[dpdk-dev] [PATCH v2] vhost: support inflight share memory protocol feature

[Li Lin <chuckylinchuckylin@gmail.com>]

From: lilin24 <lilin24@baidu.com>

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.

Firstly, qemu uses VHOST_USER_GET_INFLIGHT_FD to get the
shared buffer from backend. Then qemu should send it back
through VHOST_USER_SET_INFLIGHT_FD each time we start vhost-user.

This shared buffer is used to process inflight I/O when backend
reconnect.

Signed-off-by: lilin24 <lilin24@baidu.com>
Signed-off-by: Ni Xun <nixun@baidu.com>
Signed-off-by: Zhang Yu <zhangyu31@baidu.com>
---
 lib/librte_vhost/rte_vhost.h  |  54 +++++++++
 lib/librte_vhost/vhost.c      |  45 ++++++++
 lib/librte_vhost/vhost.h      |  12 ++
 lib/librte_vhost/vhost_user.c | 249 ++++++++++++++++++++++++++++++++++++++++++
 lib/librte_vhost/vhost_user.h |  16 ++-
 5 files changed, 374 insertions(+), 2 deletions(-)

diff --git a/lib/librte_vhost/rte_vhost.h b/lib/librte_vhost/rte_vhost.h
index d2c0c93..73acc79 100644
--- a/lib/librte_vhost/rte_vhost.h
+++ b/lib/librte_vhost/rte_vhost.h
@@ -71,6 +71,10 @@
 #define VHOST_USER_PROTOCOL_F_HOST_NOTIFIER 11
 #endif
 
+#ifndef VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
+#define VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD 12
+#endif
+
 /** Indicate whether protocol features negotiation is supported. */
 #ifndef VHOST_USER_F_PROTOCOL_FEATURES
 #define VHOST_USER_F_PROTOCOL_FEATURES	30
@@ -98,12 +102,27 @@ struct rte_vhost_memory {
 	struct rte_vhost_mem_region regions[];
 };
 
+typedef struct VhostUserInflightEntry {
+	uint8_t inflight;
+} VhostUserInflightEntry;
+
+typedef struct VhostInflightInfo {
+	uint16_t version;
+	uint16_t last_inflight_io;
+	uint16_t used_idx;
+	VhostUserInflightEntry desc[0];
+} VhostInflightInfo;
+
 struct rte_vhost_vring {
 	struct vring_desc	*desc;
 	struct vring_avail	*avail;
 	struct vring_used	*used;
 	uint64_t		log_guest_addr;
 
+	VhostInflightInfo       *inflight;
+	uint16_t                *inflight_reqs;
+	uint16_t                inflight_cnt;
+
 	/** Deprecated, use rte_vhost_vring_call() instead. */
 	int			callfd;
 
@@ -632,6 +651,41 @@ int rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 int rte_vhost_vring_call(int vid, uint16_t vring_idx);
 
 /**
+ * set inflight flag for a entry.
+ *
+ * @param vring
+ *  the structure to hold the requested vring info
+ * @param idx
+ *  inflight entry index
+ */
+void rte_vhost_set_inflight(struct rte_vhost_vring *vring,
+		uint16_t idx);
+
+/**
+ * step 1 of clear inflight flag for a entry.
+ *
+ * @param vring
+ *  the structure to hold the requested vring info
+ * @param idx
+ *  inflight entry index
+ */
+void rte_vhost_clear_inflight_begin(struct rte_vhost_vring *vring,
+		uint16_t idx);
+
+/**
+ * step 2 of clear inflight flag for a entry.
+ *
+ * @param vring
+ *  the structure to hold the requested vring info
+ * @param last_used_idx
+ *  next free used_idx
+ * @param idx
+ *  inflight entry index
+ */
+void rte_vhost_clear_inflight_end(struct rte_vhost_vring *vring,
+		uint16_t last_used_idx, uint16_t idx);
+
+/**
  * Get vhost RX queue avail count.
  *
  * @param vid
diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
index 163f459..ba5537c 100644
--- a/lib/librte_vhost/vhost.c
+++ b/lib/librte_vhost/vhost.c
@@ -76,6 +76,8 @@
 		close(vq->callfd);
 	if (vq->kickfd >= 0)
 		close(vq->kickfd);
+	if (vq->inflight)
+		vq->inflight = NULL;
 }
 
 /*
@@ -589,6 +591,14 @@
 	vring->kickfd  = vq->kickfd;
 	vring->size    = vq->size;
 
+	vring->inflight = vq->inflight;
+
+	vring->inflight_reqs = vq->inflight_reqs;
+	vq->inflight_reqs = NULL;
+
+	vring->inflight_cnt = vq->inflight_cnt;
+	vq->inflight_cnt = 0;
+
 	return 0;
 }
 
@@ -617,6 +627,41 @@
 	return 0;
 }
 
+void
+rte_vhost_set_inflight(struct rte_vhost_vring *vring, uint16_t idx)
+{
+	VhostInflightInfo *inflight = vring->inflight;
+	if (unlikely(!inflight))
+		return;
+	inflight->desc[idx].inflight = 1;
+}
+
+void
+rte_vhost_clear_inflight_begin(struct rte_vhost_vring *vring, uint16_t idx)
+{
+	VhostInflightInfo *inflight = vring->inflight;
+
+	if (unlikely(!inflight))
+		return;
+
+	inflight->last_inflight_io = idx;
+}
+
+void
+rte_vhost_clear_inflight_end(struct rte_vhost_vring *vring,
+	uint16_t last_used_idx, uint16_t idx)
+{
+	VhostInflightInfo *inflight = vring->inflight;
+
+	if (unlikely(!inflight))
+		return;
+
+	rte_compiler_barrier();
+	inflight->desc[idx].inflight = 0;
+	rte_compiler_barrier();
+	inflight->used_idx = last_used_idx;
+}
+
 uint16_t
 rte_vhost_avail_entries(int vid, uint16_t queue_id)
 {
diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index e9138df..2ab9d68 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -128,6 +128,11 @@ struct vhost_virtqueue {
 	/* Physical address of used ring, for logging */
 	uint64_t		log_guest_addr;
 
+	/* Inflight share memory info */
+	VhostInflightInfo       *inflight;
+	uint16_t                *inflight_reqs;
+	uint16_t                inflight_cnt;
+
 	uint16_t		nr_zmbuf;
 	uint16_t		zmbuf_size;
 	uint16_t		last_zmbuf_idx;
@@ -286,6 +291,12 @@ struct guest_page {
 	uint64_t size;
 };
 
+typedef struct VuDevInflightInfo {
+	int fd;
+	void *addr;
+	uint64_t size;
+} VuDevInflightInfo;
+
 /**
  * Device structure contains all configuration information relating
  * to the device.
@@ -303,6 +314,7 @@ struct virtio_net {
 	uint32_t		nr_vring;
 	int			dequeue_zero_copy;
 	struct vhost_virtqueue	*virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
+	VuDevInflightInfo inflight_info;
 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 	char			ifname[IF_NAME_SZ];
 	uint64_t		log_size;
diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c
index c9e29ec..ca937e2 100644
--- a/lib/librte_vhost/vhost_user.c
+++ b/lib/librte_vhost/vhost_user.c
@@ -31,6 +31,8 @@
 #include <sys/stat.h>
 #include <sys/syscall.h>
 #include <assert.h>
+#include <sys/syscall.h>
+#include <asm/unistd.h>
 #ifdef RTE_LIBRTE_VHOST_NUMA
 #include <numaif.h>
 #endif
@@ -49,6 +51,14 @@
 #define VIRTIO_MIN_MTU 68
 #define VIRTIO_MAX_MTU 65535
 
+#define INFLIGHT_ALIGNMENT 64
+#define INFLIGHT_VERSION 0xabcd
+
+#define CLOEXEC 0x0001U
+
+#define ALIGN_DOWN(n, m) ((n) / (m) * (m))
+#define ALIGN_UP(n, m) ALIGN_DOWN((n) + (m) - 1, (m))
+
 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 +88,8 @@
 	[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 +172,16 @@
 		dev->log_addr = 0;
 	}
 
+	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;
+	}
+
 	if (dev->slave_req_fd >= 0) {
 		close(dev->slave_req_fd);
 		dev->slave_req_fd = -1;
@@ -1165,6 +1187,184 @@
 	return 1;
 }
 
+static int mem_create(const char *name, unsigned int flags)
+{
+#ifdef __NR_memfd_create
+	return syscall(__NR_memfd_create, name, flags);
+#else
+	return -1;
+#endif
+}
+
+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;
+	mfd = mem_create(name, CLOEXEC);
+	if (mfd != -1) {
+		if (ftruncate(mfd, size) == -1) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+					"ftruncate fail for alloc inflight buffer\n");
+			close(mfd);
+			return NULL;
+		}
+	} else {
+		mfd = mkstemp(fname);
+		unlink(fname);
+
+		if (mfd == -1) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+					"mkstemp fail for alloc inflight buffer\n");
+			return NULL;
+		}
+
+		if (ftruncate(mfd, size) == -1) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+					"ftruncate fail for 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,
+				"mmap fail for alloc inflight buffer\n");
+		close(mfd);
+		return NULL;
+	}
+
+	*fd = mfd;
+	return ptr;
+}
+
+static uint32_t get_pervq_shm_size(uint16_t queue_size)
+{
+	return ALIGN_UP(sizeof(VhostUserInflightEntry) * queue_size +
+		sizeof(uint16_t) * 3, INFLIGHT_ALIGNMENT);
+}
+
+static int
+vhost_user_get_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
+		int main_fd __rte_unused)
+{
+	int fd;
+	uint64_t mmap_size;
+	void *addr;
+	uint16_t num_queues, queue_size;
+	struct virtio_net *dev = *pdev;
+
+	if (msg->size != sizeof(msg->payload.inflight)) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"Invalid get_inflight_fd message size is %d",
+			msg->size);
+		msg->payload.inflight.mmap_size = 0;
+		return 0;
+	}
+
+	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);
+	mmap_size = num_queues * get_pervq_shm_size(queue_size);
+
+	addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
+	if (!addr) {
+		RTE_LOG(ERR, VHOST_CONFIG, "Failed to alloc vhost inflight area");
+			msg->payload.inflight.mmap_size = 0;
+		return 0;
+	}
+	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;
+
+	RTE_LOG(INFO, VHOST_CONFIG,
+			"send inflight mmap_size: %lu\n",
+			msg->payload.inflight.mmap_size);
+	RTE_LOG(INFO, VHOST_CONFIG,
+			"send inflight mmap_offset: %lu\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)
+{
+	int fd, i;
+	uint64_t mmap_size, mmap_offset;
+	uint16_t num_queues, queue_size;
+	uint32_t pervq_inflight_size;
+	void *rc;
+	struct vhost_virtqueue *vq;
+	struct virtio_net *dev = *pdev;
+
+	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 -1;
+	}
+
+	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;
+	pervq_inflight_size = get_pervq_shm_size(queue_size);
+
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd mmap_size: %lu\n", mmap_size);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd mmap_offset: %lu\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.addr)
+		munmap(dev->inflight_info.addr, dev->inflight_info.size);
+
+	rc = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
+			fd, mmap_offset);
+	if (rc == MAP_FAILED) {
+		RTE_LOG(ERR, VHOST_CONFIG, "failed to mmap share memory.\n");
+		return -1;
+	}
+
+	if (dev->inflight_info.fd)
+		close(dev->inflight_info.fd);
+
+	dev->inflight_info.fd = fd;
+	dev->inflight_info.addr = rc;
+	dev->inflight_info.size = mmap_size;
+
+	for (i = 0; i < num_queues; i++) {
+		vq = dev->virtqueue[i];
+		vq->inflight = (VhostInflightInfo *)rc;
+		rc = (void *)((char *)rc + pervq_inflight_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)
@@ -1202,6 +1402,47 @@ static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
 }
 
 static int
+vhost_check_queue_inflights(struct vhost_virtqueue *vq)
+{
+	struct vring_used *used = vq->used;
+	uint16_t i = 0;
+
+	if ((!vq->inflight))
+		return RTE_VHOST_MSG_RESULT_ERR;
+
+	if (!vq->inflight->version) {
+		vq->inflight->version = INFLIGHT_VERSION;
+		return RTE_VHOST_MSG_RESULT_OK;
+	}
+
+	vq->inflight_reqs = calloc(vq->size, sizeof(uint16_t));
+	if (!vq->inflight_reqs) {
+		RTE_LOG(ERR, VHOST_CONFIG, "Failed to allocate memory for inflight reqs.\n");
+		return RTE_VHOST_MSG_RESULT_ERR;
+	}
+
+	if (vq->inflight->used_idx != used->idx) {
+		vq->inflight->desc[vq->inflight->last_inflight_io].inflight = 0;
+		rte_compiler_barrier();
+		vq->inflight->used_idx = used->idx;
+	}
+
+	for (i = 0; i < vq->size; i++) {
+		if (vq->inflight->desc[i].inflight == 1)
+			vq->inflight_reqs[vq->inflight_cnt++] = i;
+	}
+
+	if (!vq->inflight_cnt) {
+		free(vq->inflight_reqs);
+		vq->inflight_reqs = NULL;
+	}
+
+	vq->last_avail_idx += vq->inflight_cnt;
+
+	return RTE_VHOST_MSG_RESULT_OK;
+}
+
+static int
 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
 			int main_fd __rte_unused)
 {
@@ -1242,6 +1483,12 @@ static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
 		close(vq->kickfd);
 	vq->kickfd = file.fd;
 
+	if (vhost_check_queue_inflights(vq)) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"Failed to inflights for vq: %d\n", file.index);
+		return RTE_VHOST_MSG_RESULT_ERR;
+	}
+
 	return RTE_VHOST_MSG_RESULT_OK;
 }
 
@@ -1762,6 +2009,8 @@ typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
 	[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,
 };
 
 
diff --git a/lib/librte_vhost/vhost_user.h b/lib/librte_vhost/vhost_user.h
index 2a650fe..35f969b 100644
--- a/lib/librte_vhost/vhost_user.h
+++ b/lib/librte_vhost/vhost_user.h
@@ -23,7 +23,8 @@
 					 (1ULL << VHOST_USER_PROTOCOL_F_CRYPTO_SESSION) | \
 					 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD) | \
 					 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER) | \
-					 (1ULL << VHOST_USER_PROTOCOL_F_PAGEFAULT))
+					(1ULL << VHOST_USER_PROTOCOL_F_PAGEFAULT) | \
+					(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))
 
 typedef enum VhostUserRequest {
 	VHOST_USER_NONE = 0,
@@ -54,7 +55,9 @@
 	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 {
@@ -112,6 +115,13 @@
 	uint64_t offset;
 } VhostUserVringArea;
 
+typedef struct VhostUserInflight {
+	uint64_t mmap_size;
+	uint64_t mmap_offset;
+	uint16_t num_queues;
+	uint16_t queue_size;
+} VhostUserInflight;
+
 typedef struct VhostUserMsg {
 	union {
 		uint32_t master; /* a VhostUserRequest value */
@@ -131,6 +141,7 @@
 		struct vhost_vring_addr addr;
 		VhostUserMemory memory;
 		VhostUserLog    log;
+		VhostUserInflight inflight;
 		struct vhost_iotlb_msg iotlb;
 		VhostUserCryptoSessionParam crypto_session;
 		VhostUserVringArea area;
@@ -148,6 +159,7 @@
 /* vhost_user.c */
 int vhost_user_msg_handler(int vid, int fd);
 int vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm);
+void *inflight_mem_alloc(const char *name, size_t size, int *fd);
 
 /* socket.c */
 int read_fd_message(int sockfd, char *buf, int buflen, int *fds, int max_fds,
-- 
1.8.3.1