Commit 1b2b12376c ("intel-iommu: PASID support") takes PASID into
account when calculating iotlb hash like:
static guint vtd_iotlb_hash(gconstpointer v)
{
const struct vtd_iotlb_key *key = v;
return key->gfn | ((key->sid) << VTD_IOTLB_SID_SHIFT) |
(key->level) << VTD_IOTLB_LVL_SHIFT |
(key->pasid) << VTD_IOTLB_PASID_SHIFT;
}
This turns out to be problematic since:
- the shift will lose bits if not converting to uint64_t
- level should be off by one in order to fit into 2 bits
- VTD_IOTLB_PASID_SHIFT is 30 but PASID is 20 bits which will waste
some bits
- the hash result is uint64_t so we will lose bits when converting to
guint
So this patch fixes them by
- converting the keys into uint64_t before doing the shift
- off level by one to make it fit into two bits
- change the sid, lvl and pasid shift to 26, 42 and 44 in order to
take the full width of uint64_t
- perform an XOR to the top 32bit with the bottom 32bit for the final
result to fit guint
Fixes: Coverity CID 1508100
Fixes: 1b2b12376c ("intel-iommu: PASID support")
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20230412073510.7158-1-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
When vIOMMU is enabled, the vq->used_phys is actually the IOVA not
GPA. So we need to translate it to GPA before the syncing otherwise we
may hit the following crash since IOVA could be out of the scope of
the GPA log size. This could be noted when using virtio-IOMMU with
vhost using 1G memory.
Fixes: c471ad0e9b ("vhost_net: device IOTLB support")
Cc: qemu-stable@nongnu.org
Tested-by: Lei Yang <leiyang@redhat.com>
Reported-by: Yalan Zhang <yalzhang@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20221216033552.77087-1-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch introduce ECAP_PASID via "x-pasid-mode". Based on the
existing support for scalable mode, we need to implement the following
missing parts:
1) tag VTDAddressSpace with PASID and support IOMMU/DMA translation
with PASID
2) tag IOTLB with PASID
3) PASID cache and its flush
4) PASID based IOTLB invalidation
For simplicity PASID cache is not implemented so we can simply
implement the PASID cache flush as a no and leave it to be implemented
in the future. For PASID based IOTLB invalidation, since we haven't
had L1 stage support, the PASID based IOTLB invalidation is not
implemented yet. For PASID based device IOTLB invalidation, it
requires the support for vhost so we forbid enabling device IOTLB when
PASID is enabled now. Those work could be done in the future.
Note that though PASID based IOMMU translation is ready but no device
can issue PASID DMA right now. In this case, PCI_NO_PASID is used as
PASID to identify the address without PASID. vtd_find_add_as() has
been extended to provision address space with PASID which could be
utilized by the future extension of PCI core to allow device model to
use PASID based DMA translation.
This feature would be useful for:
1) prototyping PASID support for devices like virtio
2) future vPASID work
3) future PRS and vSVA work
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20221028061436.30093-5-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We used to have a macro for VTD_PE_GET_FPD_ERR() but it has an
internal goto which prevents it from being reused. This patch convert
that macro to a dedicated function and let the caller to decide what
to do (e.g using goto or not). This makes sure it can be re-used for
other function that requires fault reporting.
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20221028061436.30093-4-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Yi Liu <yi.l.liu@intel.com>
We introduce VTDBus structure as an intermediate step for searching
the address space. This works well with SID based matching/lookup. But
when we want to support SID plus PASID based address space lookup,
this intermediate steps turns out to be a burden. So the patch simply
drops the VTDBus structure and use the PCIBus and devfn as the key for
the g_hash_table(). This simplifies the codes and the future PASID
extension.
To prevent being slower for past vtd_find_as_from_bus_num() callers, a
vtd_as cache indexed by the bus number is introduced to store the last
recent search result of a vtd_as belongs to a specific bus.
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20221028061436.30093-3-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Yi Liu <yi.l.liu@intel.com>
We need check whether passthrough is enabled during
vtd_switch_address_space() by checking the context entries. This
requires the root_scalable to be set correctly otherwise we may try to
check legacy rsvd bits instead of scalable ones.
Fixing this by updating root_scalable before switching the address
spaces during post_load.
Fixes: fb43cf739e ("intel_iommu: scalable mode emulation")
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220317080522.14621-1-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
According to vtd spec v3.3 3.14:
"""
Software must not program paging-structure entries to remap any
address to the interrupt address range. Untranslated requests and
translation requests that result in an address in the interrupt range
will be blocked with condition code LGN.4 or SGN.8.
"""
This patch blocks the request that result in interrupt address range.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220210092815.45174-2-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Commit bedd7e93d0 ("virtio-net: fix use after unmap/free for sg")
tries to fix the use after free of the sg by caching the virtqueue
elements in an array and unmap them at once after receiving the
packets, But it forgot to unmap the cached elements on error which
will lead to leaking of mapping and other unexpected results.
Fixing this by detaching the cached elements on error. This addresses
CVE-2022-26353.
Reported-by: Victor Tom <vv474172261@gmail.com>
Cc: qemu-stable@nongnu.org
Fixes: CVE-2022-26353
Fixes: bedd7e93d0 ("virtio-net: fix use after unmap/free for sg")
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
SC is required for some kernel features like vhost-vDPA. So this patch
implements basic SC feature. The idea is pretty simple, for software
emulated DMA it would be always coherent. In this case we can simple
advertise ECAP_SC bit. For VFIO and vhost, thing will be more much
complicated, so this patch simply fail the IOMMU notifier
registration.
In the future, we may want to have a dedicated notifiers flag or
similar mechanism to demonstrate the coherency so VFIO could advertise
that if it has VFIO_DMA_CC_IOMMU, for vhost kernel backend we don't
need that since it's a software backend.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220214060346.72455-1-jasowang@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
When scalable mode is enabled, the passthrough more is not determined
by the context entry but PASID entry, so switch to use the logic of
vtd_dev_pt_enabled() to determine the passthrough mode in
vtd_do_iommu_translate().
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220105041945.13459-2-jasowang@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
When booting with scalable mode, I hit this error:
qemu-system-x86_64: vtd_iova_to_slpte: detected splte reserve non-zero iova=0xfffff002, level=0x1slpte=0x102681803)
qemu-system-x86_64: vtd_iommu_translate: detected translation failure (dev=01:00:00, iova=0xfffff002)
qemu-system-x86_64: New fault is not recorded due to compression of faults
This is because the SNP bit is set for second level page table since
Linux kernel commit 6c00612d0cba1 ("iommu/vt-d: Report right snoop
capability when using FL for IOVA") even if SC is not supported by the
hardware.
To unbreak the guest, ignore the leaf SNP bit for scalable mode
first. In the future we may consider to add SC support.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211129033618.3857-1-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Spec said:
"and len the total of bytes written into the buffer."
For inflateq, deflateq and statsq, we don't process in_sg so the used
length should be zero. For free_page_vq, tough the pages could be
changed by the device (in the destination), spec said:
"Note: len is particularly useful for drivers using untrusted buffers:
if a driver does not know exactly how much has been written by the
device, the driver would have to zero the buffer in advance to ensure
no data leakage occurs."
So 0 should be used as well here.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211129030841.3611-2-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
We only process the first in sg which may lead to the bitmap of the
pages belongs to following sgs were not cleared. This may result more
pages to be migrated. Fixing this by process all in sgs for
free_page_vq.
Acked-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211129030841.3611-1-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We used to access packed descriptor event and off_wrap via
address_space_{write|read}_cached(). When we hit the cache, memcpy()
is used which is not atomic which may lead a wrong value to be read or
wrote.
This patch fixes this by switching to use
virito_{stw|lduw}_phys_cached() to make sure the access is atomic.
Fixes: 683f766567 ("virtio: event suppression support for packed ring")
Cc: qemu-stable@nongnu.org
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211111063854.29060-2-jasowang@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We used to access packed descriptor flags via
address_space_{write|read}_cached(). When we hit the cache, memcpy()
is used which is not an atomic operation which may lead a wrong value
is read or wrote.
So this patch switches to use virito_{stw|lduw}_phys_cached() to make
sure the aceess is atomic.
Fixes: 86044b24e8 ("virtio: basic packed virtqueue support")
Cc: qemu-stable@nongnu.org
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211111063854.29060-1-jasowang@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch implements the multiqueue support for vhost-vdpa. This is
done simply by reading the number of queue pairs from the config space
and initialize the datapath and control path net client.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211020045600.16082-11-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch implements the control virtqueue support for vhost. This
requires virtio-net to figure out the datapath queue pairs and control
virtqueue via is_datapath and pass the number of those two types
of virtqueues to vhost_net_start()/vhost_net_stop().
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211020045600.16082-10-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch introduces a new field in the vhost_dev structure to record
the last virtqueue index for the virtio device. This will be useful
for the vhost backends with 1:N model to start or stop the device
after all the vhost_dev structures were started or stopped.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211020045600.16082-9-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We assume there's no cvq in the past, this is not true when we need
control virtqueue support for vhost-user backends. So this patch
implements the control virtqueue support for vhost-net. As datapath,
the control virtqueue is also required to be coupled with the
NetClientState. The vhost_net_start/stop() are tweaked to accept the
number of datapath queue pairs plus the the number of control
virtqueue for us to start and stop the vhost device.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211020045600.16082-7-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch introduces a boolean for the device has control queue which
can accepts control command via network queue.
The first user would be the control virtqueue support for vhost.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211020045600.16082-6-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch switches to let net_vhost_vdpa_init() to return
NetClientState *. This is used for the callers to allocate multiqueue
NetClientState for multiqueue support.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211020045600.16082-5-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Unlike vhost-kernel, vhost-vdpa adapts a single device multiqueue
model. So we need to simply use virtqueue index as the vhost virtqueue
index. This is a must for multiqueue to work for vhost-vdpa.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211020045600.16082-4-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Vhost-vdpa uses one device multiqueue queue (pairs) model. So we need
to classify the one time request (e.g SET_OWNER) and make sure those
request were only called once per device.
This is used for multiqueue support.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20211020045600.16082-3-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
When mergeable buffer is enabled, we try to set the num_buffers after
the virtqueue elem has been unmapped. This will lead several issues,
E.g a use after free when the descriptor has an address which belongs
to the non direct access region. In this case we use bounce buffer
that is allocated during address_space_map() and freed during
address_space_unmap().
Fixing this by storing the elems temporarily in an array and delay the
unmap after we set the the num_buffers.
This addresses CVE-2021-3748.
Reported-by: Alexander Bulekov <alxndr@bu.edu>
Fixes: fbe78f4f55 ("virtio-net support")
Cc: qemu-stable@nongnu.org
Signed-off-by: Jason Wang <jasowang@redhat.com>
The previous vhost_net_cleanup is sufficient for freeing, calling
vhost_vdpa_del() in this case will lead an extra round of free. Note
that this kind of "double free" is safe since vhost_dev_cleanup() zero
the whole structure.
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20210903091031.47303-8-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The VhostVDPAState is just allocated by qemu_new_net_client() via
g_malloc0() in net_vhost_vdpa_init(). So s->vhost_net is NULL for
sure, let's remove this unnecessary check in vhost_vdpa_add().
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20210903091031.47303-7-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch switches to initialize dev.nvqs from the VhostNetOptions
instead of assuming it was 2. This is useful for implementing control
virtqueue support which will be a single vhost_net structure with a
single cvq.
Note that nvqs is still set to 2 for all users and this patch does not
change functionality.
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Eli Cohen <elic@nvidia.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20210903091031.47303-6-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The nvqs and vqs have been initialized during vhost_net_init() and are
not expected to change during the life cycle of vhost_net
structure. So this patch removes the meaningless assignment.
Reviewed-by: Eli Cohen <elic@nvidia.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20210903091031.47303-4-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Vhost used to compare the dma_as against the address_space_memory to
detect whether the IOMMU is enabled or not. This might not work well
since the virito-bus may call get_dma_as if VIRTIO_F_IOMMU_PLATFORM is
set without an actual IOMMU enabled when device is plugged. In the
case of PCI where pci_get_address_space() is used, the bus master as
is returned. So vhost actually tries to enable device IOTLB even if
the IOMMU is not enabled. This will lead a lots of unnecessary
transactions between vhost and Qemu and will introduce a huge drop of
the performance.
For PCI, an ideal approach is to use pci_device_iommu_address_space()
just for get_dma_as. But Qemu may choose to initialize the IOMMU after
the virtio-pci which lead a wrong address space is returned during
device plugged. So this patch switch to use transport specific way via
iommu_enabled() to detect the IOMMU during vhost start. In this case,
we are fine since we know the IOMMU is initialized correctly.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20210804034803.1644-4-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch implements the PCI transport version of iommu_enabled. This
is done by comparing the address space returned by
pci_device_iommu_address_space() against address_space_memory.
Note that an ideal approach is to use pci_device_iommu_address_space()
in get_dma_as(), but it might not work well since the IOMMU could be
initialized after the virtio-pci device is initialized.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20210804034803.1644-3-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We used to initialize backend_features during vhost_vdpa_init()
regardless whether or not it was supported by vhost. This will lead
the unsupported features like VIRTIO_F_IN_ORDER to be included and set
to the vhost-vdpa during vhost_dev_start. Because the
VIRTIO_F_IN_ORDER is not supported by vhost-vdpa so it won't be
advertised to guest which will break the datapath.
Fix this by not initializing the backend_features, so the
acked_features could be built only from guest features via
vhost_net_ack_features().
Fixes: 108a64818e ("vhost-vdpa: introduce vhost-vdpa backend")
Cc: qemu-stable@nongnu.org
Cc: Gautam Dawar <gdawar@xilinx.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
This patch implements the vq notification mapping support for
vhost-vDPA. This is simply done by using mmap()/munmap() for the
vhost-vDPA fd during device start/stop. For the device without
notification mapping support, we fall back to eventfd based
notification gracefully.
Reviewed-by: Si-Wei Liu <si-wei.liu@oracle.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
vDPA is not tie to any specific hardware, for safety and simplicity,
vhost-vDPA doesn't allow MMIO area to be mapped via IOTLB. Only the
doorbell could be mapped via mmap(). So this patch exclude skip the
ram device from the IOTLB mapping.
Reviewed-by: Si-Wei Liu <si-wei.liu@oracle.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
There could be case that peer is NULL. This can happen when during
network device hot-add where net device needs to be added first. So
the patch check the existence of peer before trying to do the pad.
Fixes: 969e50b61a ("net: Pad short frames to minimum size before sending from SLiRP/TAP")
Signed-off-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Stefan Weil <sw@weilnetz.de>
Message-id: 20210423031803.1479-1-jasowang@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Commit 969e50b61a ("net: Pad short frames to minimum size before
sending from SLiRP/TAP") tries to pad frames but try to recyle the
local array that is used for padding to tap thread. This patch fixes
this by recyling the original buffer.
Fixes: 969e50b61a ("net: Pad short frames to minimum size before sending from SLiRP/TAP")
Tested-by: Howard Spoelstra <hsp.cat7@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Several issues has been reported for query-netdev series. Consider
it's late in the rc, this reverts commit
d32ad10a14.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Several issues has been reported for query-netdev series. Consider
it's late in the rc, this reverts commit
3c3b656885.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Several issues has been reported for query-netdev info
series. Consider it's late in the rc, this reverts commit
commit 59b5437eb7.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Several issues has been reported for query-netdev info
series. Consider it's late in the rc, this reverts commit
a0724776c5.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Several issues has been reported for query-netdev info
series. Consider it's late in the rc, this reverts commit
f2e8319d45.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Commit 4c70875372 ("pci: advertise a page aligned ATS") advertises
the page aligned via ATS capability (RO) to unbrek recent Linux IOMMU
drivers since 5.2. But it forgot the compat the capability which
breaks the migration from old machine type:
(qemu) qemu-kvm: get_pci_config_device: Bad config data: i=0x104 read:
0 device: 20 cmask: ff wmask: 0 w1cmask:0
This patch introduces a new parameter "x-ats-page-aligned" for
virtio-pci device and turns it on for machine type which is newer than
5.1.
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Cc: qemu-stable@nongnu.org
Fixes: 4c70875372 ("pci: advertise a page aligned ATS")
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20210406040330.11306-1-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch switches to use qemu_receive_receive_iov() which can detect
reentrancy and return early.
This is intended to address CVE-2021-3416.
Cc: Prasad J Pandit <ppandit@redhat.com>
Cc: qemu-stable@nongnu.org
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
This patch switches to use qemu_receive_packet() which can detect
reentrancy and return early.
This is intended to address CVE-2021-3416.
Cc: Prasad J Pandit <ppandit@redhat.com>
Cc: qemu-stable@nongnu.org
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
This patch switches to use qemu_receive_packet() which can detect
reentrancy and return early.
This is intended to address CVE-2021-3416.
Cc: Prasad J Pandit <ppandit@redhat.com>
Cc: qemu-stable@nongnu.org
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com
Signed-off-by: Jason Wang <jasowang@redhat.com>
This patch switches to use qemu_receive_packet() which can detect
reentrancy and return early.
This is intended to address CVE-2021-3416.
Cc: Prasad J Pandit <ppandit@redhat.com>
Cc: qemu-stable@nongnu.org
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Some NIC supports loopback mode and this is done by calling
nc->info->receive() directly which in fact suppresses the effort of
reentrancy check that is done in qemu_net_queue_send().
Unfortunately we can't use qemu_net_queue_send() here since for
loopback there's no sender as peer, so this patch introduce a
qemu_receive_packet() which is used for implementing loopback mode
for a NIC with this check.
NIC that supports loopback mode will be converted to this helper.
This is intended to address CVE-2021-3416.
Cc: Prasad J Pandit <ppandit@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Cc: qemu-stable@nongnu.org
Signed-off-by: Jason Wang <jasowang@redhat.com>
During procss_tx_desc(), driver can try to chain data descriptor with
legacy descriptor, when will lead underflow for the following
calculation in process_tx_desc() for bytes:
if (tp->size + bytes > msh)
bytes = msh - tp->size;
This will lead a infinite loop. So check and fail early if tp->size if
greater or equal to msh.
Reported-by: Alexander Bulekov <alxndr@bu.edu>
Reported-by: Cheolwoo Myung <cwmyung@snu.ac.kr>
Reported-by: Ruhr-University Bochum <bugs-syssec@rub.de>
Cc: Prasad J Pandit <ppandit@redhat.com>
Cc: qemu-stable@nongnu.org
Signed-off-by: Jason Wang <jasowang@redhat.com>
Currently, the default msix vectors for virtio-net-pci is 3 which is
obvious not suitable for multiqueue guest, so we depends on the user
or management tools to pass a correct vectors parameter. In fact, we
can simplifying this by calculating the number of vectors on realize.
Consider we have N queues, the number of vectors needed is 2*N + 2
(#queue pairs + plus one config interrupt and control vq). We didn't
check whether or not host support control vq because it was added
unconditionally by qemu to avoid breaking legacy guests such as Minix.
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
After Linux kernel commit 61363c1474b1 ("iommu/vt-d: Enable ATS only
if the device uses page aligned address."), ATS will be only enabled
if device advertises a page aligned request.
Unfortunately, vhost-net is the only user and we don't advertise the
aligned request capability in the past since both vhost IOTLB and
address_space_get_iotlb_entry() can support non page aligned request.
Though it's not clear that if the above kernel commit makes
sense. Let's advertise a page aligned ATS here to make vhost device
IOTLB work with Intel IOMMU again.
Note that in the future we may extend pcie_ats_init() to accept
parameters like queue depth and page alignment.
Cc: qemu-stable@nongnu.org
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20200909081731.24688-1-jasowang@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
To speed up the memory mapping updating between vhost-vDPA and vDPA
device driver, this patch passes the IOTLB batching flags via IOTLB
API. Two new flags was introduced, VHOST_IOTLB_BATCH_BEGIN is a hint
that a bathced IOTLB updating may be initiated from the
userspace. VHOST_IOTLB_BATCH_END is a hint that userspace has finished
the updating:
VHOST_IOTLB_BATCH_BEGIN
VHOST_IOTLB_UPDATE/VHOST_IOTLB_INVALIDATE
...
VHOST_IOTLB_BATCH_END
Vhost-vDPA can then know that all mappings has been set and can do
optimization like passing all the mappings to the vDPA device driver.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20200907104903.31551-4-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch tries to switch to use new kernel IOTLB format V2. Previous
version may have inconsistent ABI between 32bit and 64bit machines
because of the hole after type field. Refer kernel commit
("429711aec282 vhost: switch to use new message format") for more
information.
To enable this feature, qemu need to use a new ioctl
VHOST_SET_BACKEND_FEATURE with VHOST_BACKEND_F_IOTLB_MSG_V2 bit. A new
vhost setting backend features ops was introduced. And when we try to
set features for vhost dev, we will examine the support of new IOTLB
format and enable it. This process is total transparent to guest,
which means we can have different IOTLB message type in src and dst
during migration.
The conversion of IOTLB message is straightforward, just check the
type and behave accordingly.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20200907104903.31551-3-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The memory API allows DMA into NIC's MMIO area. This means the NIC's
RX routine must be reentrant. Instead of auditing all the NIC, we can
simply detect the reentrancy and return early. The queue->delivering
is set and cleared by qemu_net_queue_deliver() for other queue helpers
to know whether the delivering in on going (NIC's receive is being
called). We can check it and return early in qemu_net_queue_flush() to
forbid reentrant RX.
Signed-off-by: Jason Wang <jasowang@redhat.com>
We try to check whether a peer is VDPA in order to get config from
there - with no peer, this leads to a NULL
pointer dereference. Add a check before trying to access the peer
type. No peer means not VDPA.
Fixes: 108a64818e ("vhost-vdpa: introduce vhost-vdpa backend")
Cc: Cindy Lu <lulu@redhat.com>
Tested-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The sender of packet will be checked in the qemu_net_queue_purge() but
we use NetClientState not its peer when trying to purge the incoming
queue in qemu_flush_or_purge_packets(). This will trigger the assert
in virtio_net_reset since we can't pass the sender check:
hw/net/virtio-net.c:533: void virtio_net_reset(VirtIODevice *): Assertion
`!virtio_net_get_subqueue(nc)->async_tx.elem' failed.
#9 0x55a33fa31b78 in virtio_net_reset hw/net/virtio-net.c:533:13
#10 0x55a33fc88412 in virtio_reset hw/virtio/virtio.c:1919:9
#11 0x55a341d82764 in virtio_bus_reset hw/virtio/virtio-bus.c:95:9
#12 0x55a341dba2de in virtio_pci_reset hw/virtio/virtio-pci.c:1824:5
#13 0x55a341db3e02 in virtio_pci_common_write hw/virtio/virtio-pci.c:1252:13
#14 0x55a33f62117b in memory_region_write_accessor memory.c:496:5
#15 0x55a33f6205e4 in access_with_adjusted_size memory.c:557:18
#16 0x55a33f61e177 in memory_region_dispatch_write memory.c:1488:16
Reproducer:
https://www.mail-archive.com/qemu-devel@nongnu.org/msg701914.html
Fix by using the peer.
Reported-by: "Alexander Bulekov" <alxndr@bu.edu>
Acked-by: Alexander Bulekov <alxndr@bu.edu>
Fixes: ca77d85e1d ("net: complete all queued packets on VM stop")
Cc: qemu-stable@nongnu.org
Signed-off-by: Jason Wang <jasowang@redhat.com>
Spec said: The driver uses this to selectively prevent the device from
executing requests from this virtqueue. 1 - enabled; 0 - disabled.
Though write 0 to queue_enable is forbidden by the spec, we should not
assume that the value is 1.
Fix this by ignore the write value other than 1.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20200610054351.15811-1-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We turn on device IOTLB via VIRTIO_F_IOMMU_PLATFORM unconditionally on
platform without IOMMU support. This can lead unnecessary IOTLB
transactions which will damage the performance.
Fixing this by check whether the device is backed by IOMMU and disable
device IOTLB.
Reported-by: Halil Pasic <pasic@linux.ibm.com>
Tested-by: Halil Pasic <pasic@linux.ibm.com>
Reviewed-by: Halil Pasic <pasic@linux.ibm.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20200302042454.24814-1-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch implements basic support for the packed virtqueue. Compare
the split virtqueue which has three rings, packed virtqueue only have
one which is supposed to have better cache utilization and more
hardware friendly.
Please refer virtio specification for more information.
Signed-off-by: Wei Xu <wexu@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Message-Id: <20191025083527.30803-6-eperezma@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We should not raise any interrupt after VM has been stopped but this
is what e1000 currently did when mit timer is active in
pre_save(). Fixing this by scheduling a timer in post_load() which can
make sure the interrupt was raised when VM is running.
Reported-and-tested-by: Longpeng <longpeng2@huawei.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
We used to set backend unconditionally, this won't work for some
guests (e.g windows driver) who may not initialize all virtqueues. For
kernel backend, this will fail since it may try to validate the rings
during setting backend.
Fixing this by simply skipping the backend set when we find desc is
not ready.
Reviewed-by: Michael S. Tsirkin<mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
We try to detect and drop too large packet (>INT_MAX) in 1592a99470
("net: ignore packet size greater than INT_MAX") during packet
delivering. Unfortunately, this is not sufficient as we may hit
another integer overflow when trying to queue such large packet in
qemu_net_queue_append_iov():
- size of the allocation may overflow on 32bit
- packet->size is integer which may overflow even on 64bit
Fixing this by moving the check to qemu_sendv_packet_async() which is
the entrance of all networking codes and reduce the limit to
NET_BUFSIZE to be more conservative. This works since:
- For the callers that call qemu_sendv_packet_async() directly, they
only care about if zero is returned to determine whether to prevent
the source from producing more packets. A callback will be triggered
if peer can accept more then source could be enabled. This is
usually used by high speed networking implementation like virtio-net
or netmap.
- For the callers that call qemu_sendv_packet() that calls
qemu_sendv_packet_async() indirectly, they often ignore the return
value. In this case qemu will just the drop packets if peer can't
receive.
Qemu will copy the packet if it was queued. So it was safe for both
kinds of the callers to assume the packet was sent.
Since we move the check from qemu_deliver_packet_iov() to
qemu_sendv_packet_async(), it would be safer to make
qemu_deliver_packet_iov() static to prevent any external user in the
future.
This is a revised patch of CVE-2018-17963.
Cc: qemu-stable@nongnu.org
Cc: Li Qiang <liq3ea@163.com>
Fixes: 1592a99470 ("net: ignore packet size greater than INT_MAX")
Reported-by: Li Qiang <liq3ea@gmail.com>
Reviewed-by: Li Qiang <liq3ea@gmail.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 20181204035347.6148-2-jasowang@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The e1000 emulation silently discards RX packets if there's
insufficient space in the ring buffer. This leads to errors
on higher-level protocols in the guest, with no indication
about the error cause.
This patch increments the "Missed Packets Count" (MPC) and
"Receive No Buffers Count" (RNBC) HW counters in this case.
As the emulation has no FIFO for buffering packets that can't
immediately be pushed to the guest, these two registers are
practically equivalent (see 10.2.7.4, 10.2.7.33 in
https://www.intel.com/content/www/us/en/embedded/products/networking/82574l-gbe-controller-datasheet.html).
On a Linux guest, the register content will be reflected in
the "rx_missed_errors" and "rx_no_buffer_count" stats from
"ethtool -S", and in the "missed" stat from "ip -s -s link show",
giving at least some hint about the error cause inside the guest.
If the cause is known, problems like this can often be avoided
easily, by increasing the number of RX descriptors in the guest
e1000 driver (e.g under Linux, "e1000.RxDescriptors=1024").
The patch also adds a qemu trace message for this condition.
Signed-off-by: Martin Wilck <mwilck@suse.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
There should not be a reason for passing a packet size greater than
INT_MAX. It's usually a hint of bug somewhere, so ignore packet size
greater than INT_MAX in qemu_deliver_packet_iov()
CC: qemu-stable@nongnu.org
Reported-by: Daniel Shapira <daniel@twistlock.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
In pcnet_receive(), we try to assign size_ to size which converts from
size_t to integer. This will cause troubles when size_ is greater
INT_MAX, this will lead a negative value in size and it can then pass
the check of size < MIN_BUF_SIZE which may lead out of bound access
for both buf and buf1.
Fixing by converting the type of size to size_t.
CC: qemu-stable@nongnu.org
Reported-by: Daniel Shapira <daniel@twistlock.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
In rtl8139_do_receive(), we try to assign size_ to size which converts
from size_t to integer. This will cause troubles when size_ is greater
INT_MAX, this will lead a negative value in size and it can then pass
the check of size < MIN_BUF_SIZE which may lead out of bound access of
for both buf and buf1.
Fixing by converting the type of size to size_t.
CC: qemu-stable@nongnu.org
Reported-by: Daniel Shapira <daniel@twistlock.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
In ne2000_receive(), we try to assign size_ to size which converts
from size_t to integer. This will cause troubles when size_ is greater
INT_MAX, this will lead a negative value in size and it can then pass
the check of size < MIN_BUF_SIZE which may lead out of bound access of
for both buf and buf1.
Fixing by converting the type of size to size_t.
CC: qemu-stable@nongnu.org
Reported-by: Daniel Shapira <daniel@twistlock.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
When IOMMU is enabled, we store virtqueue metadata as iova (though it
may has _phys suffix) and access them through dma helpers. Any
translation failures could be reported by IOMMU.
In this case, trying to validate iova against gpa won't work and will
cause a false error reporting. So this patch bypasses the ring
verification if IOMMU is enabled which is similar to the behavior
before 0ca1fd2d68 that calls vhost_memory_map() which is a nop when
IOMMU is enabled.
Fixes: 0ca1fd2d68 ("vhost: Simplify ring verification checks")
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Guest state should not be touched if VM is stopped, unfortunately we
didn't check running state and tried to drain tx queue unconditionally
in virtio_net_set_status(). A crash was then noticed as a migration
destination when user type quit after virtqueue state is loaded but
before region cache is initialized. In this case,
virtio_net_drop_tx_queue_data() tries to access the uninitialized
region cache.
Fix this by only dropping tx queue data when vm is running.
Fixes: 283e2c2adc ("net: virtio-net discards TX data after link down")
Cc: Yuri Benditovich <yuri.benditovich@daynix.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: qemu-stable@nongnu.org
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Jason Wang