Commit 187c4944 authored Nov 15, 2019 by Alistair Delva

ANDROID: virtio_blk: Remove BUG_ON for discard/zero ops



The virtio_blk driver used by cuttlefish can crash if discard/zero ops,
added in 4.20, pass through virtio_queue_rq(). Our theory is that this
happens because multiple discards are merged, causing nr_phys_segments
to be increased beyond the sg_elem size. This makes sense as a
discard/zero is a form of I/O not backed by specific data.

This patch should be submitted upstream once the true root cause has
been tracked down and full analysis of the issue has been done.

Originally-by: Satya Tangirala <satyat@google.com>
Bug: 144592287
Test: boot cuttlefish with f2fs userdata and wait 1 minute
Change-Id: Ic73e08035c91ec9c718e6496930b326d084062a4
Signed-off-by: Alistair Delva <adelva@google.com>

parent 85261924

drivers/block/virtio_blk.c

+4 −2

Original line number	Diff line number	Diff line
		@@ -282,8 +282,6 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
		bool unmap = false;
		u32 type;

		BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);

		switch (req_op(req)) {
		case REQ_OP_READ:
		case REQ_OP_WRITE:
		@@ -311,6 +309,10 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
		return BLK_STS_IOERR;
		}

		BUG_ON(type != VIRTIO_BLK_T_DISCARD &&
		type != VIRTIO_BLK_T_WRITE_ZEROES &&
		(req->nr_phys_segments + 2 > vblk->sg_elems));

		vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
		vbr->out_hdr.sector = type ?
		0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));