Message ID | 20230119073647.76467-1-jasowang@redhat.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [V2] vhost-scsi: unbreak any layout for response | expand |
On Thu, Jan 19, 2023 at 03:36:47PM +0800, Jason Wang wrote: > Al Viro said: > > """ > Since "vhost/scsi: fix reuse of &vq->iov[out] in response" > we have this: > cmd->tvc_resp_iov = vq->iov[vc.out]; > cmd->tvc_in_iovs = vc.in; > combined with > iov_iter_init(&iov_iter, ITER_DEST, &cmd->tvc_resp_iov, > cmd->tvc_in_iovs, sizeof(v_rsp)); > in vhost_scsi_complete_cmd_work(). We used to have ->tvc_resp_iov > _pointing_ to vq->iov[vc.out]; back then iov_iter_init() asked to > set an iovec-backed iov_iter over the tail of vq->iov[], with > length being the amount of iovecs in the tail. > > Now we have a copy of one element of that array. Fortunately, the members > following it in the containing structure are two non-NULL kernel pointers, > so copy_to_iter() will not copy anything beyond the first iovec - kernel > pointer is not (on the majority of architectures) going to be accepted by > access_ok() in copyout() and it won't be skipped since the "length" (in > reality - another non-NULL kernel pointer) won't be zero. > > So it's not going to give a guest-to-qemu escalation, but it's definitely > a bug. Frankly, my preference would be to verify that the very first iovec > is long enough to hold rsp_size. Due to the above, any users that try to > give us vq->iov[vc.out].iov_len < sizeof(struct virtio_scsi_cmd_resp) > would currently get a failure in vhost_scsi_complete_cmd_work() > anyway. > """ > > However, the spec doesn't say anything about the legacy descriptor > layout for the respone. So this patch tries to not assume the response > to reside in a single separate descriptor which is what commit > 79c14141a487 ("vhost/scsi: Convert completion path to use") tries to > achieve towards to ANY_LAYOUT. > > This is done by allocating and using dedicate resp iov in the > command. To be safety, start with UIO_MAXIOV to be consistent with the > limitation that we advertise to the vhost_get_vq_desc(). > > Testing with the hacked virtio-scsi driver that use 1 descriptor for 1 > byte in the response. > > Reported-by: Al Viro <viro@zeniv.linux.org.uk> > Cc: Benjamin Coddington <bcodding@redhat.com> > Cc: Nicholas Bellinger <nab@linux-iscsi.org> > Fixes: a77ec83a5789 ("vhost/scsi: fix reuse of &vq->iov[out] in response") > Signed-off-by: Jason Wang <jasowang@redhat.com> > --- > Changes since V1: > - tweak the changelog > - fix the allocation size for tvc_resp_iov (should be sizeof(struct iovec)) > --- > drivers/vhost/scsi.c | 21 +++++++++++++++++---- > 1 file changed, 17 insertions(+), 4 deletions(-) Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
diff --git a/drivers/vhost/scsi.c b/drivers/vhost/scsi.c index dca6346d75b3..d5ecb8876fc9 100644 --- a/drivers/vhost/scsi.c +++ b/drivers/vhost/scsi.c @@ -80,7 +80,7 @@ struct vhost_scsi_cmd { struct scatterlist *tvc_prot_sgl; struct page **tvc_upages; /* Pointer to response header iovec */ - struct iovec tvc_resp_iov; + struct iovec *tvc_resp_iov; /* Pointer to vhost_scsi for our device */ struct vhost_scsi *tvc_vhost; /* Pointer to vhost_virtqueue for the cmd */ @@ -563,7 +563,7 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) memcpy(v_rsp.sense, cmd->tvc_sense_buf, se_cmd->scsi_sense_length); - iov_iter_init(&iov_iter, ITER_DEST, &cmd->tvc_resp_iov, + iov_iter_init(&iov_iter, ITER_DEST, cmd->tvc_resp_iov, cmd->tvc_in_iovs, sizeof(v_rsp)); ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter); if (likely(ret == sizeof(v_rsp))) { @@ -594,6 +594,7 @@ vhost_scsi_get_cmd(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg, struct vhost_scsi_cmd *cmd; struct vhost_scsi_nexus *tv_nexus; struct scatterlist *sg, *prot_sg; + struct iovec *tvc_resp_iov; struct page **pages; int tag; @@ -613,6 +614,7 @@ vhost_scsi_get_cmd(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg, sg = cmd->tvc_sgl; prot_sg = cmd->tvc_prot_sgl; pages = cmd->tvc_upages; + tvc_resp_iov = cmd->tvc_resp_iov; memset(cmd, 0, sizeof(*cmd)); cmd->tvc_sgl = sg; cmd->tvc_prot_sgl = prot_sg; @@ -625,6 +627,7 @@ vhost_scsi_get_cmd(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg, cmd->tvc_data_direction = data_direction; cmd->tvc_nexus = tv_nexus; cmd->inflight = vhost_scsi_get_inflight(vq); + cmd->tvc_resp_iov = tvc_resp_iov; memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE); @@ -935,7 +938,7 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) struct iov_iter in_iter, prot_iter, data_iter; u64 tag; u32 exp_data_len, data_direction; - int ret, prot_bytes, c = 0; + int ret, prot_bytes, i, c = 0; u16 lun; u8 task_attr; bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI); @@ -1092,7 +1095,8 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) } cmd->tvc_vhost = vs; cmd->tvc_vq = vq; - cmd->tvc_resp_iov = vq->iov[vc.out]; + for (i = 0; i < vc.in ; i++) + cmd->tvc_resp_iov[i] = vq->iov[vc.out + i]; cmd->tvc_in_iovs = vc.in; pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n", @@ -1461,6 +1465,7 @@ static void vhost_scsi_destroy_vq_cmds(struct vhost_virtqueue *vq) kfree(tv_cmd->tvc_sgl); kfree(tv_cmd->tvc_prot_sgl); kfree(tv_cmd->tvc_upages); + kfree(tv_cmd->tvc_resp_iov); } sbitmap_free(&svq->scsi_tags); @@ -1508,6 +1513,14 @@ static int vhost_scsi_setup_vq_cmds(struct vhost_virtqueue *vq, int max_cmds) goto out; } + tv_cmd->tvc_resp_iov = kcalloc(UIO_MAXIOV, + sizeof(struct iovec), + GFP_KERNEL); + if (!tv_cmd->tvc_resp_iov) { + pr_err("Unable to allocate tv_cmd->tvc_resp_iov\n"); + goto out; + } + tv_cmd->tvc_prot_sgl = kcalloc(VHOST_SCSI_PREALLOC_PROT_SGLS, sizeof(struct scatterlist), GFP_KERNEL);
Al Viro said: """ Since "vhost/scsi: fix reuse of &vq->iov[out] in response" we have this: cmd->tvc_resp_iov = vq->iov[vc.out]; cmd->tvc_in_iovs = vc.in; combined with iov_iter_init(&iov_iter, ITER_DEST, &cmd->tvc_resp_iov, cmd->tvc_in_iovs, sizeof(v_rsp)); in vhost_scsi_complete_cmd_work(). We used to have ->tvc_resp_iov _pointing_ to vq->iov[vc.out]; back then iov_iter_init() asked to set an iovec-backed iov_iter over the tail of vq->iov[], with length being the amount of iovecs in the tail. Now we have a copy of one element of that array. Fortunately, the members following it in the containing structure are two non-NULL kernel pointers, so copy_to_iter() will not copy anything beyond the first iovec - kernel pointer is not (on the majority of architectures) going to be accepted by access_ok() in copyout() and it won't be skipped since the "length" (in reality - another non-NULL kernel pointer) won't be zero. So it's not going to give a guest-to-qemu escalation, but it's definitely a bug. Frankly, my preference would be to verify that the very first iovec is long enough to hold rsp_size. Due to the above, any users that try to give us vq->iov[vc.out].iov_len < sizeof(struct virtio_scsi_cmd_resp) would currently get a failure in vhost_scsi_complete_cmd_work() anyway. """ However, the spec doesn't say anything about the legacy descriptor layout for the respone. So this patch tries to not assume the response to reside in a single separate descriptor which is what commit 79c14141a487 ("vhost/scsi: Convert completion path to use") tries to achieve towards to ANY_LAYOUT. This is done by allocating and using dedicate resp iov in the command. To be safety, start with UIO_MAXIOV to be consistent with the limitation that we advertise to the vhost_get_vq_desc(). Testing with the hacked virtio-scsi driver that use 1 descriptor for 1 byte in the response. Reported-by: Al Viro <viro@zeniv.linux.org.uk> Cc: Benjamin Coddington <bcodding@redhat.com> Cc: Nicholas Bellinger <nab@linux-iscsi.org> Fixes: a77ec83a5789 ("vhost/scsi: fix reuse of &vq->iov[out] in response") Signed-off-by: Jason Wang <jasowang@redhat.com> --- Changes since V1: - tweak the changelog - fix the allocation size for tvc_resp_iov (should be sizeof(struct iovec)) --- drivers/vhost/scsi.c | 21 +++++++++++++++++---- 1 file changed, 17 insertions(+), 4 deletions(-)