vhost: support inflight info sharing
authorJin Yu <jin.yu@intel.com>
Wed, 9 Oct 2019 20:48:32 +0000 (04:48 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Wed, 23 Oct 2019 14:43:09 +0000 (16:43 +0200)
This patch introduces two new messages VHOST_USER_GET_INFLIGHT_FD
and VHOST_USER_SET_INFLIGHT_FD to support transferring a shared
buffer between qemu and backend.

Signed-off-by: Lin Li <lilin24@baidu.com>
Signed-off-by: Xun Ni <nixun@baidu.com>
Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
Signed-off-by: Jin Yu <jin.yu@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
lib/librte_vhost/vhost.h
lib/librte_vhost/vhost_user.c
lib/librte_vhost/vhost_user.h

index cfea138..4ae47c1 100644 (file)
@@ -286,6 +286,12 @@ struct guest_page {
        uint64_t size;
 };
 
+struct inflight_mem_info {
+       int             fd;
+       void            *addr;
+       uint64_t        size;
+};
+
 /**
  * Device structure contains all configuration information relating
  * to the device.
@@ -303,6 +309,7 @@ struct virtio_net {
        uint32_t                nr_vring;
        int                     dequeue_zero_copy;
        struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
+       struct inflight_mem_info *inflight_info;
 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
        char                    ifname[IF_NAME_SZ];
        uint64_t                log_size;
index 5ae561e..c168c15 100644 (file)
 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
 #include <linux/userfaultfd.h>
 #endif
+#ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
+#include <linux/memfd.h>
+#define MEMFD_SUPPORTED
+#endif
 
 #include <rte_common.h>
 #include <rte_malloc.h>
@@ -49,6 +53,9 @@
 #define VIRTIO_MIN_MTU 68
 #define VIRTIO_MAX_MTU 65535
 
+#define INFLIGHT_ALIGNMENT     64
+#define INFLIGHT_VERSION       0x1
+
 static const char *vhost_message_str[VHOST_USER_MAX] = {
        [VHOST_USER_NONE] = "VHOST_USER_NONE",
        [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
@@ -78,6 +85,8 @@ static const char *vhost_message_str[VHOST_USER_MAX] = {
        [VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
        [VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
        [VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
+       [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
+       [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
 };
 
 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
@@ -160,6 +169,22 @@ vhost_backend_cleanup(struct virtio_net *dev)
                dev->log_addr = 0;
        }
 
+       if (dev->inflight_info) {
+               if (dev->inflight_info->addr) {
+                       munmap(dev->inflight_info->addr,
+                              dev->inflight_info->size);
+                       dev->inflight_info->addr = NULL;
+               }
+
+               if (dev->inflight_info->fd > 0) {
+                       close(dev->inflight_info->fd);
+                       dev->inflight_info->fd = -1;
+               }
+
+               free(dev->inflight_info);
+               dev->inflight_info = NULL;
+       }
+
        if (dev->slave_req_fd >= 0) {
                close(dev->slave_req_fd);
                dev->slave_req_fd = -1;
@@ -1218,6 +1243,221 @@ virtio_is_ready(struct virtio_net *dev)
        return 1;
 }
 
+static void *
+inflight_mem_alloc(const char *name, size_t size, int *fd)
+{
+       void *ptr;
+       int mfd = -1;
+       char fname[20] = "/tmp/memfd-XXXXXX";
+
+       *fd = -1;
+#ifdef MEMFD_SUPPORTED
+       mfd = memfd_create(name, MFD_CLOEXEC);
+#else
+       RTE_SET_USED(name);
+#endif
+       if (mfd == -1) {
+               mfd = mkstemp(fname);
+               if (mfd == -1) {
+                       RTE_LOG(ERR, VHOST_CONFIG,
+                               "failed to get inflight buffer fd\n");
+                       return NULL;
+               }
+
+               unlink(fname);
+       }
+
+       if (ftruncate(mfd, size) == -1) {
+               RTE_LOG(ERR, VHOST_CONFIG,
+                       "failed to alloc inflight buffer\n");
+               close(mfd);
+               return NULL;
+       }
+
+       ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
+       if (ptr == MAP_FAILED) {
+               RTE_LOG(ERR, VHOST_CONFIG,
+                       "failed to mmap inflight buffer\n");
+               close(mfd);
+               return NULL;
+       }
+
+       *fd = mfd;
+       return ptr;
+}
+
+static uint32_t
+get_pervq_shm_size_split(uint16_t queue_size)
+{
+       return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
+                                 queue_size + sizeof(uint64_t) +
+                                 sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
+}
+
+static uint32_t
+get_pervq_shm_size_packed(uint16_t queue_size)
+{
+       return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
+                                 * queue_size + sizeof(uint64_t) +
+                                 sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
+                                 INFLIGHT_ALIGNMENT);
+}
+
+static int
+vhost_user_get_inflight_fd(struct virtio_net **pdev,
+                          VhostUserMsg *msg,
+                          int main_fd __rte_unused)
+{
+       struct rte_vhost_inflight_info_packed *inflight_packed;
+       uint64_t pervq_inflight_size, mmap_size;
+       uint16_t num_queues, queue_size;
+       struct virtio_net *dev = *pdev;
+       int fd, i, j;
+       void *addr;
+
+       if (msg->size != sizeof(msg->payload.inflight)) {
+               RTE_LOG(ERR, VHOST_CONFIG,
+                       "invalid get_inflight_fd message size is %d\n",
+                       msg->size);
+               return RTE_VHOST_MSG_RESULT_ERR;
+       }
+
+       if (dev->inflight_info == NULL) {
+               dev->inflight_info = calloc(1,
+                                           sizeof(struct inflight_mem_info));
+               if (!dev->inflight_info) {
+                       RTE_LOG(ERR, VHOST_CONFIG,
+                               "failed to alloc dev inflight area\n");
+                       return RTE_VHOST_MSG_RESULT_ERR;
+               }
+       }
+
+       num_queues = msg->payload.inflight.num_queues;
+       queue_size = msg->payload.inflight.queue_size;
+
+       RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd num_queues: %u\n",
+               msg->payload.inflight.num_queues);
+       RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd queue_size: %u\n",
+               msg->payload.inflight.queue_size);
+
+       if (vq_is_packed(dev))
+               pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
+       else
+               pervq_inflight_size = get_pervq_shm_size_split(queue_size);
+
+       mmap_size = num_queues * pervq_inflight_size;
+       addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
+       if (!addr) {
+               RTE_LOG(ERR, VHOST_CONFIG,
+                       "failed to alloc vhost inflight area\n");
+                       msg->payload.inflight.mmap_size = 0;
+               return RTE_VHOST_MSG_RESULT_ERR;
+       }
+       memset(addr, 0, mmap_size);
+
+       dev->inflight_info->addr = addr;
+       dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
+       dev->inflight_info->fd = msg->fds[0] = fd;
+       msg->payload.inflight.mmap_offset = 0;
+       msg->fd_num = 1;
+
+       if (vq_is_packed(dev)) {
+               for (i = 0; i < num_queues; i++) {
+                       inflight_packed =
+                               (struct rte_vhost_inflight_info_packed *)addr;
+                       inflight_packed->used_wrap_counter = 1;
+                       inflight_packed->old_used_wrap_counter = 1;
+                       for (j = 0; j < queue_size; j++)
+                               inflight_packed->desc[j].next = j + 1;
+                       addr = (void *)((char *)addr + pervq_inflight_size);
+               }
+       }
+
+       RTE_LOG(INFO, VHOST_CONFIG,
+               "send inflight mmap_size: %"PRIu64"\n",
+               msg->payload.inflight.mmap_size);
+       RTE_LOG(INFO, VHOST_CONFIG,
+               "send inflight mmap_offset: %"PRIu64"\n",
+               msg->payload.inflight.mmap_offset);
+       RTE_LOG(INFO, VHOST_CONFIG,
+               "send inflight fd: %d\n", msg->fds[0]);
+
+       return RTE_VHOST_MSG_RESULT_REPLY;
+}
+
+static int
+vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
+                          int main_fd __rte_unused)
+{
+       uint64_t mmap_size, mmap_offset;
+       uint16_t num_queues, queue_size;
+       uint32_t pervq_inflight_size;
+       struct virtio_net *dev = *pdev;
+       void *addr;
+       int fd;
+
+       fd = msg->fds[0];
+       if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
+               RTE_LOG(ERR, VHOST_CONFIG,
+                       "invalid set_inflight_fd message size is %d,fd is %d\n",
+                       msg->size, fd);
+               return RTE_VHOST_MSG_RESULT_ERR;
+       }
+
+       mmap_size = msg->payload.inflight.mmap_size;
+       mmap_offset = msg->payload.inflight.mmap_offset;
+       num_queues = msg->payload.inflight.num_queues;
+       queue_size = msg->payload.inflight.queue_size;
+
+       if (vq_is_packed(dev))
+               pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
+       else
+               pervq_inflight_size = get_pervq_shm_size_split(queue_size);
+
+       RTE_LOG(INFO, VHOST_CONFIG,
+               "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
+       RTE_LOG(INFO, VHOST_CONFIG,
+               "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
+       RTE_LOG(INFO, VHOST_CONFIG,
+               "set_inflight_fd num_queues: %u\n", num_queues);
+       RTE_LOG(INFO, VHOST_CONFIG,
+               "set_inflight_fd queue_size: %u\n", queue_size);
+       RTE_LOG(INFO, VHOST_CONFIG,
+               "set_inflight_fd fd: %d\n", fd);
+       RTE_LOG(INFO, VHOST_CONFIG,
+               "set_inflight_fd pervq_inflight_size: %d\n",
+               pervq_inflight_size);
+
+       if (!dev->inflight_info) {
+               dev->inflight_info = calloc(1,
+                                           sizeof(struct inflight_mem_info));
+               if (dev->inflight_info == NULL) {
+                       RTE_LOG(ERR, VHOST_CONFIG,
+                               "failed to alloc dev inflight area\n");
+                       return RTE_VHOST_MSG_RESULT_ERR;
+               }
+       }
+
+       if (dev->inflight_info->addr)
+               munmap(dev->inflight_info->addr, dev->inflight_info->size);
+
+       addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
+                   fd, mmap_offset);
+       if (addr == MAP_FAILED) {
+               RTE_LOG(ERR, VHOST_CONFIG, "failed to mmap share memory.\n");
+               return RTE_VHOST_MSG_RESULT_ERR;
+       }
+
+       if (dev->inflight_info->fd)
+               close(dev->inflight_info->fd);
+
+       dev->inflight_info->fd = fd;
+       dev->inflight_info->addr = addr;
+       dev->inflight_info->size = mmap_size;
+
+       return RTE_VHOST_MSG_RESULT_OK;
+}
+
 static int
 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
                        int main_fd __rte_unused)
@@ -1835,9 +2075,10 @@ static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
        [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
        [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
        [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
+       [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
+       [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
 };
 
-
 /* return bytes# of read on success or negative val on failure. */
 static int
 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
index 17a1d7b..6563f73 100644 (file)
@@ -54,7 +54,9 @@ typedef enum VhostUserRequest {
        VHOST_USER_POSTCOPY_ADVISE = 28,
        VHOST_USER_POSTCOPY_LISTEN = 29,
        VHOST_USER_POSTCOPY_END = 30,
-       VHOST_USER_MAX = 31
+       VHOST_USER_GET_INFLIGHT_FD = 31,
+       VHOST_USER_SET_INFLIGHT_FD = 32,
+       VHOST_USER_MAX = 33
 } VhostUserRequest;
 
 typedef enum VhostUserSlaveRequest {