[PATCH v2 0/2] Fix a race condition triggered by submit_bio()

[PATCH v2 0/2] Fix a race condition triggered by submit_bio()

[Bart Van Assche]

Hi Jens,

One of the consequences of the switch to blk-mq is that generic_make_request()
calls, a function called by submit_bio(), must be protected by a
blk_queue_enter() / blk_queue_exit() pair to avoid that the block-cgroup
functions called by generic_make_request() trigger a race condition. This patch
series makes the kernel report a warning if that race condition is hit and also
adds the necessary protection in submit_bio(). Please consider these patches
for kernel v5.4.

Thanks,

Bart.

Changes compared to v1:
- Use the full syzbot ID in patch 2/2 instead of abbreviating it.

Bart Van Assche (2):
  block: Verify whether blk_queue_enter() is used when necessary
  block: Fix a race condition in submit_bio()

 block/blk-cgroup.c         |  2 ++
 block/blk-core.c           | 34 +++++++++++++++++++++++++++++++++-
 include/linux/blk-cgroup.h |  2 ++
 include/linux/blkdev.h     |  8 ++++++++
 4 files changed, 45 insertions(+), 1 deletion(-)

-- 
2.22.0.709.g102302147b-goog

[PATCH v2 1/2] block: Verify whether blk_queue_enter() is used when necessary

[Bart Van Assche]

It is required to protect blkg_lookup() calls with a blk_queue_enter() /
blk_queue_exit() pair. Since it is nontrivial to verify whether this is
the case, verify this at runtime. Only perform this verification if
CONFIG_LOCKDEP=y to avoid that unnecessary runtime overhead is added.

Note: using lock_acquire()/lock_release() to verify whether blkg_lookup()
is protected correctly is not possible since lock_acquire() and
lock_release() must be called from the same task and since
blk_queue_enter() and blk_queue_exit() can be called from different
tasks.

Suggested-by: Tejun Heo <tj@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: Hannes Reinecke <hare@suse.com>
Cc: Johannes Thumshirn <jthumshirn@suse.de>
Cc: Alexandru Moise <00moses.alexander00@gmail.com>
Cc: Joseph Qi <joseph.qi@linux.alibaba.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
---
 block/blk-cgroup.c         |  2 ++
 block/blk-core.c           | 21 +++++++++++++++++++++
 include/linux/blk-cgroup.h |  2 ++
 include/linux/blkdev.h     |  8 ++++++++
 4 files changed, 33 insertions(+)

diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
index 24ed26957367..04b6e962eefb 100644
--- a/block/blk-cgroup.c
+++ b/block/blk-cgroup.c
@@ -196,6 +196,8 @@ struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
 {
 	struct blkcg_gq *blkg;
 
+	WARN_ON_ONCE(!blk_entered_queue(q));
+
 	/*
 	 * Hint didn't match.  Look up from the radix tree.  Note that the
 	 * hint can only be updated under queue_lock as otherwise @blkg
diff --git a/block/blk-core.c b/block/blk-core.c
index 5878504a29af..ff27c3080348 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -389,6 +389,25 @@ struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
 }
 EXPORT_SYMBOL(blk_alloc_queue);
 
+#ifdef CONFIG_PROVE_LOCKING
+/**
+ * blk_entered_queue() - whether or not it is safe to access cgroup information
+ * @q: request queue pointer
+ *
+ * In order to avoid races between accessing cgroup information and the cgroup
+ * information removal from inside __blk_release_queue(), any code that accesses
+ * cgroup information must be protected by a blk_queue_enter()/blk_queue_exit()
+ * pair or must be called after queue cleanup progressed to a stage in which
+ * only the cleanup code accesses the queue.
+ */
+bool blk_entered_queue(struct request_queue *q)
+{
+	return percpu_ref_is_dying(&q->q_usage_counter) ||
+		!percpu_ref_is_zero(&q->q_usage_counter);
+}
+EXPORT_SYMBOL(blk_entered_queue);
+#endif
+
 /**
  * blk_queue_enter() - try to increase q->q_usage_counter
  * @q: request queue pointer
@@ -878,6 +897,8 @@ generic_make_request_checks(struct bio *bio)
 		goto end_io;
 	}
 
+	WARN_ON_ONCE(!blk_entered_queue(q));
+
 	/*
 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
 	 * if queue is not a request based queue.
diff --git a/include/linux/blk-cgroup.h b/include/linux/blk-cgroup.h
index 689a58231288..397df0719bda 100644
--- a/include/linux/blk-cgroup.h
+++ b/include/linux/blk-cgroup.h
@@ -358,6 +358,8 @@ static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
 {
 	struct blkcg_gq *blkg;
 
+	WARN_ON_ONCE(!blk_entered_queue(q));
+
 	if (blkcg == &blkcg_root)
 		return q->root_blkg;
 
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 96a29a72fd4a..e57651888450 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -856,6 +856,14 @@ extern int sg_scsi_ioctl(struct request_queue *, struct gendisk *, fmode_t,
 
 extern int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags);
 extern void blk_queue_exit(struct request_queue *q);
+#ifdef CONFIG_PROVE_LOCKING
+extern bool blk_entered_queue(struct request_queue *q);
+#else
+static inline bool blk_entered_queue(struct request_queue *q)
+{
+	return true;
+}
+#endif
 extern void blk_sync_queue(struct request_queue *q);
 extern int blk_rq_map_user(struct request_queue *, struct request *,
 			   struct rq_map_data *, void __user *, unsigned long,
-- 
2.22.0.709.g102302147b-goog

[PATCH v2 2/2] block: Fix a race condition in submit_bio()

[Bart Van Assche]

generic_make_request_checks() needs to be protected by a
blk_queue_enter() / blk_queue_exit() pair because it calls
blkcg_bio_issue_check() and because that last function calls
blkg_lookup().

This patch fixes https://syzkaller.appspot.com/bug?id=ff9ab4a23afa7553fb79f745a92be87ba4144508.

This patch also fixes the following kernel warning, triggered by
blktests:

WARNING: CPU: 5 PID: 10706 at block/blk-core.c:903 generic_make_request_checks+0x9c6/0xe60
RIP: 0010:generic_make_request_checks+0x9c6/0xe60
Call Trace:
 generic_make_request+0x7a/0x5c0
 submit_bio+0x92/0x280
 mpage_readpages+0x2b1/0x300
 blkdev_readpages+0x1d/0x20
 read_pages+0xd9/0x2c0
 __do_page_cache_readahead+0x2e0/0x310
 force_page_cache_readahead+0xfb/0x170
 page_cache_sync_readahead+0x28d/0x2a0
 generic_file_read_iter+0xc13/0x1530
 blkdev_read_iter+0x7d/0x90
 new_sync_read+0x2c5/0x3d0
 __vfs_read+0x7b/0x90
 vfs_read+0xc6/0x1f0
 ksys_read+0xc3/0x160
 __x64_sys_read+0x43/0x50
 do_syscall_64+0x71/0x270
 entry_SYSCALL_64_after_hwframe+0x49/0xbe

Cc: Tejun Heo <tj@kernel.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: Hannes Reinecke <hare@suse.com>
Cc: Johannes Thumshirn <jthumshirn@suse.de>
Cc: Alexandru Moise <00moses.alexander00@gmail.com>
Cc: Joseph Qi <joseph.qi@linux.alibaba.com>
Reported-by: syzbot+21cfe1f803e0e158acf1@syzkaller.appspotmail.com
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
---
 block/blk-core.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index ff27c3080348..cd844c54e9f1 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -1150,6 +1150,9 @@ EXPORT_SYMBOL_GPL(direct_make_request);
  */
 blk_qc_t submit_bio(struct bio *bio)
 {
+	struct request_queue *q = bio->bi_disk->queue;
+	blk_qc_t ret;
+
 	if (blkcg_punt_bio_submit(bio))
 		return BLK_QC_T_NONE;
 
@@ -1182,7 +1185,15 @@ blk_qc_t submit_bio(struct bio *bio)
 		}
 	}
 
-	return generic_make_request(bio);
+	if (unlikely(blk_queue_enter(q, 0) < 0)) {
+		bio->bi_status = BLK_STS_IOERR;
+		bio->bi_end_io(bio);
+		return BLK_QC_T_NONE;
+	}
+	ret = generic_make_request(bio);
+	blk_queue_exit(q);
+
+	return ret;
 }
 EXPORT_SYMBOL(submit_bio);
 
-- 
2.22.0.709.g102302147b-goog