by Michael Kelley (LINUX)

[permalink] [raw]

Subject: RE: [PATCH 2/6] PCI: hv: Use vmbus_requestor to generate transaction IDs for VMbus hardening

From: Andrea Parri (Microsoft) <[email protected]> Sent: Wednesday, April 6, 2022 9:30 PM
>
> Currently, pointers to guest memory are passed to Hyper-V as transaction
> IDs in hv_pci. In the face of errors or malicious behavior in Hyper-V,
> hv_pci should not expose or trust the transaction IDs returned by
> Hyper-V to be valid guest memory addresses. Instead, use small integers
> generated by vmbus_requestor as request (transaction) IDs.
>
> Suggested-by: Michael Kelley <[email protected]>
> Signed-off-by: Andrea Parri (Microsoft) <[email protected]>
> ---
> drivers/pci/controller/pci-hyperv.c | 39 +++++++++++++++++++++--------
> 1 file changed, 29 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
> index 88b3b56d05228..c1322ac37cda9 100644
> --- a/drivers/pci/controller/pci-hyperv.c
> +++ b/drivers/pci/controller/pci-hyperv.c
> @@ -91,6 +91,13 @@ static enum pci_protocol_version_t pci_protocol_versions[] = {
> /* space for 32bit serial number as string */
> #define SLOT_NAME_SIZE 11
>
> +/*
> + * Size of requestor for VMbus; the value is based on the observation
> + * that having more than one request outstanding is 'rare', and so 64
> + * should be generous in ensuring that we don't ever run out.
> + */
> +#define HV_PCI_RQSTOR_SIZE 64
> +
> /*
> * Message Types
> */
> @@ -1407,7 +1414,7 @@ static void hv_int_desc_free(struct hv_pci_dev *hpdev,
> int_pkt->wslot.slot = hpdev->desc.win_slot.slot;
> int_pkt->int_desc = *int_desc;
> vmbus_sendpacket(hpdev->hbus->hdev->channel, int_pkt, sizeof(*int_pkt),
> - (unsigned long)&ctxt.pkt, VM_PKT_DATA_INBAND, 0);
> + 0, VM_PKT_DATA_INBAND, 0);
> kfree(int_desc);
> }
>
> @@ -2649,7 +2656,7 @@ static void hv_eject_device_work(struct work_struct *work)
> ejct_pkt->message_type.type = PCI_EJECTION_COMPLETE;
> ejct_pkt->wslot.slot = hpdev->desc.win_slot.slot;
> vmbus_sendpacket(hbus->hdev->channel, ejct_pkt,
> - sizeof(*ejct_pkt), (unsigned long)&ctxt.pkt,
> + sizeof(*ejct_pkt), 0,
> VM_PKT_DATA_INBAND, 0);
>
> /* For the get_pcichild() in hv_pci_eject_device() */
> @@ -2696,8 +2703,9 @@ static void hv_pci_onchannelcallback(void *context)
> const int packet_size = 0x100;
> int ret;
> struct hv_pcibus_device *hbus = context;
> + struct vmbus_channel *chan = hbus->hdev->channel;
> u32 bytes_recvd;
> - u64 req_id;
> + u64 req_id, req_addr;
> struct vmpacket_descriptor *desc;
> unsigned char *buffer;
> int bufferlen = packet_size;
> @@ -2715,8 +2723,8 @@ static void hv_pci_onchannelcallback(void *context)
> return;
>
> while (1) {
> - ret = vmbus_recvpacket_raw(hbus->hdev->channel, buffer,
> - bufferlen, &bytes_recvd, &req_id);
> + ret = vmbus_recvpacket_raw(chan, buffer, bufferlen,
> + &bytes_recvd, &req_id);
>
> if (ret == -ENOBUFS) {
> kfree(buffer);
> @@ -2743,11 +2751,14 @@ static void hv_pci_onchannelcallback(void *context)
> switch (desc->type) {
> case VM_PKT_COMP:
>
> - /*
> - * The host is trusted, and thus it's safe to interpret
> - * this transaction ID as a pointer.
> - */
> - comp_packet = (struct pci_packet *)req_id;
> + req_addr = chan->request_addr_callback(chan, req_id);
> + if (req_addr == VMBUS_RQST_ERROR) {
> + dev_warn_ratelimited(&hbus->hdev->device,
> + "Invalid transaction ID %llx\n",
> + req_id);

This handling of a bad requestID error is a bit different from storvsc
and netvsc. They both use dev_err(). Earlier in the storvsc and netvsc
cases, I remember some discussion about whether to rate limit these errors,
and evidently we decided not to. I think we should be consistent unless
there's a specific reason not to.

> + break;
> + }
> + comp_packet = (struct pci_packet *)req_addr;
> response = (struct pci_response *)buffer;
> comp_packet->completion_func(comp_packet->compl_ctxt,
> response,
> @@ -3428,6 +3439,10 @@ static int hv_pci_probe(struct hv_device *hdev,
> goto free_dom;
> }
>
> + hdev->channel->next_request_id_callback = vmbus_next_request_id;
> + hdev->channel->request_addr_callback = vmbus_request_addr;
> + hdev->channel->rqstor_size = HV_PCI_RQSTOR_SIZE;
> +
> ret = vmbus_open(hdev->channel, pci_ring_size, pci_ring_size, NULL, 0,
> hv_pci_onchannelcallback, hbus);
> if (ret)
> @@ -3758,6 +3773,10 @@ static int hv_pci_resume(struct hv_device *hdev)
>
> hbus->state = hv_pcibus_init;
>
> + hdev->channel->next_request_id_callback = vmbus_next_request_id;
> + hdev->channel->request_addr_callback = vmbus_request_addr;
> + hdev->channel->rqstor_size = HV_PCI_RQSTOR_SIZE;
> +
> ret = vmbus_open(hdev->channel, pci_ring_size, pci_ring_size, NULL, 0,
> hv_pci_onchannelcallback, hbus);
> if (ret)
> --
> 2.25.1

2022-04-12 00:52:17

by Michael Kelley (LINUX)

[permalink] [raw]

Subject: RE: [PATCH 4/6] Drivers: hv: vmbus: Introduce vmbus_request_addr_match()

From: Andrea Parri <[email protected]> Sent: Friday, April 8, 2022 1:38 PM
>
> > > > In the case where a specific match is required, and trans_id is
> > > > valid but the addr's do not match, it looks like this function will
> > > > return the addr that didn't match, without removing the entry.
> > >
> > > Yes, that is consistent with the description on vmbus_request_addr_match():
> > >
> > > Returns the memory address stored at @trans_id, or VMBUS_RQST_ERROR if
> > > @trans_id is not contained in the requestor.
> > >
> > >
> > > > Shouldn't it return VMBUS_RQST_ERROR in that case?
> > >
> > > Can certainly be done, although I'm not sure to follow your concerns. Can
> > > you elaborate?
> > >
> >
> > Having the function return "success" when it failed to match is unexpected
> > for me. There's only one invocation where we care about matching
> > (in hv_compose_msi_msg). In that invocation the purpose for matching is to
> > not remove the wrong entry, and the return value is ignored. So I think
> > it all works correctly.
>
> You're reading it wrongly: the point is that there's nothing wrong in *not
> removing the "wrong entry" (or in failing to match). In the mentioned use,
> that means the channel callback has already processed "our" request, and
> that we don't have to worry about the ID. (Otherwise, i.e. if we do match,
> the callback will eventually scream "Invalid transaction".)
>
>
> > Just thinking out loud, maybe vmbus_request_addr_match() should be
> > renamed to vmbus_request_addr_remove(), and not have a return value?
>
> Mmh. We have vmbus_request_addr() that (always) removes the ID; it seems
> a _remove() would just add to the confusion. And removing the return value
> would mean duplicating most of vmbus_request_addr() in the "new" function.
> So, I'm not convinced that's the right thing to do. I'm inclined to leave
> this patch as is (and, as usual, happy to be proven wrong).
>

I'll defer to your judgment. I don't see anything that broken with the
patch as written, so I can live with it as is.

Michael