCH 3/4] block: re-organize fields of 'struct hd_part'

[Christoph Hellwig]

On Thu, May 07, 2020 at 04:52:38PM +0800, Ming Lei wrote:
> Put all fields accessed in IO path together at the beginning
> of the struct, so that all can be fetched in single cacheline.

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>

[PATCH 3/4] block: re-organize fields of 'struct hd_part'

[Ming Lei]

Put all fields accessed in IO path together at the beginning
of the struct, so that all can be fetched in single cacheline.

Cc: Yufen Yu <yuyufen@huawei.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hou Tao <houtao1@huawei.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 include/linux/genhd.h | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/include/linux/genhd.h b/include/linux/genhd.h
index 2732120751e8..e0fab519ce37 100644
--- a/include/linux/genhd.h
+++ b/include/linux/genhd.h
@@ -71,6 +71,14 @@ struct hd_struct {
 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
 	seqcount_t nr_sects_seq;
 #endif
+	unsigned long stamp;
+#ifdef	CONFIG_SMP
+	struct disk_stats __percpu *dkstats;
+#else
+	struct disk_stats dkstats;
+#endif
+	struct percpu_ref ref;
+
 	sector_t alignment_offset;
 	unsigned int discard_alignment;
 	struct device __dev;
@@ -80,13 +88,6 @@ struct hd_struct {
 #ifdef CONFIG_FAIL_MAKE_REQUEST
 	int make_it_fail;
 #endif
-	unsigned long stamp;
-#ifdef	CONFIG_SMP
-	struct disk_stats __percpu *dkstats;
-#else
-	struct disk_stats dkstats;
-#endif
-	struct percpu_ref ref;
 	struct gendisk *disk;
 	struct rcu_work rcu_work;
 };
-- 
2.25.2