Currently, default values of kickfd and callfd are -1.
If the values are -1, current code guesses kickfd and callfd haven't
been initialized yet. Then vhost library will guess the virtqueue isn't
ready for processing.
But callfd and kickfd will be set as -1 when "--enable-kvm"
isn't specified in QEMU command line. It means we cannot treat -1 as
uninitialized state.
The patch defines -1 and -2 as VIRTIO_INVALID_EVENTFD and
VIRTIO_UNINITIALIZED_EVENTFD, and uses VIRTIO_UNINITIALIZED_EVENTFD for
the default values of kickfd and callfd.
Signed-off-by: Tetsuya Mukawa <mukawa@igel.co.jp>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Broadcast RARP packet by injecting it to receiving mbuf array at
rte_vhost_dequeue_burst().
Commit 33226236a3 ("vhost: handle request to send RARP") iterates
all host interfaces and then broadcast it by all of them. It did
notify the switches about the new location of the migrated VM, however,
the mac learning table in the target host is wrong (at least in my
test with OVS):
$ ovs-appctl fdb/show ovsbr0
port VLAN MAC Age
1 0 b6:3c:72:71:cd:4d 10
LOCAL 0 b6:3c:72:71:cd:4e 10
LOCAL 0 52:54:00:12:34:68 9
1 0 56:f6:64:2c:bc:c0 1
Where 52:54:00:12:34:68 is the mac of the VM. As you can see from the
above, the port learned is "LOCAL", which is the "ovsbr0" port. That
is reasonable, since we indeed send the pkt by the "ovsbr0" interface.
The wrong mac table lead all the packets to the VM go to the "ovsbr0"
in the end, which ends up with all packets being lost, until the guest
send a ARP quest (or reply) to refresh the mac learning table.
Jianfeng then came up with a solution I have thought of firstly but NAKed
by myself, concerning it has potential issues [0]. The solution is as title
stated: broadcast the RARP packet by injecting it to the receiving mbuf
arrays at rte_vhost_dequeue_burst(). The re-bring of that idea made me
think it twice; it looked like a false concern to me then. And I had done
a rough verification: it worked as expected.
[0]: http://dpdk.org/ml/archives/dev/2016-February/033527.html
Another note is that while preparing this version, I found that DPDK has
some ARP related structures and macros defined. So, use them instead of
the one from standard header files here.
Cc: Thibaut Collet <thibaut.collet@6wind.com>
Suggested-by: Jianfeng Tan <jianfeng.tan@intel.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
The vhost_net_device_ops indirection is unnecessary because there is only
one implementation of the vhost common code.
Removing it makes the code more readable.
Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
The common vhost code only supported a single mmap per device. vhost-user
worked around this by saving the address/length/fd of each mmap after the end
of the rte_virtio_memory struct. This only works if the vhost-user code frees
dev->mem, since the common code is unaware of the extra info. The
VHOST_USER_RESET_OWNER message is one situation where the common code frees
dev->mem and leaks the fds and mappings. This happens every time I shut down a
VM.
The new code calls back into the implementation (vhost-user or vhost-cuse) to
clean up these resources.
The vhost-cuse changes are only compile tested.
Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
To claim that we support vhost-user live migration support:
SET_LOG_BASE request will be send only when this feature flag
is set.
Besides this flag, we actually need another feature flag set
to make vhost-user live migration work: VHOST_F_LOG_ALL.
Which, however, has been enabled long time ago.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Tested-by: Pavel Fedin <p.fedin@samsung.com>
While in former patch we enabled GUEST_ANNOUNCE feature, so that the
guest OS will broadcast a GARP message after migration to notify the
switch about the new location of migrated VM, the thing is that
GUEST_ANNOUNCE is enabled since kernel v3.5 only. For older kernel,
VHOST_USER_SEND_RARP request comes to rescue.
The payload of this new request is the mac address of the migrated VM,
with that, we could construct a RARP message, and then broadcast it
to host interfaces.
That's how this patch works:
- list all interfaces, with the help of SIOCGIFCONF ioctl command
- construct an RARP message and broadcast it
Cc: Thibaut Collet <thibaut.collet@6wind.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
VHOST_USER_SET_LOG_BASE request is used to tell the backend (dpdk
vhost-user) where we should log dirty pages, and how big the log
buffer is.
This request introduces a new payload:
typedef struct VhostUserLog {
uint64_t mmap_size;
uint64_t mmap_offset;
} VhostUserLog;
Also, a fd is delivered from QEMU by ancillary data.
With those info given, an area of memory is mmaped, assigned
to dev->log_base, for logging dirty pages.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Signed-off-by: Victor Kaplansky <victork@redhat.com>
Tested-by: Pavel Fedin <p.fedin@samsung.com>
The VHOST_USER_SET_VRING_ENABLE request was sent for each queue-pair.
However, it's changed to be sent per queue in the queue-pair by QEMU
commit dc3db6ad ("vhost-user: start/stop all rings"). The change
is reasonable, as we send all other request per queue, instead of
queue-pair.
Hence we should do proper changes to adapt to the QEMU change here.
Otherwise, a segfault will be triggered when last TX queue was enabled.
Signed-off-by: Victor Kaplansky <victork@redhat.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
This patch fixes a bug under lower version linux kernel, mmap()
fails when length is not aligned with hugepage size. mmap()
without flag of MAP_ANONYMOUS, should be called with length
argument aligned with hugepagesz at older longterm version
Linux, like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
This bug was fixed in Linux kernel by commit:
dab2d3dc45ae7343216635d981d43637e1cb7d45
To avoid failure, make sure in caller to keep length aligned.
Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
Fix build error:
virtio-net.c:80:89: error: ‘VIRTIO_NET_F_MQ’ undeclared here
rte_virtio_net.h:109: error: ‘VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX’ undeclared here
Above two virtio-net MQ macros are introduced since kernel v3.8.
For older kernel, we should not reference them directly, hence,
this patch introduced two wrapper macros, with proper values
being set depending on we support MQ or not.
Fixes: b09b198bfb ("vhost-user: announce queue number in message")
Reported-by: Yongjie Gu <yongjiex.gu@intel.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Tested-by: David Marchand <david.marchand@6wind.com>
This message is used to enable/disable a specific vring queue pair.
The first queue pair is enabled by default.
Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Acked-by: Flavio Leitner <fbl@sysclose.org>
Acked-by: Huawei Xie <huawei.xie@intel.com>
All queue pairs, including the default (the first) queue pair,
are allocated dynamically, when a vring_call message is received
first time for a specific queue pair.
This is a refactor work for enabling vhost-user multiple queue;
it should not break anything as it does no functional changes:
we don't support mq set, so there is only one mq at max.
This patch is based on Changchun's patch.
Signed-off-by: Ouyang Changchun <changchun.ouyang@intel.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Acked-by: Flavio Leitner <fbl@sysclose.org>
Acked-by: Huawei Xie <huawei.xie@intel.com>
Add VHOST_USER_GET_QUEUE_NUM message
to tell the frontend (qemu) how many queue pairs we support.
And it is initiated to VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Acked-by: Flavio Leitner <fbl@sysclose.org>
Acked-by: Huawei Xie <huawei.xie@intel.com>
The two protocol features messages are introduced by qemu vhost
maintainer(Michael) for extendting vhost-user interface. Here is
an excerpta from the vhost-user spec:
Any protocol extensions are gated by protocol feature bits,
which allows full backwards compatibility on both master
and slave.
The vhost-user multiple queue features will be treated as a vhost-user
extension, hence, we have to implement the two messages first.
VHOST_USER_PROTOCOL_FEATURES is initialized to 0, as we don't support
any yet.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Acked-by: Flavio Leitner <fbl@sysclose.org>
Acked-by: Huawei Xie <huawei.xie@intel.com>
virtio-net search for it's device in reset_owner.
The function don't check the return result of get_config_ll_entry.
Using get_config_ll_entry in reset_owner don't show any error when the
device is not found. This patch fix this by using get_device instead
instead of get_config_ll_entry.
In user_get_vring_base, get_device return is not checked and may cause
segfault when device is not found.
Signed-off-by: Jerome Jutteau <jerome.jutteau@outscale.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
According to eventfd man page:
typedef uint64_t eventfd_t;
int eventfd_read(int fd, eventfd_t *value);
int eventfd_write(int fd, eventfd_t value);
eventfd_t is defined for the second arg(value), but not for fd.
Here I redefine those fd fields to `int' type, which also removes
the redundant (int) cast. And as the man page stated, we should
cast 1 to eventfd_t type for eventfd_write().
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
It adds more readable log info if a socket fails to bind to
local socket file name.
Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
rte_vhost_driver_unregister API will remove the listenfd from event list,
and then close it.
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Signed-off-by: Peng Sun <peng.a.sun@intel.com>
In the event handler of connection fd, the connection fd could be possibly
closed. The event dispatch loop would then try to remove the fd from fdset.
Between these two actions, another thread might register a new listenfd
reusing the val of just closed fd, so we couldn't call fdset_del which would
wrongly clean up the new listenfd. A new function fdset_del_slot is provided
to cleanup the fd at the specified location.
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Fix the error "missing initializer" and "cast to pointer from integer of different size".
For the pointer to integer cast issue, need to investigate changing the typeof mapped_address.
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Acked-by: Changchun Ouyang <changchun.ouyang@intel.com>
Previous vhost implementation wrongly name kickfd as callfd and callfd as kickfd.
It is functional correct, but causes confusion.
Exchange kickfd and callfd to avoid confusion.
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Acked-by: Changchun Ouyang <changchun.ouyang@intel.com>
Acked-by: Tetsuya Mukawa <mukawa@igel.co.jp>
This patch reorder the code a bit to use loop instead of goto.
Besides, remove abudant check 'fd != -1'.
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
combine sleep into select when there is no file descriptors to be monitored.
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
This patch fixes the segfault issue in the case vhost receives
new VHOST_SET_MEM_TABLE message without VHOST_VRING_GET_VRING_BASE
(which we uses as the stop message).
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Acked-by: Tommy Long <thomas.long@intel.com>
* support calling rte_vhost_driver_register after rte_vhost_driver_session_start
* add mutext to protect fdset from concurrent access
* add busy flag in fdentry. this flag is set before cb and cleared after cb is finished.
mutex lock scenario in vhost:
* event_dispatch(in rte_vhost_driver_session_start) runs in a separate thread, infinitely
processing vhost messages through cb(callback).
* event_dispatch acquires the lock, get the cb and its context, mark the busy flag,
and releases the mutex.
* vserver_new_vq_conn cb calls fdset_add, which acquires the mutex and add new fd into fdset.
* vserver_message_handler cb frees data context, marks remove flag to request to delete
connfd(connection fd) from fdset.
* after cb returns, event_dispatch
1. clears busy flag.
2. if there is remove request, call fdset_del, which acquires mutex, checks busy flag, and
removes connfd from fdset.
* rte_vhost_driver_unregister(not implemented) runs in another thread, acquires the mutex,
calls fdset_del to remove fd(listenerfd) from fdset. Then it could free data context.
The above steps ensures fd data context isn't freed when cb is using.
VM(s) should have been shutdown before rte_vhost_driver_unregister.
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Tetsuya Mukawa <mukawa@igel.co.jp>
for vhost-cuse, ifname is the name of the tap device
for vhost-user, ifname is the name of the unix domain socket path
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Signed-off-by: Przemyslaw Czesnowicz <przemyslaw.czesnowicz@intel.com>
Acked-by: Tetsuya Mukawa <mukawa@igel.co.jp>
In rte_vhost_driver_register(), vhost unix domain socket listener fd is created
and added to polled(based on select) fdset.
In rte_vhost_driver_session_start(), fds in the fdset are checked for
processing. If there is new connection from qemu, connection fd accepted is
added to polled fdset. The listener and connection fds in the fdset are
then both checked. When there is message on the connection fd, its
callback vserver_message_handler is called to process vhost-user messages.
To support identifying which virtio is from which guest VM, we could call
rte_vhost_driver_register with different socket path. Virtio devices from
same VM will connect to VM specific socket. The socket path information is
stored in the virtio_net structure.
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Signed-off-by: Przemyslaw Czesnowicz <przemyslaw.czesnowicz@intel.com>
Acked-by: Tetsuya Mukawa <mukawa@igel.co.jp>
