1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
5 #include <sys/socket.h>
15 #include "virtio_user_dev.h"
17 /* The version of the protocol we support */
18 #define VHOST_USER_VERSION 0x1
20 #define VHOST_MEMORY_MAX_NREGIONS 8
24 struct vhost_memory_region regions[VHOST_MEMORY_MAX_NREGIONS];
27 struct vhost_user_msg {
28 enum vhost_user_request request;
30 #define VHOST_USER_VERSION_MASK 0x3
31 #define VHOST_USER_REPLY_MASK (0x1 << 2)
33 uint32_t size; /* the following payload size */
35 #define VHOST_USER_VRING_IDX_MASK 0xff
36 #define VHOST_USER_VRING_NOFD_MASK (0x1 << 8)
38 struct vhost_vring_state state;
39 struct vhost_vring_addr addr;
40 struct vhost_memory memory;
42 int fds[VHOST_MEMORY_MAX_NREGIONS];
43 } __attribute((packed));
45 #define VHOST_USER_HDR_SIZE offsetof(struct vhost_user_msg, payload.u64)
46 #define VHOST_USER_PAYLOAD_SIZE \
47 (sizeof(struct vhost_user_msg) - VHOST_USER_HDR_SIZE)
50 vhost_user_write(int fd, void *buf, int len, int *fds, int fd_num)
55 size_t fd_size = fd_num * sizeof(int);
56 char control[CMSG_SPACE(fd_size)];
59 memset(&msgh, 0, sizeof(msgh));
60 memset(control, 0, sizeof(control));
62 iov.iov_base = (uint8_t *)buf;
67 msgh.msg_control = control;
68 msgh.msg_controllen = sizeof(control);
70 cmsg = CMSG_FIRSTHDR(&msgh);
71 cmsg->cmsg_len = CMSG_LEN(fd_size);
72 cmsg->cmsg_level = SOL_SOCKET;
73 cmsg->cmsg_type = SCM_RIGHTS;
74 memcpy(CMSG_DATA(cmsg), fds, fd_size);
77 r = sendmsg(fd, &msgh, 0);
78 } while (r < 0 && errno == EINTR);
84 vhost_user_read(int fd, struct vhost_user_msg *msg)
86 uint32_t valid_flags = VHOST_USER_REPLY_MASK | VHOST_USER_VERSION;
87 int ret, sz_hdr = VHOST_USER_HDR_SIZE, sz_payload;
89 ret = recv(fd, (void *)msg, sz_hdr, 0);
91 PMD_DRV_LOG(ERR, "Failed to recv msg hdr: %d instead of %d.",
96 /* validate msg flags */
97 if (msg->flags != (valid_flags)) {
98 PMD_DRV_LOG(ERR, "Failed to recv msg: flags %x instead of %x.",
99 msg->flags, valid_flags);
103 sz_payload = msg->size;
105 if ((size_t)sz_payload > sizeof(msg->payload))
109 ret = recv(fd, (void *)((char *)msg + sz_hdr), sz_payload, 0);
110 if (ret < sz_payload) {
112 "Failed to recv msg payload: %d instead of %d.",
124 struct hugepage_file_info {
125 uint64_t addr; /**< virtual addr */
126 size_t size; /**< the file size */
127 char path[PATH_MAX]; /**< path to backing file */
130 /* Two possible options:
131 * 1. Match HUGEPAGE_INFO_FMT to find the file storing struct hugepage_file
132 * array. This is simple but cannot be used in secondary process because
133 * secondary process will close and munmap that file.
134 * 2. Match HUGEFILE_FMT to find hugepage files directly.
136 * We choose option 2.
139 get_hugepage_file_info(struct hugepage_file_info huges[], int max)
143 char buf[BUFSIZ], *tmp, *tail;
144 char *str_underline, *str_start;
146 uint64_t v_start, v_end;
149 f = fopen("/proc/self/maps", "r");
151 PMD_DRV_LOG(ERR, "cannot open /proc/self/maps");
156 while (fgets(buf, sizeof(buf), f) != NULL) {
157 if (sscanf(buf, "%" PRIx64 "-%" PRIx64, &v_start, &v_end) < 2) {
158 PMD_DRV_LOG(ERR, "Failed to parse address");
162 tmp = strchr(buf, ' ') + 1; /** skip address */
163 tmp = strchr(tmp, ' ') + 1; /** skip perm */
164 tmp = strchr(tmp, ' ') + 1; /** skip offset */
165 tmp = strchr(tmp, ' ') + 1; /** skip dev */
166 tmp = strchr(tmp, ' ') + 1; /** skip inode */
167 while (*tmp == ' ') /** skip spaces */
169 tail = strrchr(tmp, '\n'); /** remove newline if exists */
173 /* Match HUGEFILE_FMT, aka "%s/%smap_%d",
174 * which is defined in eal_filesystem.h
176 str_underline = strrchr(tmp, '_');
180 str_start = str_underline - strlen("map");
184 if (sscanf(str_start, "map_%d", &huge_index) != 1)
187 /* skip duplicated file which is mapped to different regions */
188 for (k = 0, exist = -1; k < idx; ++k) {
189 if (!strcmp(huges[k].path, tmp)) {
198 PMD_DRV_LOG(ERR, "Exceed maximum of %d", max);
202 huges[idx].addr = v_start;
203 huges[idx].size = v_end - v_start; /* To be corrected later */
204 snprintf(huges[idx].path, PATH_MAX, "%s", tmp);
208 /* correct the size for files who have many regions */
209 for (k = 0; k < idx; ++k) {
210 if (stat(huges[k].path, &stats) < 0) {
211 PMD_DRV_LOG(ERR, "Failed to stat %s, %s\n",
212 huges[k].path, strerror(errno));
215 huges[k].size = stats.st_size;
216 PMD_DRV_LOG(INFO, "file %s, size %zx\n",
217 huges[k].path, huges[k].size);
229 prepare_vhost_memory_user(struct vhost_user_msg *msg, int fds[])
232 struct hugepage_file_info huges[VHOST_MEMORY_MAX_NREGIONS];
233 struct vhost_memory_region *mr;
235 num = get_hugepage_file_info(huges, VHOST_MEMORY_MAX_NREGIONS);
237 PMD_INIT_LOG(ERR, "Failed to prepare memory for vhost-user");
241 for (i = 0; i < num; ++i) {
242 mr = &msg->payload.memory.regions[i];
243 mr->guest_phys_addr = huges[i].addr; /* use vaddr! */
244 mr->userspace_addr = huges[i].addr;
245 mr->memory_size = huges[i].size;
247 fds[i] = open(huges[i].path, O_RDWR);
250 msg->payload.memory.nregions = num;
251 msg->payload.memory.padding = 0;
256 static struct vhost_user_msg m;
258 const char * const vhost_msg_strings[] = {
259 [VHOST_USER_SET_OWNER] = "VHOST_SET_OWNER",
260 [VHOST_USER_RESET_OWNER] = "VHOST_RESET_OWNER",
261 [VHOST_USER_SET_FEATURES] = "VHOST_SET_FEATURES",
262 [VHOST_USER_GET_FEATURES] = "VHOST_GET_FEATURES",
263 [VHOST_USER_SET_VRING_CALL] = "VHOST_SET_VRING_CALL",
264 [VHOST_USER_SET_VRING_NUM] = "VHOST_SET_VRING_NUM",
265 [VHOST_USER_SET_VRING_BASE] = "VHOST_SET_VRING_BASE",
266 [VHOST_USER_GET_VRING_BASE] = "VHOST_GET_VRING_BASE",
267 [VHOST_USER_SET_VRING_ADDR] = "VHOST_SET_VRING_ADDR",
268 [VHOST_USER_SET_VRING_KICK] = "VHOST_SET_VRING_KICK",
269 [VHOST_USER_SET_MEM_TABLE] = "VHOST_SET_MEM_TABLE",
270 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_SET_VRING_ENABLE",
274 vhost_user_sock(struct virtio_user_dev *dev,
275 enum vhost_user_request req,
278 struct vhost_user_msg msg;
279 struct vhost_vring_file *file = 0;
281 int fds[VHOST_MEMORY_MAX_NREGIONS];
284 int vhostfd = dev->vhostfd;
288 PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
290 if (dev->is_server && vhostfd < 0)
294 msg.flags = VHOST_USER_VERSION;
298 case VHOST_USER_GET_FEATURES:
302 case VHOST_USER_SET_FEATURES:
303 case VHOST_USER_SET_LOG_BASE:
304 msg.payload.u64 = *((__u64 *)arg);
305 msg.size = sizeof(m.payload.u64);
308 case VHOST_USER_SET_OWNER:
309 case VHOST_USER_RESET_OWNER:
312 case VHOST_USER_SET_MEM_TABLE:
313 if (prepare_vhost_memory_user(&msg, fds) < 0)
315 fd_num = msg.payload.memory.nregions;
316 msg.size = sizeof(m.payload.memory.nregions);
317 msg.size += sizeof(m.payload.memory.padding);
318 msg.size += fd_num * sizeof(struct vhost_memory_region);
321 case VHOST_USER_SET_LOG_FD:
322 fds[fd_num++] = *((int *)arg);
325 case VHOST_USER_SET_VRING_NUM:
326 case VHOST_USER_SET_VRING_BASE:
327 case VHOST_USER_SET_VRING_ENABLE:
328 memcpy(&msg.payload.state, arg, sizeof(msg.payload.state));
329 msg.size = sizeof(m.payload.state);
332 case VHOST_USER_GET_VRING_BASE:
333 memcpy(&msg.payload.state, arg, sizeof(msg.payload.state));
334 msg.size = sizeof(m.payload.state);
338 case VHOST_USER_SET_VRING_ADDR:
339 memcpy(&msg.payload.addr, arg, sizeof(msg.payload.addr));
340 msg.size = sizeof(m.payload.addr);
343 case VHOST_USER_SET_VRING_KICK:
344 case VHOST_USER_SET_VRING_CALL:
345 case VHOST_USER_SET_VRING_ERR:
347 msg.payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
348 msg.size = sizeof(m.payload.u64);
350 fds[fd_num++] = file->fd;
352 msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
356 PMD_DRV_LOG(ERR, "trying to send unhandled msg type");
360 len = VHOST_USER_HDR_SIZE + msg.size;
361 if (vhost_user_write(vhostfd, &msg, len, fds, fd_num) < 0) {
362 PMD_DRV_LOG(ERR, "%s failed: %s",
363 vhost_msg_strings[req], strerror(errno));
367 if (req == VHOST_USER_SET_MEM_TABLE)
368 for (i = 0; i < fd_num; ++i)
372 if (vhost_user_read(vhostfd, &msg) < 0) {
373 PMD_DRV_LOG(ERR, "Received msg failed: %s",
378 if (req != msg.request) {
379 PMD_DRV_LOG(ERR, "Received unexpected msg type");
384 case VHOST_USER_GET_FEATURES:
385 if (msg.size != sizeof(m.payload.u64)) {
386 PMD_DRV_LOG(ERR, "Received bad msg size");
389 *((__u64 *)arg) = msg.payload.u64;
391 case VHOST_USER_GET_VRING_BASE:
392 if (msg.size != sizeof(m.payload.state)) {
393 PMD_DRV_LOG(ERR, "Received bad msg size");
396 memcpy(arg, &msg.payload.state,
397 sizeof(struct vhost_vring_state));
400 PMD_DRV_LOG(ERR, "Received unexpected msg type");
408 #define MAX_VIRTIO_USER_BACKLOG 1
410 virtio_user_start_server(struct virtio_user_dev *dev, struct sockaddr_un *un)
414 int fd = dev->listenfd;
416 ret = bind(fd, (struct sockaddr *)un, sizeof(*un));
418 PMD_DRV_LOG(ERR, "failed to bind to %s: %s; remove it and try again\n",
419 dev->path, strerror(errno));
422 ret = listen(fd, MAX_VIRTIO_USER_BACKLOG);
426 flag = fcntl(fd, F_GETFL);
427 fcntl(fd, F_SETFL, flag | O_NONBLOCK);
433 * Set up environment to talk with a vhost user backend.
440 vhost_user_setup(struct virtio_user_dev *dev)
444 struct sockaddr_un un;
446 fd = socket(AF_UNIX, SOCK_STREAM, 0);
448 PMD_DRV_LOG(ERR, "socket() error, %s", strerror(errno));
452 flag = fcntl(fd, F_GETFD);
453 if (fcntl(fd, F_SETFD, flag | FD_CLOEXEC) < 0)
454 PMD_DRV_LOG(WARNING, "fcntl failed, %s", strerror(errno));
456 memset(&un, 0, sizeof(un));
457 un.sun_family = AF_UNIX;
458 snprintf(un.sun_path, sizeof(un.sun_path), "%s", dev->path);
460 if (dev->is_server) {
462 if (virtio_user_start_server(dev, &un) < 0) {
463 PMD_DRV_LOG(ERR, "virtio-user startup fails in server mode");
469 if (connect(fd, (struct sockaddr *)&un, sizeof(un)) < 0) {
470 PMD_DRV_LOG(ERR, "connect error, %s", strerror(errno));
481 vhost_user_enable_queue_pair(struct virtio_user_dev *dev,
487 for (i = 0; i < 2; ++i) {
488 struct vhost_vring_state state = {
489 .index = pair_idx * 2 + i,
493 if (vhost_user_sock(dev, VHOST_USER_SET_VRING_ENABLE, &state))
500 struct virtio_user_backend_ops ops_user = {
501 .setup = vhost_user_setup,
502 .send_request = vhost_user_sock,
503 .enable_qp = vhost_user_enable_queue_pair