net/virtio: add virtio-user status ops
[dpdk.git] / drivers / net / virtio / virtio_user / vhost_user.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4
5 #include <sys/socket.h>
6 #include <sys/types.h>
7 #include <sys/stat.h>
8 #include <unistd.h>
9 #include <fcntl.h>
10 #include <sys/un.h>
11 #include <string.h>
12 #include <errno.h>
13
14 #include <rte_string_fns.h>
15 #include <rte_fbarray.h>
16
17 #include "vhost.h"
18 #include "virtio_user_dev.h"
19
20 /* The version of the protocol we support */
21 #define VHOST_USER_VERSION    0x1
22
23 #define VHOST_MEMORY_MAX_NREGIONS 8
24 struct vhost_memory {
25         uint32_t nregions;
26         uint32_t padding;
27         struct vhost_memory_region regions[VHOST_MEMORY_MAX_NREGIONS];
28 };
29
30 struct vhost_user_msg {
31         enum vhost_user_request request;
32
33 #define VHOST_USER_VERSION_MASK     0x3
34 #define VHOST_USER_REPLY_MASK       (0x1 << 2)
35 #define VHOST_USER_NEED_REPLY_MASK  (0x1 << 3)
36         uint32_t flags;
37         uint32_t size; /* the following payload size */
38         union {
39 #define VHOST_USER_VRING_IDX_MASK   0xff
40 #define VHOST_USER_VRING_NOFD_MASK  (0x1 << 8)
41                 uint64_t u64;
42                 struct vhost_vring_state state;
43                 struct vhost_vring_addr addr;
44                 struct vhost_memory memory;
45         } payload;
46         int fds[VHOST_MEMORY_MAX_NREGIONS];
47 } __rte_packed;
48
49 #define VHOST_USER_HDR_SIZE offsetof(struct vhost_user_msg, payload.u64)
50 #define VHOST_USER_PAYLOAD_SIZE \
51         (sizeof(struct vhost_user_msg) - VHOST_USER_HDR_SIZE)
52
53 static int
54 vhost_user_write(int fd, struct vhost_user_msg *msg, int *fds, int fd_num)
55 {
56         int r;
57         struct msghdr msgh;
58         struct iovec iov;
59         size_t fd_size = fd_num * sizeof(int);
60         char control[CMSG_SPACE(fd_size)];
61         struct cmsghdr *cmsg;
62
63         memset(&msgh, 0, sizeof(msgh));
64         memset(control, 0, sizeof(control));
65
66         iov.iov_base = (uint8_t *)msg;
67         iov.iov_len = VHOST_USER_HDR_SIZE + msg->size;
68
69         msgh.msg_iov = &iov;
70         msgh.msg_iovlen = 1;
71         msgh.msg_control = control;
72         msgh.msg_controllen = sizeof(control);
73
74         cmsg = CMSG_FIRSTHDR(&msgh);
75         cmsg->cmsg_len = CMSG_LEN(fd_size);
76         cmsg->cmsg_level = SOL_SOCKET;
77         cmsg->cmsg_type = SCM_RIGHTS;
78         memcpy(CMSG_DATA(cmsg), fds, fd_size);
79
80         do {
81                 r = sendmsg(fd, &msgh, 0);
82         } while (r < 0 && errno == EINTR);
83
84         if (r < 0)
85                 PMD_DRV_LOG(ERR, "Failed to send msg: %s", strerror(errno));
86
87         return r;
88 }
89
90 static int
91 vhost_user_read(int fd, struct vhost_user_msg *msg)
92 {
93         uint32_t valid_flags = VHOST_USER_REPLY_MASK | VHOST_USER_VERSION;
94         int ret, sz_hdr = VHOST_USER_HDR_SIZE, sz_payload;
95
96         ret = recv(fd, (void *)msg, sz_hdr, 0);
97         if (ret < sz_hdr) {
98                 PMD_DRV_LOG(ERR, "Failed to recv msg hdr: %d instead of %d.",
99                             ret, sz_hdr);
100                 goto fail;
101         }
102
103         /* validate msg flags */
104         if (msg->flags != (valid_flags)) {
105                 PMD_DRV_LOG(ERR, "Failed to recv msg: flags %x instead of %x.",
106                             msg->flags, valid_flags);
107                 goto fail;
108         }
109
110         sz_payload = msg->size;
111
112         if ((size_t)sz_payload > sizeof(msg->payload))
113                 goto fail;
114
115         if (sz_payload) {
116                 ret = recv(fd, (void *)((char *)msg + sz_hdr), sz_payload, 0);
117                 if (ret < sz_payload) {
118                         PMD_DRV_LOG(ERR,
119                                 "Failed to recv msg payload: %d instead of %d.",
120                                 ret, msg->size);
121                         goto fail;
122                 }
123         }
124
125         return 0;
126
127 fail:
128         return -1;
129 }
130
131 static int
132 vhost_user_check_reply_ack(struct virtio_user_dev *dev, struct vhost_user_msg *msg)
133 {
134         enum vhost_user_request req = msg->request;
135         int ret;
136
137         if (!(msg->flags & VHOST_USER_NEED_REPLY_MASK))
138                 return 0;
139
140         ret = vhost_user_read(dev->vhostfd, msg);
141         if (ret < 0) {
142                 PMD_DRV_LOG(ERR, "Failed to read reply-ack");
143                 return -1;
144         }
145
146         if (req != msg->request) {
147                 PMD_DRV_LOG(ERR, "Unexpected reply-ack request type (%d)", msg->request);
148                 return -1;
149         }
150
151         if (msg->size != sizeof(msg->payload.u64)) {
152                 PMD_DRV_LOG(ERR, "Unexpected reply-ack payload size (%u)", msg->size);
153                 return -1;
154         }
155
156         if (msg->payload.u64) {
157                 PMD_DRV_LOG(ERR, "Slave replied NACK to request type (%d)", msg->request);
158                 return -1;
159         }
160
161         return 0;
162 }
163
164 static int
165 vhost_user_set_owner(struct virtio_user_dev *dev)
166 {
167         int ret;
168         struct vhost_user_msg msg = {
169                 .request = VHOST_USER_SET_OWNER,
170                 .flags = VHOST_USER_VERSION,
171         };
172
173         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
174         if (ret < 0) {
175                 PMD_DRV_LOG(ERR, "Failed to set owner");
176                 return -1;
177         }
178
179         return 0;
180 }
181
182 static int
183 vhost_user_get_features(struct virtio_user_dev *dev, uint64_t *features)
184 {
185         int ret;
186         struct vhost_user_msg msg = {
187                 .request = VHOST_USER_GET_FEATURES,
188                 .flags = VHOST_USER_VERSION,
189         };
190
191         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
192         if (ret < 0)
193                 goto err;
194
195         ret = vhost_user_read(dev->vhostfd, &msg);
196         if (ret < 0)
197                 goto err;
198
199         if (msg.request != VHOST_USER_GET_FEATURES) {
200                 PMD_DRV_LOG(ERR, "Unexpected request type (%d)", msg.request);
201                 goto err;
202         }
203
204         if (msg.size != sizeof(*features)) {
205                 PMD_DRV_LOG(ERR, "Unexpected payload size (%u)", msg.size);
206                 goto err;
207         }
208
209         *features = msg.payload.u64;
210
211         return 0;
212 err:
213         PMD_DRV_LOG(ERR, "Failed to get backend features");
214
215         return -1;
216 }
217
218 static int
219 vhost_user_set_features(struct virtio_user_dev *dev, uint64_t features)
220 {
221         int ret;
222         struct vhost_user_msg msg = {
223                 .request = VHOST_USER_SET_FEATURES,
224                 .flags = VHOST_USER_VERSION,
225                 .size = sizeof(features),
226                 .payload.u64 = features,
227         };
228
229         msg.payload.u64 |= dev->device_features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES);
230
231         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
232         if (ret < 0) {
233                 PMD_DRV_LOG(ERR, "Failed to set features");
234                 return -1;
235         }
236
237         return 0;
238 }
239
240 static int
241 vhost_user_get_protocol_features(struct virtio_user_dev *dev, uint64_t *features)
242 {
243         int ret;
244         struct vhost_user_msg msg = {
245                 .request = VHOST_USER_GET_PROTOCOL_FEATURES,
246                 .flags = VHOST_USER_VERSION,
247         };
248
249         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
250         if (ret < 0)
251                 goto err;
252
253         ret = vhost_user_read(dev->vhostfd, &msg);
254         if (ret < 0)
255                 goto err;
256
257         if (msg.request != VHOST_USER_GET_PROTOCOL_FEATURES) {
258                 PMD_DRV_LOG(ERR, "Unexpected request type (%d)", msg.request);
259                 goto err;
260         }
261
262         if (msg.size != sizeof(*features)) {
263                 PMD_DRV_LOG(ERR, "Unexpected payload size (%u)", msg.size);
264                 goto err;
265         }
266
267         *features = msg.payload.u64;
268
269         return 0;
270 err:
271         PMD_DRV_LOG(ERR, "Failed to get backend protocol features");
272
273         return -1;
274 }
275
276 static int
277 vhost_user_set_protocol_features(struct virtio_user_dev *dev, uint64_t features)
278 {
279         int ret;
280         struct vhost_user_msg msg = {
281                 .request = VHOST_USER_SET_PROTOCOL_FEATURES,
282                 .flags = VHOST_USER_VERSION,
283                 .size = sizeof(features),
284                 .payload.u64 = features,
285         };
286
287         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
288         if (ret < 0) {
289                 PMD_DRV_LOG(ERR, "Failed to set protocol features");
290                 return -1;
291         }
292
293         return 0;
294 }
295
296 struct walk_arg {
297         struct vhost_memory *vm;
298         int *fds;
299         int region_nr;
300 };
301
302 static int
303 update_memory_region(const struct rte_memseg_list *msl __rte_unused,
304                 const struct rte_memseg *ms, void *arg)
305 {
306         struct walk_arg *wa = arg;
307         struct vhost_memory_region *mr;
308         uint64_t start_addr, end_addr;
309         size_t offset;
310         int i, fd;
311
312         fd = rte_memseg_get_fd_thread_unsafe(ms);
313         if (fd < 0) {
314                 PMD_DRV_LOG(ERR, "Failed to get fd, ms=%p rte_errno=%d",
315                         ms, rte_errno);
316                 return -1;
317         }
318
319         if (rte_memseg_get_fd_offset_thread_unsafe(ms, &offset) < 0) {
320                 PMD_DRV_LOG(ERR, "Failed to get offset, ms=%p rte_errno=%d",
321                         ms, rte_errno);
322                 return -1;
323         }
324
325         start_addr = (uint64_t)(uintptr_t)ms->addr;
326         end_addr = start_addr + ms->len;
327
328         for (i = 0; i < wa->region_nr; i++) {
329                 if (wa->fds[i] != fd)
330                         continue;
331
332                 mr = &wa->vm->regions[i];
333
334                 if (mr->userspace_addr + mr->memory_size < end_addr)
335                         mr->memory_size = end_addr - mr->userspace_addr;
336
337                 if (mr->userspace_addr > start_addr) {
338                         mr->userspace_addr = start_addr;
339                         mr->guest_phys_addr = start_addr;
340                 }
341
342                 if (mr->mmap_offset > offset)
343                         mr->mmap_offset = offset;
344
345                 PMD_DRV_LOG(DEBUG, "index=%d fd=%d offset=0x%" PRIx64
346                         " addr=0x%" PRIx64 " len=%" PRIu64, i, fd,
347                         mr->mmap_offset, mr->userspace_addr,
348                         mr->memory_size);
349
350                 return 0;
351         }
352
353         if (i >= VHOST_MEMORY_MAX_NREGIONS) {
354                 PMD_DRV_LOG(ERR, "Too many memory regions");
355                 return -1;
356         }
357
358         mr = &wa->vm->regions[i];
359         wa->fds[i] = fd;
360
361         mr->guest_phys_addr = start_addr;
362         mr->userspace_addr = start_addr;
363         mr->memory_size = ms->len;
364         mr->mmap_offset = offset;
365
366         PMD_DRV_LOG(DEBUG, "index=%d fd=%d offset=0x%" PRIx64
367                 " addr=0x%" PRIx64 " len=%" PRIu64, i, fd,
368                 mr->mmap_offset, mr->userspace_addr,
369                 mr->memory_size);
370
371         wa->region_nr++;
372
373         return 0;
374 }
375
376 static int
377 vhost_user_set_memory_table(struct virtio_user_dev *dev)
378 {
379         struct walk_arg wa;
380         int fds[VHOST_MEMORY_MAX_NREGIONS];
381         int ret, fd_num;
382         struct vhost_user_msg msg = {
383                 .request = VHOST_USER_SET_MEM_TABLE,
384                 .flags = VHOST_USER_VERSION,
385         };
386
387         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK))
388                 msg.flags |= VHOST_USER_NEED_REPLY_MASK;
389
390         wa.region_nr = 0;
391         wa.vm = &msg.payload.memory;
392         wa.fds = fds;
393
394         /*
395          * The memory lock has already been taken by memory subsystem
396          * or virtio_user_start_device().
397          */
398         ret = rte_memseg_walk_thread_unsafe(update_memory_region, &wa);
399         if (ret < 0)
400                 goto err;
401
402         fd_num = wa.region_nr;
403         msg.payload.memory.nregions = wa.region_nr;
404         msg.payload.memory.padding = 0;
405
406         msg.size = sizeof(msg.payload.memory.nregions);
407         msg.size += sizeof(msg.payload.memory.padding);
408         msg.size += fd_num * sizeof(struct vhost_memory_region);
409
410         ret = vhost_user_write(dev->vhostfd, &msg, fds, fd_num);
411         if (ret < 0)
412                 goto err;
413
414         return vhost_user_check_reply_ack(dev, &msg);
415 err:
416         PMD_DRV_LOG(ERR, "Failed to set memory table");
417         return -1;
418 }
419
420 static int
421 vhost_user_set_vring(struct virtio_user_dev *dev, enum vhost_user_request req,
422                 struct vhost_vring_state *state)
423 {
424         int ret;
425         struct vhost_user_msg msg = {
426                 .request = req,
427                 .flags = VHOST_USER_VERSION,
428                 .size = sizeof(*state),
429                 .payload.state = *state,
430         };
431
432         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
433         if (ret < 0) {
434                 PMD_DRV_LOG(ERR, "Failed to set vring state (request %d)", req);
435                 return -1;
436         }
437
438         return 0;
439 }
440
441 static int
442 vhost_user_set_vring_enable(struct virtio_user_dev *dev, struct vhost_vring_state *state)
443 {
444         return vhost_user_set_vring(dev, VHOST_USER_SET_VRING_ENABLE, state);
445 }
446
447 static int
448 vhost_user_set_vring_num(struct virtio_user_dev *dev, struct vhost_vring_state *state)
449 {
450         return vhost_user_set_vring(dev, VHOST_USER_SET_VRING_NUM, state);
451 }
452
453 static int
454 vhost_user_set_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
455 {
456         return vhost_user_set_vring(dev, VHOST_USER_SET_VRING_BASE, state);
457 }
458
459 static int
460 vhost_user_get_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
461 {
462         int ret;
463         struct vhost_user_msg msg;
464         unsigned int index = state->index;
465
466         ret = vhost_user_set_vring(dev, VHOST_USER_GET_VRING_BASE, state);
467         if (ret < 0) {
468                 PMD_DRV_LOG(ERR, "Failed to send request");
469                 goto err;
470         }
471
472         ret = vhost_user_read(dev->vhostfd, &msg);
473         if (ret < 0) {
474                 PMD_DRV_LOG(ERR, "Failed to read reply");
475                 goto err;
476         }
477
478         if (msg.request != VHOST_USER_GET_VRING_BASE) {
479                 PMD_DRV_LOG(ERR, "Unexpected request type (%d)", msg.request);
480                 goto err;
481         }
482
483         if (msg.size != sizeof(*state)) {
484                 PMD_DRV_LOG(ERR, "Unexpected payload size (%u)", msg.size);
485                 goto err;
486         }
487
488         if (msg.payload.state.index != index) {
489                 PMD_DRV_LOG(ERR, "Unexpected ring index (%u)", state->index);
490                 goto err;
491         }
492
493         *state = msg.payload.state;
494
495         return 0;
496 err:
497         PMD_DRV_LOG(ERR, "Failed to get vring base");
498         return -1;
499 }
500
501 static int
502 vhost_user_set_vring_file(struct virtio_user_dev *dev, enum vhost_user_request req,
503                 struct vhost_vring_file *file)
504 {
505         int ret;
506         int fd = file->fd;
507         int num_fd = 0;
508         struct vhost_user_msg msg = {
509                 .request = req,
510                 .flags = VHOST_USER_VERSION,
511                 .size = sizeof(msg.payload.u64),
512                 .payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK,
513         };
514
515         if (fd >= 0)
516                 num_fd++;
517         else
518                 msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
519
520         ret = vhost_user_write(dev->vhostfd, &msg, &fd, num_fd);
521         if (ret < 0) {
522                 PMD_DRV_LOG(ERR, "Failed to set vring file (request %d)", req);
523                 return -1;
524         }
525
526         return 0;
527 }
528
529 static int
530 vhost_user_set_vring_call(struct virtio_user_dev *dev, struct vhost_vring_file *file)
531 {
532         return vhost_user_set_vring_file(dev, VHOST_USER_SET_VRING_CALL, file);
533 }
534
535 static int
536 vhost_user_set_vring_kick(struct virtio_user_dev *dev, struct vhost_vring_file *file)
537 {
538         return vhost_user_set_vring_file(dev, VHOST_USER_SET_VRING_KICK, file);
539 }
540
541
542 static int
543 vhost_user_set_vring_addr(struct virtio_user_dev *dev, struct vhost_vring_addr *addr)
544 {
545         int ret;
546         struct vhost_user_msg msg = {
547                 .request = VHOST_USER_SET_VRING_ADDR,
548                 .flags = VHOST_USER_VERSION,
549                 .size = sizeof(*addr),
550                 .payload.addr = *addr,
551         };
552
553         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
554         if (ret < 0) {
555                 PMD_DRV_LOG(ERR, "Failed to send vring addresses");
556                 return -1;
557         }
558
559         return 0;
560 }
561
562 static int
563 vhost_user_get_status(struct virtio_user_dev *dev, uint8_t *status)
564 {
565         int ret;
566         struct vhost_user_msg msg = {
567                 .request = VHOST_USER_GET_STATUS,
568                 .flags = VHOST_USER_VERSION,
569         };
570
571         /*
572          * If features have not been negotiated, we don't know if the backend
573          * supports protocol features
574          */
575         if (!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK))
576                 return -ENOTSUP;
577
578         /* Status protocol feature requires protocol features support */
579         if (!(dev->device_features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
580                 return -ENOTSUP;
581
582         if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS)))
583                 return -ENOTSUP;
584
585         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
586         if (ret < 0) {
587                 PMD_DRV_LOG(ERR, "Failed to send request");
588                 goto err;
589         }
590
591         ret = vhost_user_read(dev->vhostfd, &msg);
592         if (ret < 0) {
593                 PMD_DRV_LOG(ERR, "Failed to recv request");
594                 goto err;
595         }
596
597         if (msg.request != VHOST_USER_GET_STATUS) {
598                 PMD_DRV_LOG(ERR, "Unexpected request type (%d)", msg.request);
599                 goto err;
600         }
601
602         if (msg.size != sizeof(msg.payload.u64)) {
603                 PMD_DRV_LOG(ERR, "Unexpected payload size (%u)", msg.size);
604                 goto err;
605         }
606
607         *status = (uint8_t)msg.payload.u64;
608
609         return 0;
610 err:
611         PMD_DRV_LOG(ERR, "Failed to get device status");
612         return -1;
613 }
614
615 static int
616 vhost_user_set_status(struct virtio_user_dev *dev, uint8_t status)
617 {
618         int ret;
619         struct vhost_user_msg msg = {
620                 .request = VHOST_USER_SET_STATUS,
621                 .flags = VHOST_USER_VERSION,
622                 .size = sizeof(msg.payload.u64),
623                 .payload.u64 = status,
624         };
625
626         /*
627          * If features have not been negotiated, we don't know if the backend
628          * supports protocol features
629          */
630         if (!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK))
631                 return -ENOTSUP;
632
633         /* Status protocol feature requires protocol features support */
634         if (!(dev->device_features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
635                 return -ENOTSUP;
636
637         if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS)))
638                 return -ENOTSUP;
639
640         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK))
641                 msg.flags |= VHOST_USER_NEED_REPLY_MASK;
642
643         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
644         if (ret < 0) {
645                 PMD_DRV_LOG(ERR, "Failed to send get status request");
646                 return -1;
647         }
648
649         return vhost_user_check_reply_ack(dev, &msg);
650 }
651
652 static struct vhost_user_msg m;
653
654 const char * const vhost_msg_strings[] = {
655         [VHOST_USER_RESET_OWNER] = "VHOST_RESET_OWNER",
656 };
657
658 static int
659 vhost_user_sock(struct virtio_user_dev *dev,
660                 enum vhost_user_request req,
661                 void *arg)
662 {
663         struct vhost_user_msg msg;
664         struct vhost_vring_file *file = 0;
665         int need_reply = 0;
666         int fds[VHOST_MEMORY_MAX_NREGIONS];
667         int fd_num = 0;
668         int vhostfd = dev->vhostfd;
669
670         RTE_SET_USED(m);
671
672         PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
673
674         if (dev->is_server && vhostfd < 0)
675                 return -1;
676
677         msg.request = req;
678         msg.flags = VHOST_USER_VERSION;
679         msg.size = 0;
680
681         switch (req) {
682         case VHOST_USER_SET_LOG_BASE:
683                 msg.payload.u64 = *((__u64 *)arg);
684                 msg.size = sizeof(m.payload.u64);
685                 break;
686
687         case VHOST_USER_SET_FEATURES:
688                 msg.payload.u64 = *((__u64 *)arg) | (dev->device_features &
689                         (1ULL << VHOST_USER_F_PROTOCOL_FEATURES));
690                 msg.size = sizeof(m.payload.u64);
691                 break;
692
693         case VHOST_USER_RESET_OWNER:
694                 break;
695
696         case VHOST_USER_SET_LOG_FD:
697                 fds[fd_num++] = *((int *)arg);
698                 break;
699
700         case VHOST_USER_SET_VRING_ERR:
701                 file = arg;
702                 msg.payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
703                 msg.size = sizeof(m.payload.u64);
704                 if (file->fd > 0)
705                         fds[fd_num++] = file->fd;
706                 else
707                         msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
708                 break;
709
710         default:
711                 PMD_DRV_LOG(ERR, "trying to send unhandled msg type");
712                 return -1;
713         }
714
715         if (vhost_user_write(vhostfd, &msg, fds, fd_num) < 0) {
716                 PMD_DRV_LOG(ERR, "%s failed: %s",
717                             vhost_msg_strings[req], strerror(errno));
718                 return -1;
719         }
720
721         if (need_reply || msg.flags & VHOST_USER_NEED_REPLY_MASK) {
722                 if (vhost_user_read(vhostfd, &msg) < 0) {
723                         PMD_DRV_LOG(ERR, "Received msg failed: %s",
724                                     strerror(errno));
725                         return -1;
726                 }
727
728                 if (req != msg.request) {
729                         PMD_DRV_LOG(ERR, "Received unexpected msg type");
730                         return -1;
731                 }
732
733                 switch (req) {
734                 default:
735                         /* Reply-ack handling */
736                         if (msg.size != sizeof(m.payload.u64)) {
737                                 PMD_DRV_LOG(ERR, "Received bad msg size");
738                                 return -1;
739                         }
740
741                         if (msg.payload.u64 != 0) {
742                                 PMD_DRV_LOG(ERR, "Slave replied NACK");
743                                 return -1;
744                         }
745
746                         break;
747                 }
748         }
749
750         return 0;
751 }
752
753 #define MAX_VIRTIO_USER_BACKLOG 1
754 static int
755 virtio_user_start_server(struct virtio_user_dev *dev, struct sockaddr_un *un)
756 {
757         int ret;
758         int flag;
759         int fd = dev->listenfd;
760
761         ret = bind(fd, (struct sockaddr *)un, sizeof(*un));
762         if (ret < 0) {
763                 PMD_DRV_LOG(ERR, "failed to bind to %s: %s; remove it and try again\n",
764                             dev->path, strerror(errno));
765                 return -1;
766         }
767         ret = listen(fd, MAX_VIRTIO_USER_BACKLOG);
768         if (ret < 0)
769                 return -1;
770
771         flag = fcntl(fd, F_GETFL);
772         if (fcntl(fd, F_SETFL, flag | O_NONBLOCK) < 0) {
773                 PMD_DRV_LOG(ERR, "fcntl failed, %s", strerror(errno));
774                 return -1;
775         }
776
777         return 0;
778 }
779
780 /**
781  * Set up environment to talk with a vhost user backend.
782  *
783  * @return
784  *   - (-1) if fail;
785  *   - (0) if succeed.
786  */
787 static int
788 vhost_user_setup(struct virtio_user_dev *dev)
789 {
790         int fd;
791         int flag;
792         struct sockaddr_un un;
793
794         fd = socket(AF_UNIX, SOCK_STREAM, 0);
795         if (fd < 0) {
796                 PMD_DRV_LOG(ERR, "socket() error, %s", strerror(errno));
797                 return -1;
798         }
799
800         flag = fcntl(fd, F_GETFD);
801         if (fcntl(fd, F_SETFD, flag | FD_CLOEXEC) < 0)
802                 PMD_DRV_LOG(WARNING, "fcntl failed, %s", strerror(errno));
803
804         memset(&un, 0, sizeof(un));
805         un.sun_family = AF_UNIX;
806         strlcpy(un.sun_path, dev->path, sizeof(un.sun_path));
807
808         if (dev->is_server) {
809                 dev->listenfd = fd;
810                 if (virtio_user_start_server(dev, &un) < 0) {
811                         PMD_DRV_LOG(ERR, "virtio-user startup fails in server mode");
812                         close(fd);
813                         return -1;
814                 }
815                 dev->vhostfd = -1;
816         } else {
817                 if (connect(fd, (struct sockaddr *)&un, sizeof(un)) < 0) {
818                         PMD_DRV_LOG(ERR, "connect error, %s", strerror(errno));
819                         close(fd);
820                         return -1;
821                 }
822                 dev->vhostfd = fd;
823         }
824
825         return 0;
826 }
827
828 static int
829 vhost_user_enable_queue_pair(struct virtio_user_dev *dev,
830                              uint16_t pair_idx,
831                              int enable)
832 {
833         int i;
834
835         if (dev->qp_enabled[pair_idx] == enable)
836                 return 0;
837
838         for (i = 0; i < 2; ++i) {
839                 struct vhost_vring_state state = {
840                         .index = pair_idx * 2 + i,
841                         .num = enable,
842                 };
843
844                 if (vhost_user_set_vring_enable(dev, &state))
845                         return -1;
846         }
847
848         dev->qp_enabled[pair_idx] = enable;
849         return 0;
850 }
851
852 struct virtio_user_backend_ops virtio_ops_user = {
853         .setup = vhost_user_setup,
854         .set_owner = vhost_user_set_owner,
855         .get_features = vhost_user_get_features,
856         .set_features = vhost_user_set_features,
857         .get_protocol_features = vhost_user_get_protocol_features,
858         .set_protocol_features = vhost_user_set_protocol_features,
859         .set_memory_table = vhost_user_set_memory_table,
860         .set_vring_num = vhost_user_set_vring_num,
861         .set_vring_base = vhost_user_set_vring_base,
862         .get_vring_base = vhost_user_get_vring_base,
863         .set_vring_call = vhost_user_set_vring_call,
864         .set_vring_kick = vhost_user_set_vring_kick,
865         .set_vring_addr = vhost_user_set_vring_addr,
866         .get_status = vhost_user_get_status,
867         .set_status = vhost_user_set_status,
868         .send_request = vhost_user_sock,
869         .enable_qp = vhost_user_enable_queue_pair
870 };