Re: [PATCH] nvme: fix use-after-free during booting

[Tong Zhang <ztong0001@gmail.com>]

Hi Christoph,
I modified the patch a bit and now it works.
Best,
- Tong

---
 block/genhd.c         | 7 +++++--
 include/linux/genhd.h | 1 +
 2 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/block/genhd.c b/block/genhd.c
index 99c64641c314..8c432e5f97ea 100644
--- a/block/genhd.c
+++ b/block/genhd.c
@@ -835,7 +835,10 @@ static void __device_add_disk(struct device *parent, struct gendisk *disk,
 	 * Take an extra ref on queue which will be put on disk_release()
 	 * so that it sticks around as long as @disk is there.
 	 */
-	WARN_ON_ONCE(!blk_get_queue(disk->queue));
+	if (blk_get_queue(disk->queue))
+		disk->flags |= GENHD_FL_QUEUE_REF;
+	else
+		WARN_ON_ONCE(1);
 
 	disk_add_events(disk);
 	blk_integrity_add(disk);
@@ -1567,7 +1570,7 @@ static void disk_release(struct device *dev)
 	kfree(disk->random);
 	disk_replace_part_tbl(disk, NULL);
 	hd_free_part(&disk->part0);
-	if (disk->queue)
+	if (disk->flags & GENHD_FL_QUEUE_REF)
 		blk_put_queue(disk->queue);
 	kfree(disk);
 }
diff --git a/include/linux/genhd.h b/include/linux/genhd.h
index 4ab853461dff..9441077ee103 100644
--- a/include/linux/genhd.h
+++ b/include/linux/genhd.h
@@ -135,6 +135,7 @@ struct hd_struct {
 #define GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE	0x0100
 #define GENHD_FL_NO_PART_SCAN			0x0200
 #define GENHD_FL_HIDDEN				0x0400
+#define GENHD_FL_QUEUE_REF			0x0800
 
 enum {
 	DISK_EVENT_MEDIA_CHANGE			= 1 << 0, /* media changed */
-- 
2.25.1


> On Sep 22, 2020, at 12:41 PM, Christoph Hellwig <hch@lst.de> wrote:
> 
> On Tue, Sep 22, 2020 at 12:19:06PM -0400, Tong Zhang wrote:
>> Thank you Christoph for providing the patch.
>> However, the test shows that the issue still persists even with the new patch.
>> The call trace is the same as in my first email.
> 
> Weird.  What baseline did you test?  You need to use the patch on top of
> the latest for-next or for-5.10/block tree from Jens for full effect.
> 
> If that still feels can you send me a longer dmesg that also shows what
> happens before the sniplet you sent?