Support host notifier in libvhost-user and vhost-user-bridge

index beeed0c43f..d7872d2df6 100644
@@ -314,11 +314,6 @@ vu_message_write(VuDev *dev, int conn_fd, VhostUserMsg *vmsg)
         msg.msg_controllen = 0;
     }
 
-    /* Set the version in the flags when sending the reply */
-    vmsg->flags &= ~VHOST_USER_VERSION_MASK;
-    vmsg->flags |= VHOST_USER_VERSION;
-    vmsg->flags |= VHOST_USER_REPLY_MASK;
-
     do {
         rc = sendmsg(conn_fd, &msg, 0);
     } while (rc < 0 && (errno == EINTR || errno == EAGAIN));

@@ -339,6 +334,39 @@ vu_message_write(VuDev *dev, int conn_fd, VhostUserMsg *vmsg)
     return true;
 }
 
+static bool
+vu_send_reply(VuDev *dev, int conn_fd, VhostUserMsg *vmsg)
+{
+    /* Set the version in the flags when sending the reply */
+    vmsg->flags &= ~VHOST_USER_VERSION_MASK;
+    vmsg->flags |= VHOST_USER_VERSION;
+    vmsg->flags |= VHOST_USER_REPLY_MASK;
+
+    return vu_message_write(dev, conn_fd, vmsg);
+}
+
+static bool
+vu_process_message_reply(VuDev *dev, const VhostUserMsg *vmsg)
+{
+    VhostUserMsg msg_reply;
+
+    if ((vmsg->flags & VHOST_USER_NEED_REPLY_MASK) == 0) {
+        return true;
+    }
+
+    if (!vu_message_read(dev, dev->slave_fd, &msg_reply)) {
+        return false;
+    }
+
+    if (msg_reply.request != vmsg->request) {
+        DPRINT("Received unexpected msg type. Expected %d received %d",
+               vmsg->request, msg_reply.request);
+        return false;
+    }
+
+    return msg_reply.payload.u64 == 0;
+}
+
 /* Kick the log_call_fd if required. */
 static void
 vu_log_kick(VuDev *dev)

@@ -534,7 +562,7 @@ vu_set_mem_table_exec_postcopy(VuDev *dev, VhostUserMsg *vmsg)
 
     /* Send the message back to qemu with the addresses filled in */
     vmsg->fd_num = 0;
-    if (!vu_message_write(dev, dev->sock, vmsg)) {
+    if (!vu_send_reply(dev, dev->sock, vmsg)) {
         vu_panic(dev, "failed to respond to set-mem-table for postcopy");
         return false;
     }

@@ -914,6 +942,37 @@ void vu_set_queue_handler(VuDev *dev, VuVirtq *vq,
     }
 }
 
+bool vu_set_queue_host_notifier(VuDev *dev, VuVirtq *vq, int fd,
+                                int size, int offset)
+{
+    int qidx = vq - dev->vq;
+    int fd_num = 0;
+    VhostUserMsg vmsg = {
+        .request = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
+        .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
+        .size = sizeof(vmsg.payload.area),
+        .payload.area = {
+            .u64 = qidx & VHOST_USER_VRING_IDX_MASK,
+            .size = size,
+            .offset = offset,
+        },
+    };
+
+    if (fd == -1) {
+        vmsg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
+    } else {
+        vmsg.fds[fd_num++] = fd;
+    }
+
+    vmsg.fd_num = fd_num;
+
+    if (!vu_message_write(dev, dev->slave_fd, &vmsg)) {
+        return false;
+    }
+
+    return vu_process_message_reply(dev, &vmsg);
+}
+
 static bool
 vu_set_vring_call_exec(VuDev *dev, VhostUserMsg *vmsg)
 {

@@ -966,7 +1025,8 @@ static bool
 vu_get_protocol_features_exec(VuDev *dev, VhostUserMsg *vmsg)
 {
     uint64_t features = 1ULL << VHOST_USER_PROTOCOL_F_LOG_SHMFD |
-                        1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ;
+                        1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ |
+                        1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER;
 
     if (have_userfault()) {
         features |= 1ULL << VHOST_USER_PROTOCOL_F_PAGEFAULT;

@@ -1250,7 +1310,7 @@ vu_dispatch(VuDev *dev)
         goto end;
     }
 
-    if (!vu_message_write(dev, dev->sock, &vmsg)) {
+    if (!vu_send_reply(dev, dev->sock, &vmsg)) {
         goto end;
     }
 

index b27075ea3b..90806a25f9 100644
@@ -51,6 +51,7 @@ enum VhostUserProtocolFeature {
     VHOST_USER_PROTOCOL_F_CRYPTO_SESSION = 7,
     VHOST_USER_PROTOCOL_F_PAGEFAULT = 8,
     VHOST_USER_PROTOCOL_F_CONFIG = 9,
+    VHOST_USER_PROTOCOL_F_HOST_NOTIFIER = 10,
 
     VHOST_USER_PROTOCOL_F_MAX
 };

@@ -92,6 +93,14 @@ typedef enum VhostUserRequest {
     VHOST_USER_MAX
 } VhostUserRequest;
 
+typedef enum VhostUserSlaveRequest {
+    VHOST_USER_SLAVE_NONE = 0,
+    VHOST_USER_SLAVE_IOTLB_MSG = 1,
+    VHOST_USER_SLAVE_CONFIG_CHANGE_MSG = 2,
+    VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG = 3,
+    VHOST_USER_SLAVE_MAX
+}  VhostUserSlaveRequest;
+
 typedef struct VhostUserMemoryRegion {
     uint64_t guest_phys_addr;
     uint64_t memory_size;

@@ -122,6 +131,12 @@ static VhostUserConfig c __attribute__ ((unused));
                                    + sizeof(c.size) \
                                    + sizeof(c.flags))
 
+typedef struct VhostUserVringArea {
+    uint64_t u64;
+    uint64_t size;
+    uint64_t offset;
+} VhostUserVringArea;
+
 #if defined(_WIN32)
 # define VU_PACKED __attribute__((gcc_struct, packed))
 #else

@@ -133,6 +148,7 @@ typedef struct VhostUserMsg {
 
 #define VHOST_USER_VERSION_MASK     (0x3)
 #define VHOST_USER_REPLY_MASK       (0x1 << 2)
+#define VHOST_USER_NEED_REPLY_MASK  (0x1 << 3)
     uint32_t flags;
     uint32_t size; /* the following payload size */
 

@@ -145,6 +161,7 @@ typedef struct VhostUserMsg {
         VhostUserMemory memory;
         VhostUserLog log;
         VhostUserConfig config;
+        VhostUserVringArea area;
     } payload;
 
     int fds[VHOST_MEMORY_MAX_NREGIONS];

@@ -368,6 +385,20 @@ VuVirtq *vu_get_queue(VuDev *dev, int qidx);
 void vu_set_queue_handler(VuDev *dev, VuVirtq *vq,
                           vu_queue_handler_cb handler);
 
+/**
+ * vu_set_queue_host_notifier:
+ * @dev: a VuDev context
+ * @vq: a VuVirtq queue
+ * @fd: a file descriptor
+ * @size: host page size
+ * @offset: notifier offset in @fd file
+ *
+ * Set queue's host notifier. This function may be called several
+ * times for the same queue. If called with -1 @fd, the notifier
+ * is removed.
+ */
+bool vu_set_queue_host_notifier(VuDev *dev, VuVirtq *vq, int fd,
+                                int size, int offset);
 
 /**
  * vu_queue_set_notification:

index e0605a529e..0884294141 100644
@@ -29,6 +29,7 @@
 
 #define _FILE_OFFSET_BITS 64
 
+#include "qemu/atomic.h"
 #include "qemu/osdep.h"
 #include "qemu/iov.h"
 #include "standard-headers/linux/virtio_net.h"

@@ -65,6 +66,11 @@ typedef struct VubrDev {
     int sock;
     int ready;
     int quit;
+    struct {
+        int fd;
+        void *addr;
+        pthread_t thread;
+    } notifier;
 } VubrDev;
 
 static void

@@ -445,14 +451,22 @@ static uint64_t
 vubr_get_features(VuDev *dev)
 {
     return 1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE |
-        1ULL << VIRTIO_NET_F_MRG_RXBUF;
+        1ULL << VIRTIO_NET_F_MRG_RXBUF |
+        1ULL << VIRTIO_F_VERSION_1;
 }
 
 static void
 vubr_queue_set_started(VuDev *dev, int qidx, bool started)
 {
+    VubrDev *vubr = container_of(dev, VubrDev, vudev);
     VuVirtq *vq = vu_get_queue(dev, qidx);
 
+    if (started && vubr->notifier.fd >= 0) {
+        vu_set_queue_host_notifier(dev, vq, vubr->notifier.fd,
+                                   getpagesize(),
+                                   qidx * getpagesize());
+    }
+
     if (qidx % 2 == 1) {
         vu_set_queue_handler(dev, vq, started ? vubr_handle_tx : NULL);
     }

@@ -522,6 +536,8 @@ vubr_new(const char *path, bool client)
         vubr_die("socket");
     }
 
+    dev->notifier.fd = -1;
+
     un.sun_family = AF_UNIX;
     strcpy(un.sun_path, path);
     len = sizeof(un.sun_family) + strlen(path);

@@ -559,6 +575,73 @@ vubr_new(const char *path, bool client)
     return dev;
 }
 
+static void *notifier_thread(void *arg)
+{
+    VuDev *dev = (VuDev *)arg;
+    VubrDev *vubr = container_of(dev, VubrDev, vudev);
+    int pagesize = getpagesize();
+    int qidx;
+
+    while (true) {
+        for (qidx = 0; qidx < VHOST_MAX_NR_VIRTQUEUE; qidx++) {
+            uint16_t *n = vubr->notifier.addr + pagesize * qidx;
+
+            if (*n == qidx) {
+                *n = 0xffff;
+                /* We won't miss notifications if we reset
+                 * the memory first. */
+                smp_mb();
+
+                DPRINT("Got a notification for queue%d via host notifier.\n",
+                       qidx);
+
+                if (qidx % 2 == 1) {
+                    vubr_handle_tx(dev, qidx);
+                }
+            }
+            usleep(1000);
+        }
+    }
+
+    return NULL;
+}
+
+static void
+vubr_host_notifier_setup(VubrDev *dev)
+{
+    char template[] = "/tmp/vubr-XXXXXX";
+    pthread_t thread;
+    size_t length;
+    void *addr;
+    int fd;
+
+    length = getpagesize() * VHOST_MAX_NR_VIRTQUEUE;
+
+    fd = mkstemp(template);
+    if (fd < 0) {
+        vubr_die("mkstemp()");
+    }
+
+    if (posix_fallocate(fd, 0, length) != 0) {
+        vubr_die("posix_fallocate()");
+    }
+
+    addr = mmap(NULL, length, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
+    if (addr == MAP_FAILED) {
+        vubr_die("mmap()");
+    }
+
+    memset(addr, 0xff, length);
+
+    if (pthread_create(&thread, NULL, notifier_thread, &dev->vudev) != 0) {
+        vubr_die("pthread_create()");
+    }
+
+    dev->notifier.fd = fd;
+    dev->notifier.addr = addr;
+    dev->notifier.thread = thread;
+}
+
 static void
 vubr_set_host(struct sockaddr_in *saddr, const char *host)
 {

@@ -673,8 +756,9 @@ main(int argc, char *argv[])
     VubrDev *dev;
     int opt;
     bool client = false;
+    bool host_notifier = false;
 
-    while ((opt = getopt(argc, argv, "l:r:u:c")) != -1) {
+    while ((opt = getopt(argc, argv, "l:r:u:cH")) != -1) {
 
         switch (opt) {
         case 'l':

@@ -693,6 +777,9 @@ main(int argc, char *argv[])
         case 'c':
             client = true;
             break;
+        case 'H':
+            host_notifier = true;
+            break;
         default:
             goto out;
         }

@@ -708,6 +795,10 @@ main(int argc, char *argv[])
         return 1;
     }
 
+    if (host_notifier) {
+        vubr_host_notifier_setup(dev);
+    }
+
     vubr_backend_udp_setup(dev, lhost, lport, rhost, rport);
     vubr_run(dev);
 

@@ -717,7 +808,7 @@ main(int argc, char *argv[])
 
 out:
     fprintf(stderr, "Usage: %s ", argv[0]);
-    fprintf(stderr, "[-c] [-u ud_socket_path] [-l lhost:lport] [-r rhost:rport]\n");
+    fprintf(stderr, "[-c] [-H] [-u ud_socket_path] [-l lhost:lport] [-r rhost:rport]\n");
     fprintf(stderr, "\t-u path to unix doman socket. default: %s\n",
             DEFAULT_UD_SOCKET);
     fprintf(stderr, "\t-l local host and port. default: %s:%s\n",

@@ -725,6 +816,7 @@ out:
     fprintf(stderr, "\t-r remote host and port. default: %s:%s\n",
             DEFAULT_RHOST, DEFAULT_RPORT);
     fprintf(stderr, "\t-c client mode\n");
+    fprintf(stderr, "\t-H use host notifier\n");
 
     return 1;
 }